Latest / Oxide and Friends / Is NVIDIA like Sun from the Dot Com Bubble?

Transcript

Is NVIDIA like Sun from the Dot Com Bubble? · Oxide and Friends

• 0:01
  Bryan Cantrill: Hello?
• 0:02
  Adam Leventhal: Uh-oh. Hello, Bryan
• 0:04
  Bryan Cantrill: You can't hear me?
• 0:05
  Adam Leventhal: I can hear you great. I can hear you great. I was just I was cutting you off.
• 0:11
  Bryan Cantrill: Okay. There's no-- is there an, uh-oh?
• 0:12
  you know, I'm, I'm so audio sensitive.
• 0:15
  Adam Leventhal: I know. I know. I feel bad. No, You were saying hello, Adam. I cut you off.
• 0:19
  Okay. I didn't know how we were gonna get out of that situation. It just felt like we were gonna be stuck there for a while.
• 0:24
  Bryan Cantrill: Okay. Good. We're not stuck. We're not stuck. We're, we're back after a little, a little hiatus.
• 0:29
  We had a couple weeks off here. So Yeah. Little little bit of summer vacations, but, it's good to be back. I we're we're back and we're being trolled by the Internet. Instantly.
• 0:40
  Instantly trolled, Instantly trolled. And I feel like, you know, as we kinda think of, like, the buckets for Oxide and Friends episodes, we got, like, our hot takes. We got kind of our quarterly hot takes, maybe a little more frequently than that. They're not they're not as frequently as monthly, I don't think. I think we can spare people from that.
• 0:56
  But, we do have our occasional hot takes, and today is a hot take episode. But I pledge I I I have listened. Have you listened to any of our previous hot takes?
• 1:07
  Adam Leventhal: I know exactly what you're talking about. And I, the ones
• 1:09
  you call hot takes,
• 1:11
  I call Bryan gets trolled on Twitter, but you know, potato, potato.
• 1:16
  Bryan Cantrill: Fine. But, but we, we both know the fundamental problem with this. It's like, can you read the tweet, please, you moron? Can you please That's right. Can someone please read the tweet so we know what we're talking about?
• 1:28
  When we when folks are I mean, I I feel sorry for the the poor folks that are trying to consume this is a podcast. Like, I do not know what we're talking about. Can someone please read the entire tweet that we're angry about? So I I I want yanking part of it.
• 1:40
  Adam Leventhal: Like, that'd be a good start. I don't know.
• 1:41
  Bryan Cantrill: Just a part of it.
• 1:43
  Just a part of it. So I now, unfortunately, because, you know, Elon Musk has flown this thing into the into the sun, He now allows, like, blog entries as tweets. I got it. Yeah. There's, like, no upper bound on the end.
• 2:00
  We have here an essay, and I think I
• 2:03
  I do think we have
• 2:03
  to read it in its entirety. I think it'd be a mistake to not read this in its entirety. Now okay. But I've got some questions for you about, like, protocol before we get going.
• 2:11
  Adam Leventhal: Yeah. Yeah.
• 2:11
  Bryan Cantrill: Yeah. Do we there are there there there are many, many there's several mistakes in here, and then there are many things we've got commentary on. But we need to we we have to reserve that commentary. Obviously, I'm speaking to myself here. Obviously, not speaking.
• 2:29
  Adam Leventhal: You think you can read through all of these 7 or 8 paragraphs with no commentary?
• 2:36
  Bryan Cantrill: Okay. No, I don't think I can. What if I offer so but okay, here, I'm just concerned. This is how we end up in the situation. If I offer and already someone is listening to this on the podcast being, like, you can't
• 2:47
  Adam Leventhal: Read the tweet. Why are they not reading
• 2:49
  the tweet?
• 2:50
  Bryan Cantrill: Are they having a debate about how they're gonna read the tweet? And it's like, if you if you pre rage before you read the tweet, that's gonna be the whole podcast, you idiots. And you're never gonna read the tweet. So I and I also also, I would like to say just to also channel you in terms of you know, I like to channel your criticisms of me to, like, get part of the, but, you know, he banged them a little bit. And I also feel that the you know, I think that you would be right.
• 3:16
  You know, I I had I tweeted this out today despite my best attempts towards just talking about this. I I'm giving into the temptation of talking about it. I feel like you'd be right to be like, oh, oh, your your best attempts
• 3:26
  towards this.
• 3:27
  Adam Leventhal: I mean, I did. I I of course. I'm sure even as you wrote that, you could hear me saying really, like, that was
• 3:34
  Bryan Cantrill: Good. That's best effort
• 3:35
  Adam Leventhal: for you?
• 3:36
  Bryan Cantrill: Name 5 things you did. You know what? Actually actually name 3. Name 2, actually. Just name 2.
• 3:41
  Name 2 things that you did to resist talking about this. I don't think you'd are resisting talking about it.
• 3:47
  Adam Leventhal: So tell you what, why why don't you
• 3:49
  Bryan Cantrill: I think you're desperate to talk about it.
• 3:50
  Adam Leventhal: That's right. Why why don't you read as many paragraphs as you can? And then I think we could take a break. I think it'd be safe to take a break for some commentary as long as we promise
• 3:58
  Bryan Cantrill: to discuss it. We'll get back to it. Here's what I'm gonna do. No. Instead of that, I am gonna read through the whole thing.
• 4:02
  Adam Leventhal: Okay.
• 4:03
  Bryan Cantrill: I am going you would god, you would appreciate this. My 17 year old who is, definitely, I mean, god bless the kid, but I definitely got a little bit too much of my own genetic material through on that one. And his counterargument to any argument I might have about literally anything is to repeat exactly what I just said, but sound like a complete idiot. It's it's like,
• 4:32
  Bryan Cantrill as his son as him: oh, we should take the call today.
• 4:34
  Bryan Cantrill: It's like, no. No. That okay. That's not that's not my counterargument. That's just you, me, you're just now saying this.
• 4:38
  You're just this is just just mocking me. And he also loves to do an impersonation of me that you would find that just kills with the rest of the family. So the other thing he likes to do is repeat exactly the argument that I've made, but impersonating me in a way that just leaves everyone else howling. I'm like,
• 4:57
  Adam Leventhal: this is We've gotta have him in for the next Oxcon to just, like, deliver the keynote for the internal crowd, man. It was slight.
• 5:04
  Bryan Cantrill: I kid. So I the so I'm I'm going to try to read this earnestly. What I'm trying to is what I'm trying to say is I'm not gonna adopt his technique. I'm gonna try to read this earnestly.
• 5:14
  Adam Leventhal: You're not gonna read it in in the voice of him pretending to be you, comma, an idiot.
• 5:19
  Bryan Cantrill: I'd be right. That's I'm not going to do that. But in but in order to that the one thing I does the capitalization bother you at all? The so this person has spelled NVIDIA with a solitary capital v, lowercase n, capital v, lowercase the rest. And I think this is not correct.
• 5:38
  And I I acknowledge that NVIDIA is a strangely named company that seems to be inconsistent about their own capitalization. And, you know, the it it seems to, in many places, be, in all caps as kind of a brand, which is how I've always spelled it. And but you also see a capital n, lowercase video, but you'd never see it, like, lowercase n, capital v. Come on. That's not it.
• 6:01
  Does that
• 6:02
  Adam Leventhal: Well, it didn't because I you know, I thought that, like, way back in the day, that was their branding. Like, in the Okay. In, you know, buying graphics cards
• 6:09
  in them.
• 6:10
  Bryan Cantrill: So maybe there's gotta be some because I he is not the only person I've seen do this. So alright. I'm gonna read this thing.
• 6:16
  Adam Leventhal: Let's do it.
• 6:17
  Bryan Cantrill: Good good luck to us all. This is from Robert Graham, x, Arattle Rob, says NVIDIA is in the same position as Sun Microsystems was in the early days of the dotcom bubble. Sun had the leading edge web servers, the smartest engineers. Okay. We can agree on that point.
• 6:35
  No. No. It's alright. I'm doing it again. The smartest engineers, the most respect in the industry.
• 6:39
  If you were dotcom startup, you bought Sun Servers. Smart engineers wouldn't come work for your startup if you were dumb enough not to buy Sun Servers, and they charged a premium. Their profits went through the roof, except, well, there wasn't anything actually special about them. They weren't actually the best. Nothing really was.
• 7:02
  It's just that they were the least risky option. You knew they'd work, that newly hired employees would be familiar with them, that they were good enough. As a start up, you don't optimize for the efficiency of your systems, you optimize for building a business, like selling pet food, doing auctions, selling books online, and so on. You want growth, not profits. Once you've dominated the market and have steady revenue, you can afford to go back and fix the efficiency problems.
• 7:27
  It's funny because back in 1996, Windows NT 4 running on Pentium Pro was a vastly better web server than Sun. It's just alright. Getting through this. It's just that Silicon Valley startups couldn't find anybody who knew the system. Techies looked down on windows and considered it a quote toy operating system compared to the mighty Solaris, and Intel CPUs were CISC when everyone knew risk was better.
• 7:55
  Everybody was wrong, of course. NVIDIA is in the same position. Everybody wants NVIDIA chips for AI because they are known to work. The techies know how to program for them and so on. But Intel, AMD, and others make competitive chips for part or all of the AI stack that cost a lot a lots less.
• 8:14
  Indeed, Apple's own chips are quite good. Their private cloud could, in theory, be serviced by racks of Mac Ultra Servers, but they probably are buying NVIDIA too. When the dotcom bubble burst, Sun crashed and never recovered. Right now, VCs are throwing vast amounts of money at startups who are in turn sending it to NVIDIA. At some point, this will stop.
• 8:37
  Unsuccessful startups will go bankrupt and sell NVIDIA hardware and office chairs on eBay. Successful companies will now work to attain profitability by reducing costs.
• 8:48
  Adam Leventhal: Bravo. I did it. That that was really impressive. Really impressive.
• 8:52
  Bryan Cantrill: I think that is impressive. I only coughed once. I think it was really impressive. You know, and as with many things, I realized, of course, I didn't actually read every word of this thing, and there was definitely, there's there's even more to be said about this. So, one is, like, I definitely miss the character limit.
• 9:10
  I think the character limit is an important it's an it's an important part of the That's right.
• 9:15
  Adam Leventhal: Like saying things in as few words as possible is the soul of wit.
• 9:20
  Bryan Cantrill: It is. One might say. One might say. I I don't have another way of phrasing that, but if I did, we actually said that brevity is assault. What is that?
• 9:29
  That an EB white ism? I I feel like I don't know. I don't know. Okay. So you and I saw this because, you know, apparently, Adam, you and I are the keepers of the flame for for Sun Microsystems.
• 9:44
  That's right. Indeed. So I we could I I get just I just got tagged by a lot of folks about, like, okay. Anything sun, we got I'm like the the living fossil, but and they despite the fact that, like, I you know, I was I was at Sun for 14 years. You were at Sun for a decade.
• 9:58
  Right? Yep. And the, and, you know, we weren't we Sun was a very established company when I came there in 1996. I mean, I was, you know, my badge number is 30,992. Oh, yeah.
• 10:10
  But we were there for the back half. So, you know, it it definitely saw plenty, and and certainly happy to there's plenty in this that is not gonna resonate with our own experience. So that's the tweet. We I do wonder if people wasn't podcast like, you know what? On second thought, go back to not reading the tweet, actually.
• 10:33
  Yeah. I think I kind of preferred it when you guys weren't reading the tweet. You know, I think it was, there's a lot in here that is a very strange read of history. And I wanna kinda get to a bunch of different pieces of it. What's what kinda struck you here, rating this?
• 10:51
  Adam Leventhal: Man, so a couple pieces. I think, at the at the jump, the implication that the dotcombubble is like a, a reading of on the AI bubble, which I I think is is pretty far off. I think that that's maybe a reasonable place to start.
• 11:08
  Bryan Cantrill: Is the kind of this AI boom looks a lot like the dotcomboom. And I mean, there are obviously some commonalities, but it feels like there are a lot of differences too. It sounds like you've lost it. Oh, yeah.
• 11:20
  Adam Leventhal: Absolutely. I mean, I I just, I don't know. Even at the time, it felt like a bubble, like things didn't make sense on their face. And then just like the the outcomes of AI are are are very exciting and real, I think that even I think we're pretty I think we're both pretty excited about the prospects here, and they seem very tangible. But then when it comes to NVIDIA, it just doesn't seem like the same lessons could conceivably apply.
• 11:50
  Like, even if it turns out that there is this big, that there is a big con contraction associated with AI, that the money going into it, I I just it it feels like NVIDIA has, a lot of leads when it comes to the general properties of accelerated compute for sure for AI, but for other stuff as well.
• 12:13
  Bryan Cantrill: Yeah. I mean, I I just to put it out there, I do not think that Sun was never as dominant as NVIDIA's right now. I mean, I just don't think Yeah. I mean, just I mean,
• 12:23
  Adam Leventhal: that that's another great point in here, which was the implication that, that folks were buying Sun Systems and could only buy Sun Systems. But there was never this shortage, I don't think. There was never this thought of, like, you know, who is gonna get the next allocation of Sun Systems? I think Sun was making them pretty quickly, but it wasn't like, well, there's there's only enough to feed this this teeny fragment of the market, which is what it feels like within video right now.
• 12:51
  Bryan Cantrill: Yes. That's right. And I think that the it it so, like, the kind of the lead in on this is is a bit suspect just because Sun never felt as as dominant as NVIDIA is now. And, you know, Sun had kind of a leading edge web servers. It's like, well, it's a lot more than serving the web.
• 13:07
  I mean, that we had the I mean, I remember the would be like the 1999 all hands when, you know, Greg Papadopoulos got up in front of the company and said the the three most important business 3 most important applications for Son are databases, database, databases. So I I I actually think that, like, the the database boom was every bit as important as the web server boom, kind of the dotcom era. The so you like, that kind of is is kind of, rickety. I I I don't really buy that. And then you're, like, getting into the the the windows and t 4 running on a Pentium Pro is a vastly better web server than sending 1996.
• 13:42
  It's like, no. Sorry. No. The for a bunch of reasons. I mean, I think that, like, that that kind of ignores the fact that that x86 passed the risk manufacturers with respect to clock speed.
• 13:55
  It did not happen in 1996. It happened quite a bit later than that. And it was happening in as we got into the late nineties for sure. And then with with the the the the dotcom bust, then then, you know, clearly, x86 was was ahead of not just Spark, but more or less every other risk vendor, maybe with the with the possible exception of of power. But the the idea that, like, that it mean, in many ways, this that read on history is super, super strange because in 1996 in particular, and I remember that year vividly because it's the year that I went to go work for Sun.
• 14:34
  Really, that that is this is before that I mean, yes, the dotcomboomstartsinprobably1993.It starts growing in 1993. The net the the Netscape IPO kinda kicks things off. But it is not, like, full throttle in 1996. No. That that that would, that would come quite a few years later.
• 14:55
  That would come in like 97, 98, 99, 2000. And the in, it was still true in 1996, and I would say into 1997 that UNIX was viewed as antiquarian and that everybody felt that Windows NT was the future. So, I mean, this this kind of read on history is exactly wrong, in that it Windows was actually viewed as being completely dominant in the mid nineties. And, and I remember the the n t the anytime there was any kind of criticism of of Windows, it was always coming in n t 5. N t 5 was one of these great vaporware releases, and ever all the computer manufacturers had abandoned their unit experience.
• 15:38
  I mean, they had their unit experience, but this is kinda very, you know, very infamously when SGI, which had been one of the real UNIX innovators, they start moving to Windows in 96, 97.
• 15:49
  Adam Leventhal: And and it was right?
• 15:51
  Bryan Cantrill: Yeah. Absolutely. Yeah. Yeah. And so, you know, that was really not and then it was really only when the the the Internet really begins to take off, and it's because Windows is a crappy web server, and it's because Windows isn't actually good at these Internet workloads.
• 16:12
  It's not good at database workloads at that time. Not good very much, you know, still, having kind of residues of that personal computer heritage, and, it was because UNIX was the model that very much the model that grew up with the Internet. It was the model the Internet wanted, and the the Sun was UNIX Central. Sun had made the bet on UNIX and on SMP UNIX in particular at a time when many of the people have walked away from it. So the boom at at Sun is actually it's a it it it is exactly opposite of what's being claimed here, I think.
• 16:49
  Adam Leventhal: Yeah. Yeah. Well, we're certainly in in that mid nineties era, it did feel like Windows NT was it just, the hype of it had been, was like almost insurmountable. Like, it was the answer, regardless of the question. And I think then in maybe 97, it started to come down as you're saying, as people got more experience on it, more miles on those tires.
• 17:11
  Bryan Cantrill: Yeah. And I think begin to realize that, you know, I I've always viewed that one of those turning points was this thing called scalability day. And because what the thing that that had been, the the criticism of Microsoft had been its lack of scale, so they decided to, like, okay. We're gonna and actually, Jim Gray led this. Like, we're gonna have this great demonstration of the scalability of Windows, but they did it in a in a very rigged fashion.
• 17:37
  They basically it was a kind of an embarrassingly parallel workload that they had across many, many machines, and it wasn't actually that interesting. It was, you you know, it it was not and it it it kind of actually you know, it's one of the dangers of of overshooting. It put a spotlight on the fact that in the narrative, like, even in the Wall Street Journal I remember, because this is 1997, May of 1997, if memory serves. And even the Wall Street Journal was like, okay. This is actually kinda revealing that there's not a there there for Microsoft.
• 18:07
  And that to me, I was viewed as a bit of a turning point when begin people begin to realize, like, well, actually, maybe this maybe these these sun servers are maybe this this is this is what actually makes sense. And I even SGI was realizing that, like, okay. We actually, you know, we're we're trying to kinda have a a foot in both camps here, and and, that's where things really begin to to take off. Up until then, it was very, very suffocating. And, I mean, ironically, in from a period of, like, dominance to lost dominance, Microsoft felt absolutely dominant in that era, You know, leading up to say, like, 92 to 97, it felt like it that the future of everything was Microsoft and was very proprietary.
• 18:53
  And that actually does feel a bit like Nvidia, honestly. The I think that Nvidia is executing better than Microsoft was executing then. I think Nvidia is executing broadly, really well. But so it it mean, I I think that the kind of the that bit of the narrative is, is super weird. And I do, I mean, the replies of this tweet are like, that is a very strange read of history.
• 19:20
  Adam Leventhal: And it Did you see the one where you said, I mean, IO completion queues?
• 19:25
  Bryan Cantrill: I mean, that's
• 19:26
  Adam Leventhal: IO completion.
• 19:27
  Bryan Cantrill: Right. That was right.
• 19:29
  Adam Leventhal: So the what a what a strange way to try to win an argument. I don't know. It just felt like that's not the Trump card you've been holding on to, is it?
• 19:37
  Bryan Cantrill: That was a very bizarre and it's like okay. Look. Like, Windows NT had IO completion ports. Absolutely. And which is great.
• 19:46
  And this is, you know, all operating systems developed this kind of, of ability to have it's a support the the this kind of asynchrony. And, unfortunately, everyone did it slightly differently, and you had kinda k event from BSDs, and you had, we did event ports, in in Silveros, now Lumos, and and NT at IO completion ports, and then, of course, Linux did e poll. But the like, that is a that's a very inside baseball feature to cite. And so it would it really, it was it so that bit is like, that's just that's that okay. That's bizarre.
• 20:23
  And I think that the the other thing it was, like, bizarre is that that the the whole excess versus risk thing was, I mean, that
• 20:32
  Adam Leventhal: was want to ask about that because I feel like I am the last risk dead ender. And I'm sure that's not the case, but I just I think because I spent so much time with the Spark instruction set, like, decoding it in my brain, that I just, like, fell in love with risk.
• 20:48
  Bryan Cantrill: Yeah. I think well, you're a risk dead dead ender because you wanna go modify everyone's address space. You wanna go modify, like, program text. And like Exactly. Yeah.
• 20:57
  Of course, you're just talking your book with risk. That's right.
• 21:00
  Adam Leventhal: That's right. And yeah. But I'm I'm glad that that trolled you at least some. Maybe not as much or maybe this is just you're you're giving me a little present here because you know how much I I love risks secretly.
• 21:10
  Bryan Cantrill: Oh, I I we we all love risk. I mean, I think the thing that that is funny about risk is, I mean, I definitely love risk, and then every risk every risk construction site, which I guess is a redundancy, but the every specific ISA had its own little, like, very you know, for something that putatively was about the elegance of the thing, there'd be there were some seriously hairy warts on on basically every one of them. You know, every what every instructor said, I guess you're not gonna get I do love I think you and I both love RISC 5. Oh, yeah. Did you see
• 21:42
  Adam Leventhal: I mean, tying in with another episode. There's a RISC 5 motherboard for the Framework laptop. So
• 21:48
  Bryan Cantrill: Okay. Wow. This is, yeah. That's that does it run anything? What do you run on it?
• 21:56
  Adam Leventhal: Hey. Hey. Hey.
• 21:57
  Oh, you could say the same thing about my Tadpole laptop with its SPARC processor, I guess.
• 22:02
  Bryan Cantrill: Right. Like, look. Running things on it is not a constraint on the problem. So, yeah, that is kinda that's kinda wild. Yeah.
• 22:09
  I mean, I and I do love lots of elements of risk. I I think that they've done a very good job on the engineering of the instruction set of RISC 5.
• 22:19
  Adam Leventhal: Yeah. And because I
• 22:20
  Bryan Cantrill: think they they really tried, you know, I I and I think kind of the Cisco versus risk debates were I think that that is a bit of a false dichotomy. I mean, yes, that existed, but Intel had kind of figured out the instruction decode in the kind of mid nineties getting into the late nineties. The in order for x86 to surpass risk performance, they had to basically decode it internally into effectively risk instructions that they're executing. So Yeah. And then they've got big caches of decoded instructions.
• 22:51
  Because if it's a small, we'd always joke about, like, ASCII ad before multiplies, kinda my my my my poor whipping boy for the ridiculousness of Sisk. And you could just see in every passing generation that would take more cycles. And ASCII add for multiply is actually, you know what? You know what? It's an instruction I've never used.
• 23:10
  For all of my abuse of ASCII add for multiply, I almost wanna speak out on Asciiad for multiplies defense. Be like, hey, you know what? Why don't you why don't you execute this instruction before you knock it? You know? Yeah.
• 23:22
  You might like it.
• 23:22
  Adam Leventhal: You you should do it.
• 23:23
  Bryan Cantrill: I would throw up,
• 23:24
  Adam Leventhal: get that ASM macro and and give it a shot, see how it works.
• 23:27
  Bryan Cantrill: I might. I I you know, I just might. And this is for the binary encoder decimal support. It's a crazy legacy NX86. And, I mean, you know it still works, but at every generation, that would take more and more cycles because who cares about the performance of the stuff?
• 23:43
  In fact, you don't you you want to decode that into smaller micro ops that you can then optimize. So, I mean, like, risk 1 in that regard, in that that's the implementation of of Intel as well, and just or in the the implementation of x86 under the hood. But, yeah, I thought that was a little bit ridiculous. The and then, of course, like, everybody was wrong, of course. Like, everybody was wrong about what?
• 24:07
  Everybody Yeah. Like,
• 24:09
  Adam Leventhal: Just in general?
• 24:10
  Bryan Cantrill: It was right. Just in general? I think because I think on, like, on this particular point, like, actually, like, you're wrong on this. Yeah. Like, everybody else everyone else is kinda right.
• 24:19
  So and then the the other thing that which is super surprising, if you wanna talk about kind of the, you know, the the the end of that dominance of Sun such as it was, and again, nowhere near as dominant NVIDIA. I mean, it was it was x86 and the performance of x86, and then the availability of open source operating systems on x86, Linux specifically. I mean, that's what I mean, because people what people wanted was, like, fast Unix, and fast Unix was Solaris on Spark for a period of time, and then it was not Solaris on Spark. Right. And you and I were both in a bad relationship with Spark.
• 24:57
  I think is that is that fair to say? We were we needed to
• 25:01
  Adam Leventhal: Well, how dare you? Yeah. I mean, yes. I think that's right. Well, actually, you know, we we got out of that relationship, you know, with the Fishworks box.
• 25:08
  Bryan Cantrill: We did.
• 25:09
  Adam Leventhal: You know We did get
• 25:10
  Bryan Cantrill: out of that relationship. And you know what? The like and it it was better for Spark that we got out of that relationship. I was better for Spark that it, you know, it needed to go, it needed to figure some things out. Spark had some thoughts.
• 25:23
  Adam Leventhal: We were mad. We were just disappointed.
• 25:25
  Bryan Cantrill: We were disappointed. It was not good. The and, you know, Spark was there were other ideas that Spark was experimenting with that were very important. I do think that, like, it does kinda get into one of the challenges that Sun had. Because Sun was nowhere near as dominant as NVIDIA, but things were going well for Sun, for those, you know, what feels like a very brief period of time.
• 25:49
  And the I feel when things are going, one of the things that I learned when because, Adam, you interned in
• 25:58
  Adam Leventhal: 2,000. That's right.
• 25:59
  Bryan Cantrill: Right.
• 26:00
  Adam Leventhal: And then he turned I mean, arguably at I mean, it must have been totally at height. Like, I mean, just organic note. Me and my other intern, like, we were almost unhoused. Like, we could not find an apartment. Like, we were squatting in 1, buddies from college who had apartment.
• 26:18
  We had to have, like, my folks cosign and have some whole, like, power of attorney, like, craziness to just, like, we we were go we were, like, one day away from sleeping in our office.
• 26:29
  Bryan Cantrill: And, well, it just would, like, couch it in modern terms. Getting an apartment in Silicon Valley in 2000 was like getting a GPU today.
• 26:37
  Adam Leventhal: That's right.
• 26:38
  Bryan Cantrill: You're you're just like, it can't be had for love for money. Like, you can't even I mean, you just, and especially, you you you've got this kind of temporary housing. I guess, was Sun Sun didn't set you or Sun was trying
• 26:47
  to set
• 26:47
  you up, were they? No. They I don't I don't
• 26:49
  Adam Leventhal: it is vexing to me that that Sun didn't. Like, my my other buddies who are interns at the same time, but now Sun Sun couldn't figure out how where to put us other than, like, under our desks.
• 26:58
  Bryan Cantrill: Under your desks. So where did you end up? I guess I'm sorry. I don't think I concern for where you were spending every night. No.
• 27:09
  Adam Leventhal: We we were like, 1 buddy had, like, a month of overlap. So we had, like, mattresses and lamps that we had borrowed from work. And then we got some, like, corporate housing that, they were very reluctant to rent to us for just 2 months.
• 27:26
  Bryan Cantrill: So that is 2,000. That is 2,000. Yeah. It is go go times in Silicon Valley. And then you go back to school, and you graduate in 2001, and then you come out in 2001, and the world has changed.
• 27:41
  Adam Leventhal: Yeah. So, while I was a summer intern in 2,000, I I just wanted to know, like, one of, our shared professors from Brown was at a startup and tried to get me and the, Pete or the other intern to drop out of school to join his startup. That's that's how kind of bananas it was. I don't think I ever told you that.
• 28:01
  Bryan Cantrill: I you did right. I had you mentioned that, and I had forgotten. I think I returned I urge you to return to school. It's what I would like.
• 28:08
  Adam Leventhal: Yeah. Yeah. I think that would've I think I mentioned that to him again recently, and he that is like borderline child abuse.
• 28:18
  Bryan Cantrill: A borderline child abuse. I know it feels
• 28:19
  Adam Leventhal: like that. Oh, that feels like a Yeah. So, yeah, so there's return to sun. And I think I they gave me some stock options for my internship priced at the time of my internship. So at, like, 50.
• 28:33
  And the the recruiter did explain to me that those would actually probably be more valuable. Bear with him. Because usually, you could just depend on stock options doubling in value over 2 years. So if they started at 50 and they doubled in value and I really did that did not sound right to me. That felt I felt like there were some sort of fallacy in there.
• 28:53
  Bryan Cantrill: Correct. I can't put my finger on it, but I don't know. You seem to know what you're talking about. So doubling value.
• 28:58
  Adam Leventhal: He was a professional,
• 29:00
  Bryan Cantrill: so, but yeah. It did not double in value. No. No. To say It's it's okay.
• 29:09
  So this actually gets to, like, one immediate because this is actually one of the things I wanted to talk about In the there's a lot that's wrong with the street, but I also did wanna talk about, like, you know, booms and busts, dominance, the rise of dominance, and the fall of dominance. Because I think in tech, we do see this more. Like, there there you have this kind of rise to, like, absolute dominance, and then companies fade again. And and more so I think than in other domains. You know, I mean, not not unexclusively, but, like, you know, you think you, you know, you look at, like, you know, aircraft manufacturers are there have been start ups, but basically the same.
• 29:43
  Auto manufacturers, there have been start ups, and I guess Tesla would be a counter example of this, but, like, basically the same. And the in tech, like, we've companies that literally did not exist become absolutely dominant, and then cease to exist again in the span of, like, 30 40 years, which is a really compressed time frame.
• 29:59
  Adam Leventhal: Yeah.
• 30:00
  Bryan Cantrill: And is so there are there are also 2 different aspects of that, but you're hitting on one of them I want to hit on, which is for a public company, there is a valuation that that it has built into it future expectations. And the valuation is like can be and there were, like, the valuations in the dotcom boom were ludicrous. And there was absolutely no way, Even if things went well, there were no there was no and I think I do think that is a parallel with I mean, I think this is a real challenge that NVIDIA's got is the valuation is really, really steep. I mean, it's like how you surely NVIDIA interns are being told the same thing that, like, don't worry. The stock's gonna double.
• 30:45
  And it's like, is that right? Like, is the stock gonna double? Because, like, it's already I mean, that's wow. That would be and, you know, I mean, I think that the, the stock has been on a kind of on a bit of a slide because it's like the market cap right now is, like, close to $3,000,000,000,000. Is it gonna be a $6,000,000,000,000 company?
• 31:06
  Is it gonna be a $100,000,000,000,000 company? I mean, if you listen to the acquired episodes, yes. If you and I are fans of Acquired. I love Acquired.
• 31:13
  Adam Leventhal: Oh, yeah. Yeah. You listened to the interview with Jensen?
• 31:17
  Bryan Cantrill: I did. It's ex and it's like, they're great. And everybody I mean, in
• 31:22
  Adam Leventhal: a man after your own heart, he has over 40 direct reports.
• 31:26
  Bryan Cantrill: He is a man after my own heart.
• 31:27
  I
• 31:27
  do love the 40 direct reports. I actually I so, you know, I actually there's so much I actually do love about NVIDIA, and, I I I okay. Look. I do love to throw out 40 dark reports. I also you know, and this has been kinda quiet that there have been a couple of companies that have been very loud.
• 31:47
  Many companies have been very loud about the return to the office mandate. Less loud have been those companies that do not have those mandates, and those companies include Nvidia. And, you know, the whether it's the most viable company on earth right now, or it's definitely up there. But the fact that they, you so people are kinda looking around all these other companies being like, oh, everyone's doing a return to the ops mandate. It's like, not everybody actually.
• 32:09
  And I know, Adam, we believe very strongly the power of remote work. And it's kind of interesting. There are a lot of these beliefs that people look at in oxide as being totally iconoclastic. It's like, well, also, like, by the way, the most valuable company on earth also agrees with this. So it's like, it can't be that way.
• 32:23
  I mean, it's not that strange that 4100 reports. Please, let's not. Let's not, you know, I'm not a freak show over here. So, yes. It's, there are a lot of things that I really like about it, but it's very hard to imagine how that doubles in value.
• 32:38
  And you look at, and I think this is still true, that if you look at Cisco, which is a like, Cisco, unlike Sun, Cisco did did not perish. Right? And Cisco Cisco has not gotten close. I mean, what is right now, just looking at it, it is I mean, it it is closer than it's been in a long time, but it is not at its valuation from 2,000, from its its peak valuation in in like the that summer of 2000 when you could not get your apartment out of. It was at, like, that exact moment.
• 33:15
  You could've sold all of your Cisco shares.
• 33:20
  Adam Leventhal: Yeah. But
• 33:21
  Bryan Cantrill: it's kinda wild to think that, like, it is still you could be at Cisco for, you know, 25 years and a company that's, like, not a mess. Right? And is, like, not has done has done well commercially.
• 33:33
  Adam Leventhal: Yeah. And are like, arguably important even. Right? Like, both in terms of what they've done and the acquisitions they've made and and all that. And and to not get back to that threshold of 2,000 is remarkable.
• 33:44
  Bryan Cantrill: Isn't that crazy? And I which I think shows you that, like, you can have and you they I think they were blessed by kind of this this crash, which allows them I mean, in some ways, like, you don't what you don't want is like the kind of the slow drain or the slow bleed because whether we like it or not, the, the, the price of the equity does affect the way people think of the company. So,
• 34:12
  Adam Leventhal: while we're alluding to like some of my questionable financial decisions, I just wanna share, do you know who one of the early investors in Nvidia was?
• 34:20
  Bryan Cantrill: No, no. I mean
• 34:22
  Adam Leventhal: Yeah. So Sutter Hill, where I was an EIR. And there was some a lot of water cooler talk when I was there in 2016 about how, you know, they were about to kill it and so forth. I didn't I hadn't really paid too much attention to it. But I was getting, you know, some, like, I didn't commit insider trading.
• 34:39
  What I'm saying is there was a lot of encouragement to do things that might have been regarded as insider trading. And that would have been a good investment. It turns out To, to to buy Nvidia in 2016.
• 34:52
  Bryan Cantrill: To buy Nvidia in 2016. Yes. That would have been a good investment. Yeah. I mean, it's the the, so I I I can't tell if that's a confession or, like, an anti confession.
• 35:01
  What what what was that? Was that to say No.
• 35:02
  Adam Leventhal: Just it was just like everyone everyone else was talking about how they were gonna kill this thing, and I wasn't paying too much attention. And they thought, you know, graphics card company, I guess, or whatever. Not What
• 35:12
  Bryan Cantrill: do you mean to kill? I think I'm misunderstanding the verb
• 35:16
  Adam Leventhal: kill in the sentence.
• 35:17
  Bryan Cantrill: Oh, do well. Okay. Yes. That yes. That explains it.
• 35:21
  Yes. So they they everyone the this is NVIDIA is gonna kill it, and NVIDIA is going they're not gonna kill NVIDIA. Yeah. Right now, I would've Yeah. That would've been that that, you know, that would've been good.
• 35:32
  I I I'm sure there's been no bad tips that have happened in this
• 35:35
  Adam Leventhal: whole That's probably true. For the that's probably true.
• 35:37
  Bryan Cantrill: There are there are there are no zeros at Sutter Hill. There there are only winners. That's right. I'm sure they they don't bury their but, I mean, you look I mean, even I mean, and and, you know, Microsoft has had a a real resurgence obviously under Satya Nadell, but Microsoft also peaked in in 2000 and took into and was flat for years years years years years. And it's like that does when equity is a huge component of compensation, and compensation is why people are at a company, and I'm not accusing NVIDIA that.
• 36:10
  I don't necessarily think that's necessarily the case, but I think there there have been other companies where, you know, the comp is the expectation is that the equity is gonna increase in value because that's how you that's a significant portion of your compensation. When the stock is kind of a mess, it can it can be a downer for morale. It's a downer at some for morale. I mean, I'm I, you know, we did, I mean, we did work for a public company that lost 98% of its value. So, you know, that's
• 36:35
  a that's a thing.
• 36:36
  That's a downer.
• 36:37
  Adam Leventhal: It's a bit of a bummer. Yeah.
• 36:38
  It's a
• 36:38
  Bryan Cantrill: bit of a bummer. And I I kinda think do you also remember when when things actually do begin to unspool, you get the floors that people create for themselves. Like, oh, it can't go below 40. It's like, well, okay. Like, why not?
• 36:53
  Well, no. Because it's like, obviously, it can't go below 40. It's like, this is not this is not, you know, like, a a, this can definitely go below 40. Yeah. I mean, even when
• 37:05
  Adam Leventhal: I think people were saying, you know, can't go below the cash we have in the bank, but that turned out to not be a floor either.
• 37:12
  Bryan Cantrill: Yes. That was a, that one was a little bit, debilitating. I mean, in terms of like that was dispiriting, I would say, that, that we, when you, when you're being, when you're trading for less than your cash, I feel like, like you're treating us like we're, like a drug addict. I mean, we're being treated. I think we we are we're we're worth our cash.
• 37:34
  But so but that that can be a drain. And I think that this is a challenge when you got a really, really rich validation of that valuation. Like, that is a challenge. And I don't that is the part of Nvidia that I think is actually in many ways, like, the the the most fragile is actually not the company. I think it's the valuation.
• 37:55
  I don't know. I mean, this is obviously this is not stock market advice, but I guess it is. I don't know. I am not sure. I'm the and, normally, I go to my my my trusted financial analyst and adviser, Adam Levithal.
• 38:06
  Also, my my lawyer, hardware engineer,
• 38:10
  Adam Leventhal: with And postmortem consultant.
• 38:12
  Bryan Cantrill: And postmortem I fired you from the postmortem. The the the the postmortem consultant gig was super short. Post Adam Levinthal postmortem lawyer did not last very long. That turned
• 38:20
  Adam Leventhal: into
• 38:21
  Bryan Cantrill: you, that was a very short
• 38:23
  Adam Leventhal: I could be pretty quick.
• 38:24
  Bryan Cantrill: Yeah. Yeah. Well, pretty much turned into a finger pointing exercise where all the fingers pointed right at me. So, that was a reference to our data center outage episode a couple weeks ago where, so I like that, that aspect of that kind of equity aspect of the bust, I feel it was, I mean, it was very broad. I mean, it was so it kinda had happened to everybody.
• 38:50
  And I also felt like I mean, I don't know, Adam, how you felt. So you see you arrived back at Sun postapocalypse in 2001. Yeah. And, and how does it, I mean, what were some of the differences that you noticed? Other than it wasn't possible.
• 39:05
  Adam Leventhal: Oh, the post I mean, the traffic was easy. I got a department. No problem. So there were certainly some benefits of the bust. But you're right that the folks who were, like, explaining how things couldn't get lower than they were, it it it you know, as someone coming in, it did all seem kind of ridiculous.
• 39:25
  It's like, why did you think that? Like, why like, I I mean, I am just a child, so maybe you know better than I do, but, like, why why would that be the case? Why is there some fundamental law of gravity about the valuation here?
• 39:36
  Bryan Cantrill: Right. And there is no fundamental law of valuation, even though about the cash you have on hand, because so much of it is your future expectations. And the future expectations for NVIDIA are obviously really, really high right now in a way that it seems impossible for them to live up to the valuation, but, you know, who knows? I've been wrong so frequently with that company. I just, you know, who knows?
• 39:55
  Adam Leventhal: So so Todd joined us on stage and there was one part of of this in particular that I was hoping Todd would join us for, which was that the the tweet mentions how Intel and AMD have competitive chips that are basically just as good and cheaper and so forth. Is that true? I mean, that doesn't seem true. Todd, what's your experience there?
• 40:20
  Todd Gamblin: Oh, okay. I had a question. But yeah. Sure. I mean, is that true?
• 40:26
  If you look at have you looked at the top 500 recently? No. I mean, so Frontier's number 1. It's a big, MI 250x system. And it's I mean, they're getting performance out of it.
• 40:39
  My 250x is the software stack is not quite where it is with with NVIDIA, but you can get good FP 64 performance out of it. And, I mean, that's been our experience on the in my 300 a's as well. Nice. So so, yeah, I'd say I mean, m I three hundred is really the first, well, that's not true. M I two hundred was really their first HPC competitive.
• 41:02
  Like, we are serious about doing floating point chip. I mean, the the the capability was clearly, you know, competitive with NVIDIA's current chips. If you look at, like, the the flops they're offering. And and my 300 is is, I don't know, the one where they're, like, you know, we're we're actually here.
• 41:19
  Bryan Cantrill: Right.
• 41:22
  Todd Gamblin: So, yeah, I'd I'd say they're viable. AMD in particular.
• 41:24
  Bryan Cantrill: Nice.
• 41:26
  Todd Gamblin: Intel, they, it has been harder to get performance out of Ponte Vecchio on Aurora, which is currently number 2. They started out and it's not clear if this is because of the the GPUs or because of the network on the machine or what. But I mean, Aurora debuted on the top 500 with, like, half an exaflop of of performance when it was supposed to get 1. Now it's slightly over 1. Their percent peak is way lower, than what we're getting out of AMD chips right now.
• 41:56
  So I don't know. Intel has some catching up to do, and I don't know. We'll we'll see what they do. They're canceling or what? They're sunsetting Ponte Vecchio and moving on to Falcon Shores now.
• 42:07
  So, and you're supposed to use Gaudi in the meantime or something like that. So I I don't know. It it it will be interesting if they commit to
• 42:17
  Bryan Cantrill: a chip. Well, but I think it's suffice it to say you've got a lot of folks that are, and AMD, the probably the most viable among them, that are really that see the same opportunity that NVIDIA sees.
• 42:29
  Todd Gamblin: Well, Well, and they're making progress. Right? Like, I mean, they have made inroads into data center GPUs, and Microsoft's is deploying m I three hundreds in Azure.
• 42:38
  Bryan Cantrill: Right. And I I think that, like, the the part of what leads to the undoing of dominance is when companies believe that that dominance is a birthright or will last in perpetuity. And Adam, one of the things that I felt was that when Sun was dominant, everybody felt it was due to their piece of the puzzle.
• 43:00
  Adam Leventhal: It's all because of Java 3 d.
• 43:02
  Bryan Cantrill: A 100%. A 100%. I remember I no. I remember being in the so in 1997, I remember what Arthur Van Hoff telling me, every e ten k that Sun sells, they sell because of Java. And I'm like, are you fucking kidding me?
• 43:22
  I mean, no. I I I I strong disagree. He even left Sun at that point. I mean, he didn't even, and, you know, this is a younger man. I don't think I took that very well.
• 43:31
  I don't think I was invited back. I think we did not that was a social event that was not repeated. But the, I mean, the Java folks thought it was due to Java. The, the Spark folks thought it was due to Spark. The SMCC folks thought it was due to the machine.
• 43:45
  In the Solaris folks out of I mean, the operating folks thought it was I mean, everybody kind of feels like it is and I think this is one of the challenges of when a company is really successful, it is very, very easy to lose focus, especially at Sun. Sun lost focus when we were successful. And one of the things that I always occurred to me is like kind of interesting is that I always felt Adam, that we had, there, we were much more focused during the trough of the bust than we were in the boom. I mean, we developed DTrace in the bust, CFS in the bust. We the the the like I mean,
• 44:23
  Adam Leventhal: I was not there during the boom, but I would say that it it felt like we had a lot of opportunity to focus, shall we say, because, like, v VPs and product managers who might have wanted things kind of had wandered off into areas that they thought would be more productive.
• 44:38
  Bryan Cantrill: Yeah. And I remember we just had, like, so many dumb arguments, it felt like, during the boom, which ultimately and kind of hindsight boiled down to everyone believing that their piece was the piece that was the most responsible for any of Sun's success. And, you know, we were kind of like at some level all wrong and I guess all right in that it was this I mean, I always felt it was the system and and putting the the that that comprehensive system that we are putting together. But I think it's really hard to and I think one of the things that, I mean, NVIDIA, at least from the outside in has done well is retained focus during the and I think this is where, like, you really do have to give it I mean, the the story of Nvidia and those acquired episodes really go in the depth on it, but they had so many run ins with true death in a way that sun never did. You know, sun had had one real run-in with death in 1989, and I that that had cast a long shadow.
• 45:36
  And it was due to some operational issues. There were and there there were a bunch of, like, issues that contributed to that. But Sun basically is Tom, you know, when when Tom's been on here, it's kind of and I know Tom, you're out there in the audience, you can you can hop on stage, don't disagree. But, I mean, Tom the the Sun hit product market fit, like, straight out of the shoot, and it was just, I mean explosive. It was in the right place, right time, right team, right technology, right everything.
• 45:59
  The market was ready for it, just took off. And the and I mean, not not to downplay, I'm sure, what felt like the risk at that time, but Nvidia had real run ins with, like, like death knocking at the door.
• 46:14
  Adam Leventhal: Do you, do you remember the anecdote about the Rivo 128 from, the acquired podcast with Jensen?
• 46:19
  Bryan Cantrill: Yeah. You should retell that because that was the part
• 46:21
  Adam Leventhal: of it. I loved it because, so they were really, really, really low on cash, and were taping out this new chip and basically only had the cash for 1 iteration. And so spent a bunch of time and money, with emulation to get all the software right. And then were of the mindset, like, this chip is
• 46:42
  Bryan Cantrill: going to be perfect.
• 46:42
  Adam Leventhal: Why was the chip going to
• 46:42
  be perfect? Because it they not only did that, but double down saying, okay, because we're about to, or else they were dead. And then they not only did that, but doubled down saying, okay, because we're about to tape out a perfect chip, what are we gonna do next? It was like, well, we're gonna build it into a system and we're gonna do all the marketing around it. And all of this predicated on this just put all of the chips on on, like, one number and spin the spin the wheel and hope it lands.
• 47:12
  And in the in the Wikipedia article, it it talks about the Riva 128. It says the unofficial motto for the company today is still our company is 30 days from going out of business. So I don't know if it's true inside or not, but but certainly, that is, I mean, the opposite of complacency, certainly.
• 47:31
  Bryan Cantrill: The opposite of complacency. I I feel like I mean, kudos to Amazon. Same way. Right? Amazon, I think, has has also the retain I mean, any I think you can argue that that recently has maybe shipped at Amazon, and that was maybe more true a decade ago than it is now.
• 47:46
  But, you know, that stuff like keeps you that stuff casts a very long shadow. And I think it's part of the reason why NVIDIA hasn't really taken their eye off the ball from an execution is because they've had these run ins with that. And I mean, you know, Todd, you mentioned AMD. AMD also had run ins with that. I mean, AMD was, I mean, I guess you got run ins with death and then you have, like, actually was clinically dead, which I feel was more like a AMD was actually like dead on the operating table.
• 48:12
  Adam Leventhal: But when when do you feel like AMD was particularly dead?
• 48:16
  Bryan Cantrill: Oh, no. I did before Naples. AMD was dead. Bulldozer was I mean, it was dead dead.
• 48:21
  Adam Leventhal: So this was like this was you know, they they were first out with the 64 bit, you know, x86. Was that a hammer?
• 48:29
  Bryan Cantrill: Yes. That would you're right. That's right.
• 48:31
  Adam Leventhal: And that was the real that was a huge resurgence for them in, like, what it was, like, 2,004, that kind of neighborhood?
• 48:37
  Bryan Cantrill: The oh, yeah. And because Intel had misexuted. Intel the Intel had, and, you know, suffering from its own problems about what an Intel has. This particular variant of misexecution is like a specialty of the house at Intel, where they're like, no. No.
• 48:56
  X86 is not what we we are we are actually gonna go do the grand new architecture that all software will be rewritten in. You're like, didn't you just do this with APX 432? But they did this is IE 64, and and what was Merced, and was just an absolute disaster. And a very the wait. How do you feel about VVW?
• 49:18
  Does VVW leave you with complicated feelings? You never had to work on a VIOW machine?
• 49:24
  Adam Leventhal: No. No. I haven't.
• 49:26
  Bryan Cantrill: Except VIOW is kind of like you can you can view it as, like, the best and worst of both worlds. I'm not sure. But VIW is basically an idea that did it was kind of the wrong idea
• 49:34
  Adam Leventhal: at the wrong time.
• 49:35
  Bryan Cantrill: Yeah. And, and then Merced was a disaster, and McKinley, everything was gonna be a McKinley. This is when I wanted to name I wanted Sun to name its chip that was gonna be concurrent with McKinley after McKinley's assassin? Leon Zolgaz? Did you believe I
• 49:56
  Adam Leventhal: don't know. No. I wasn't. I wasn't part of that Zolgaz.
• 50:00
  Bryan Cantrill: The Zolgaz? I'm sorry.
• 50:02
  Adam Leventhal: Marketing campaign that you were on.
• 50:04
  Bryan Cantrill: Yeah. As it turns out, like, Leon Zolgaz I mean, there's a reason that history has kinda forgot this guy. But it'd be it's Zolgaz has got I mean, it's it's a lot of lot of consonants in that name. It's a lot a lot of season z. It's a
• 50:15
  Adam Leventhal: good scrabble word. Agreed.
• 50:17
  Bryan Cantrill: But yeah. Yeah. Good scrabble word. Not not such a great and then arguably, you know, a bit a bit off color. But I've come, like, come on.
• 50:24
  I mean, McKinley's assassin. It's all it's it's it's it's straight there. Just an, it's in the distant past now. But the, so that the they took their eye off the ball. AMD came in and did whatever we wanted wanted them to do.
• 50:36
  We still called AMD 64, and actually extend the x 86 architecture to b 64 bit, and and take that take away some of the register pressure, which is really acute. Get some get some actual registers as opposed opposed to just, you know, having the 4 registers, whatever I get some 6. And the, and then Intel had, and, you know, kind of had heard that rumor that Intel had this project Yam Hill internally where they were. And so it, AMD had a kind of this very brief window and then AMD misexecuted. Right?
• 51:08
  And AMD had this very brief period of kind of dominance. It, again, nowhere near what NVIDIA has today, but they, they took their eye off the ball. And this is in the you know, so, Adam, you recall that we were on this is on, Shanghai and Barcelona, and, they were they did not hyper transport did not keep up. And, Barcelona, I believe it was Barcelona, was one of that that was really late. And Intel then had, at this point, put had had kinda, killed, i64, and, the and proceeded to execute very well, with Nihaylo.
• 51:56
  Right. And so the but and then AMD is dead dead, by the time you fast forward 10 years. And AMD is so dead that Intel is I remember Intel presenting us a road map in like 2012. That was a very aggressive road map. And I'm like, I remember telling Robert, I'm like, do they not know that AMD is dead?
• 52:19
  They are acting don't tell them that AMD is dead. Like, they are acting like AMD is still alive, which is kind of amazing. You think like that That's only that's not that long ago. Like, that's only 12 years ago when Intel Especially in
• 52:32
  Adam Leventhal: the in the amount of time it takes to make a chip. That's not that long ago.
• 52:35
  Bryan Cantrill: Time may makes make a chip. Exactly. You think like, wow, that wasn't that long ago. Intel was so dominant that I I, like, didn't wanna tell them. That I I I didn't want them to know how dominant they were for fear that they would stop with this very aggressive roadmap.
• 52:52
  But then Intel and just shows how these things can change. Intel made misexecuted, and they and you can misexecute, like, once or twice, but it's when you cut start linking misexecutions together, and they misegeted at 10 nanometer was an absolute disaster for them, and then they misegeted, and they just totally missed chiplets. And by the time we're starting a company in 2019, for us in 2019, it you know, we're doing the analysis, and it's very clear that we're gonna go all in on AMD versus Intel 4x86. And at that at that time, I felt like it was I don't know. Because you've been kinda outside the CPU space a little bit for a couple years.
• 53:36
  Coming back in at that time, how, risky did that feel?
• 53:42
  Adam Leventhal: You know, it's so funny because, like, I was, I felt like I had gone I had totally missed the trough of AMD. So the fact that we were just putting those up next to each other, you know, felt like all normal. And the fact that AMD was kinda shaking out on top also felt normal. And the fact that everyone was squinting at the spreadsheet because there was a time when it was really clear that Intel was gonna pour a ton of help into it, that, you know, that we were gonna get, like, engineering help. That there there was lots of of support that Intel was promising that I don't think we felt like we were gonna get the same kind of support from AMD, but that everyone was squinting at it and trying to see in the spreadsheet some opportunity, to work with Intel, you know, despite the just overwhelming, you know, numerical evidence of the the strength of the AMD parts.
• 54:33
  That it was just sort of surprising, but it makes sense based on folks' history.
• 54:38
  Bryan Cantrill: Well, it just can show you that, like, in a very small number of generations, things can really, really flip. And we had some advisors of the company in 2019 who were like, you are making a mistake by betting on AMD. And, like, the like, the world like, of course, AMD is not going to out execute Intel, but that's exactly what's happened in the last 5 years. Right? And Yeah.
• 54:57
  They and, you know, you'd be very hard pressed to to argue that x86 that Intel x86 is dominant right now, and really AMD x86 is really, much more dominant, and there's been, you know, the there, so I mean, which kind of amazing how quickly you can be, like, rise and fall and rise and fall and rise and fall again. In fact, you actually missed I you you missed AMD being dead on the operating table. Exactly. You know, like, right, everyone's way repeat
• 55:25
  Adam Leventhal: Right.
• 55:25
  Bryan Cantrill: Person's alive. I'm not sure why.
• 55:26
  Adam Leventhal: Just look away for a decade and all of a sudden. Right.
• 55:29
  Bryan Cantrill: Right. Exactly. And which which I think does show you that, like, I and, you know, we we talk with you. I think you kinda getting back to NVIDIA's dominance. It does feel very dumb, but just to Todd's point about, you know, AMD really chipping away there.
• 55:43
  And it it it require if NVIDIA as NVIDIA NVIDIA will be deprived of margin to misexecute is kind of the way I would phrase it. Yeah. That the and if you do misexecute, you've got a bunch of folks that are gonna pounce. And certainly, that's felt what it felt like happened to Microsoft, which had been so dominant, and then did misexecute. They they they they pretty grievously misexecuted on I mean, they misexecuted on a bunch of things.
• 56:12
  They they misexecuted for sure on Unix and Linux and open source. And, I mean, even as late as 2012, they the fact that you could run Linux guests in Azure was something that was not made available to the public. It's like a it was like a secret inside of Microsoft. And, the and it was really with the departure of Ballmer that they could kinda enter a new era, which has been very important for them. And it just showed you, like, you can't count these companies out either.
• 56:43
  So that I you you know, we've got, in terms of, like, what what the future may hold for NVIDIA, it's gonna be, I, I, it's just, it's hard to imagine that Nvidia stays this dominant in perpetuity. So I think So
• 56:59
  Adam Leventhal: I'm with you. I mean, I think that seems right. I mean, think that that is sort of a safe bet given that they're, like, at what what seems like a real peak. And yet, CUDA seems like such a a huge moat. In a moat that, like, for I mean, we talked about Sun was never as dominant.
• 57:15
  Part of the reason why it didn't feel as dominant is I don't think that the switching costs ever felt as high.
• 57:21
  Bryan Cantrill: Whereas what we what we've
• 57:23
  Adam Leventhal: heard in the market over and over again is how pervasive and useful and good CUDA is. And, you know, how people wish Rock'em were better.
• 57:33
  Bryan Cantrill: Were better. But to Todd's point, you know, this thing being used, heavily at Microsoft, heavily at, you know, elsewhere, there's a lot there are a lot of reasons for the software to improve. But, yeah, I mean, it's like, is the software gonna get there? I do kinda wonder if, you know, the the AMD's approach, I I and I think this is I I also think is is a reason why NVIDIA may become a misexecution for NVIDIA. They they are very, very proprietary.
• 57:59
  And
• 58:00
  Adam Leventhal: Mhmm.
• 58:00
  Bryan Cantrill: It does kind of like it it's almost as if x86 only supported Windows. You know? Like if you can imagine that world for a moment where, you know, a world that didn't exist, but if you can imagine where in order to run an X86 chip, you had to run windows. And it would've been a different world, you know, and it would've, I think it would've extended the dominance at some level of Microsoft, but it also would have when it collapsed, it would've kind of it would've collapsed more completely because ultimately, it was the fact that x86 was a an open architecture. I mean, open, I guess, in air quotes.
• 58:40
  But the fact that you could that you could put other software stacks on top of x 86, that you are not well that that Intel was very deliberately not forcing you into software choices. Even necessarily even like, you know, CUDA, you can argue is a good one, but the the the fact that you had that flexibility. So I I I think that I think the fact that the the the the proprietariness of NVIDIA, I think is a challenge for them, and and is a is a potential area of misexecution.
• 59:09
  Adam Leventhal: Yeah. It's interesting because it does we have seen, you know, at least in my career, this, this fight against proprietary components and how it may take a while, but the market finds a way to route around some of these things. And you're right that it does potentially set up maybe even like a Sun like hurdle. I mean, we you know, keep in mind, we had a proprietary CPU, proprietary operating system. We were, you know, always I mean, we were more open than Nvidia, but there was a lot proprietary about us even until, you know, open Solaris in 2003 or whenever it was, 2004.
• 59:47
  Yes.
• 59:47
  Bryan Cantrill: 2005.
• 59:49
  Adam Leventhal: There we go.
• 59:50
  Bryan Cantrill: Too late.
• 59:51
  Adam Leventhal: Yeah. Much too late. Much too late.
• 59:53
  Bryan Cantrill: I mean, because I think this you kinda get into the power game of, like, I think that Sun's proprietariness, despite the fact that being it was born as the open systems company, I think it was proprietary far too long. And I think that we I mean, if we'd open source the operating system in, like, 98, I think the you know? I don't know. The way the the different things happen. Yeah.
• 1:00:12
  For sure. Yeah. Would be would have been different. So I I that's something I definitely wonder about. You know, one of the one of the questions was, what what is the signal that we saw in oxide that about AMD to make the bet?
• 1:00:25
  And this is we we were not splitting the atom here. This is like this is really clear.
• 1:00:31
  Adam Leventhal: If you just Power in, instructions out. It was it was it was very, very clear.
• 1:00:37
  Bryan Cantrill: Yeah. If we looked at like, the things we cared about were things like power, performance, thread count, clock speed, you know, and on more or less every sing and then cost, and on every one of those axes. There was just no comparison. We should actually I wonder if we should should we open our f d 12? I need to go read our f d 12.
• 1:01:00
  Adam Leventhal: I'm sure it's filled with proprietary information.
• 1:01:03
  Bryan Cantrill: I know that it is the proprietary information. The r t 12 was the r t that we had that went into this analysis. I'm more concerned that it's filled with infective. Actually I guess, I guess, someone who may have contributed some some punchy lines to our
• 1:01:16
  Adam Leventhal: It's probably a healthy blend of both proprietary information and infective.
• 1:01:19
  Bryan Cantrill: That that's right. Yeah. Right. It is. The I've been I come for the proprietary information.
• 1:01:25
  Stay for the Invective. But that's I mean, just a low number denotes how I mean, this is an early decision for us. We wanna make sure we were right, and it was just very clear. And, oh, the thing that I didn't realize until we were talking to Intel was that Intel had been anti chiplet. That was a huge surprise, because I'd like, okay.
• 1:01:48
  They're anti chiplet, and they had this kind of this tiled architecture, and it was just very clear that the the chiplet architecture was the right architecture. I don't see and, you know, the obviously, the GPU is not the domain of expertise. I but I don't see NVIDIA making mistakes like that. I do wonder so, Adam, another thing to put in front of you, you know, and, you know, you and I talk I I know we've talked in the past about because the moment was so a kinda catalytic for both of us when you, I think, is an intern. Right?
• 1:02:19
  When you would we had just seen the price tag
• 1:02:22
  Adam Leventhal: on Yeah.
• 1:02:22
  Bryan Cantrill: On a Serengeti system. You're like, wow, $800,000 for 8 CPU machine. Like, why would I not buy, like, 8,000 to 1 CPU machines? I'm like, I don't know. That's a good question.
• 1:02:33
  Adam Leventhal: I don't know. I'm sure someone's thought of it. Right? Whatever.
• 1:02:36
  Bryan Cantrill: Yeah. I
• 1:02:36
  don't know. Someone else picture. I I I don't know. I think this they tell me the stock is gonna double in value, so I'm pretty sure that's what it is. But anyway,
• 1:02:41
  let's go on.
• 1:02:42
  Adam Leventhal: Regardless of what the value is and the and the initial option, doesn't make sense to me. But
• 1:02:46
  Bryan Cantrill: Yeah. It makes sense to that one. And of course, like, as it turns out through the, you know, out of the mouth of babes, like, oh my God, we were right. I do really wonder about and, you know, obviously, we had that terrific episode and conversation with Simon Wilson about, you know, the the acquisition of Mellanox by NVIDIA is so, strategic with respect to understanding the importance of very low latency networking to do this kind of massive training because these models are so large. And it it feels like there's gonna be a lot of economic pressure on finding it's certainly on the inference side.
• 1:03:23
  It's like, can we can we just get these things can we get the high speed ultra high speed networking out of the picture? Is there a way to go, like, make these models small enough to live in either HBM or or DRAM? And you wonder if, like, the economics because I do feel like the economics of Sun's servers, coupled with the bust, forced companies like Google to do something else. Google could not have been built on the Google had to be built on something cheaper, and Google could not have been built on Windows. Google had to have been built on open source software because they just didn't have, you know, the the resources in early the early 2000s.
• 1:03:59
  And you you do wonder if a you know, in in this regard, like, an AI bus could be super interesting because you could have not because it would wipe out companies, but it would just make companies much more focused on, like, I gotta run on whatever I can get. And if I, and I'm actually way, I care a lot more about price performance than I do maybe today. And I think that's another potential point of vulnerability.
• 1:04:28
  Adam Leventhal: Interesting. I mean, even the scarcity may be a point of vulnerability because it Right. It is forcing folks to to invest more, both in alternatives and in these kinds of interposition layers where you're not committed to one architecture or another, just by the necessity of of the NVIDIA chips being complete unobtainium.
• 1:04:48
  Bryan Cantrill: Okay. So to make a crazy analogy, I think that the there was a an era of absolute dominance of AWS. And I okay. I say AWS less dominant than they were, say, in 2014.
• 1:05:04
  Adam Leventhal: Okay.
• 1:05:04
  Bryan Cantrill: That's hot take number 1. I'm not sure if that's if the
• 1:05:08
  Adam Leventhal: Go on.
• 1:05:09
  Bryan Cantrill: Hot take number 2.
• 1:05:10
  Adam Leventhal: There are other alternatives, but it it does feel like they're they are still almost synonymous with public cloud. But,
• 1:05:16
  Bryan Cantrill: Yes. They are. But, like, we we we have run into plenty of GCP and Azure customers.
• 1:05:22
  Adam Leventhal: Yeah. Even Oracle Cloud customers.
• 1:05:25
  Bryan Cantrill: Come on.
• 1:05:25
  Adam Leventhal: No. Really?
• 1:05:26
  Bryan Cantrill: This is not a comedy hour. Who's who's an Oracle Cloud customer? Name 10.
• 1:05:33
  Adam Leventhal: I can't name 10 publicly.
• 1:05:35
  Bryan Cantrill: Okay. Okay then. Okay. The the we got Oracle. You're just trying to, like, we'd a list doesn't need to have 4 things on it.
• 1:05:41
  You can just have 3 things on it. Okay. You don't need the, like, flesh UCS. Oracle, Liven. Right.
• 1:05:47
  Yeah. This is like, I've tried really hard to resist being trolled by this tweet. It's like, you know, I've tried very hard. Come on. But the I mean, we see a lot there's just and I and I think so hot take number 2 is that the rise of Kubernetes tracks that to a degree.
• 1:06:04
  Adam Leventhal: Interesting. Because it makes it less the the specific choice becomes less relevant. You're getting more abstraction between you and the infrastructure provider.
• 1:06:11
  Bryan Cantrill: Well and I'm not sure this is correlation or causation. I think this may just be correlation. I think a lot of the energy behind Kubernetes was people having energy for not being so beholden to AWS. And at the it it the I mean, I I would I I would love to get, the others I others takes on my hot take, but that that's the, I I think that AWS became margin I mean, it's obviously still, like, wildly successful. It's still dominant, but it it it, like, it did not there was an era when where it had all the oxygen in the room.
• 1:06:48
  And when Adrian Cockcroft is talking about how you gotta build on, you know, everything needs to be built on every knee like, they would announce a new service, and everyone would say you have to build on this new serve. Everything has to be built on this new service. It feels like that's not the era that we're
• 1:07:01
  Adam Leventhal: we're you know? Yeah. I think
• 1:07:02
  Bryan Cantrill: that's right.
• 1:07:02
  Adam Leventhal: I think you're you're right. It's like there's a reinvent where there's, like, 18,000 new services.
• 1:07:07
  Bryan Cantrill: Right.
• 1:07:08
  Adam Leventhal: And it is not like, if you don't do this, you'll be behind. You're right. I I I agree.
• 1:07:12
  Bryan Cantrill: That's right. And it felt like that was true for, like, a like, a scratch of, like, 4 or 5 years.
• 1:07:17
  Adam Leventhal: Yeah. I think that's right. I I I agree with that.
• 1:07:20
  Bryan Cantrill: Where it's like, oh god, like, we had to, like and, you know, okay. Now I gotta use Kinesis or now I've gotta use, you know, Greengrass or, you know, Fargate or, you know, I think it's there was a I've gotta get Corey on here. We got Corey could be like, come on. What's the guy we could do with the the he could best us both on
• 1:07:40
  Adam Leventhal: on Rattling off the obscure AWS services?
• 1:07:43
  Bryan Cantrill: The obscure AWS services, that's to their credit. They never seem to turn off. I really admire that about AWS. They support support everything in Praktuity.
• 1:07:50
  Adam Leventhal: Yeah.
• 1:07:50
  Bryan Cantrill: But I I so I you you wonder if, like and I think that that is kind you know, ultimately, the the companies lose their dominance because they become arrogant at some level, and because they view it as their birthright. And I think each son don't know if it was quite as, you know, I think it was arrogance, for sure. And then I think it was also just like losing track of, no longer losing some of that hunger, losing some of that that kinda connection with why someone would buy the product or why they wouldn't buy the product, losing some of that connection with the customer, not seeing some of these other economic pressures. I think that's what leads to the demise of these things. And then Yeah.
• 1:08:30
  And then the demise seems, I mean, as expectations, high expectations on the upside will make up for tons of of defects. Right? You can paper over a lot of things. People just kind of assume that you're going to, you know, on everything, like you're gonna get there if you're not there yet. When, man, when it, when the whole narrative flips, it doesn't seem to matter like how much great stuff you're doing because like you're actually the narrative is flipped and Yeah.
• 1:08:59
  And it's, it can be really, really hard, I think. I mean, obviously, NVIDIA's a long way from that because the the, I again, they continue to execute very, very well. So I think that, like, that that the the that's the bit that and then I will will say, like, the final bit of history that is the the misread that is not correct is Sun's crash after the dotcom bubble bath after the bust. That is not what led to the demise of Sun. I don't think.
• 1:09:31
  Adam Leventhal: I mean, it didn't help.
• 1:09:33
  Bryan Cantrill: It didn't help.
• 1:09:34
  Adam Leventhal: So so What do you
• 1:09:37
  Bryan Cantrill: I'm Levitt Hall's CFO.
• 1:09:40
  Adam Leventhal: I'm just saying, what do you think, was it was more on the scene there?
• 1:09:46
  Bryan Cantrill: Well, I I don't I don't think it helped. Obviously, it didn't help, but I think that like Sun could have etched a path that was Cisco, for example. Sun did not need to die with the .com bust. We were not a Yeah. We were not web fan.
• 1:10:02
  It's not like we evaporated overnight. I mean, Sun was acquired by Oracle in in 2 in 2010, long after the dotcom bust. Yeah. Yeah.
• 1:10:12
  Adam Leventhal: So You know, I I I mean, I I don't wanna put it all on on Jonathan Schwartz's shoulders, but it did seem like there was a a continuity problem where, you know, I'm not sure that Scott had done the Scott McNealy founding or CEO of Sono. I'm not sure he had done the things that needed to be done to change the culture, to adapt post the dotcom boom, the bust rather. But but when you talk about, regarding successes of birthright, that felt more like the Jonathan era without ever figuring out what what we needed to be different.
• 1:10:47
  Bryan Cantrill: Well, and I think that this is always a challenge for when a company is no longer founder led. It's very easy to lose the way. Yeah.
• 1:10:56
  Adam Leventhal: And I That's exactly what I
• 1:10:58
  was thinking. You have to think about Nvidia, which is, I mean, I'm sure Jensen is a long way from handing off the rings to anyone. But, but that's gonna be a real inflection point.
• 1:11:07
  Bryan Cantrill: That's gonna be an inflection point. And I think, you know, it it's you know, sometimes that's needed. I think, like, getting Ballmer out of Microsoft is actually needed for Microsoft Yeah. To be able to to kind of to regain relevance. But the but Ballmer's on a founder also.
• 1:11:22
  And, you know, I it it feels like Amazon has done pretty well in the post Bezos era, but I would point out that, like, the end of Amazon's total dominance is in the departure of Bezos. And, you know, I I definitely feel that, I mean, the the departure of Intel's founders and, and then I think, I mean, Andy Grove mean, I the Andy Grove is just is merely is Andy Grove just very early intel? I'm not sure. But so much of the the the culture, I guess, the very early, I know that he was he was at Fairchild, before going Intel, but so it means super early Intel. Right?
• 1:12:05
  And, and the, the the kind of the the zeitgeist of Intel really shifted after Andy Grove, was very important for that kind of just relentless. Staying relentless and hungry when a company is dominant is a real challenge, and I think that it is really hard when you've got kind of an inherited company and often when you're past, and I so agree with you. I think that like, even though there were, there were, lots of things that I personally liked about Jonathan, I think that he was not in the right spot. And, I I think that if if McNeely, I think McNeely and I mean, hey, like the guy didn't find the company for me, it's like, the guy's got the right to be like, Hey, I kind of wanna go off and do something else at this point. And, but ultimately he had to come back as chairman and it was mess.
• 1:12:59
  And, and you got I mean, someone in the chat is also talking about, you know, the, the that founder led companies can also flounder. Founder led is not the the right answer, and Steve Jobs. I mean, Jobs is so interesting. Right? Because Jobs just kinda shows you jobs shows you whatever it it whatever you wanna see in that founder arc because on one hand, you know, had the the what was forced out of Apple.
• 1:13:26
  And on the other hand, like, there is I mean, surely, it is not a hot or controversial take that if Jobs does not return to Apple, Apple does not return to
• 1:13:35
  Adam Leventhal: dominance. Yeah. Not controversial at all. And, a throwback to one of our earliest episodes about Steve Jobs and the Next
• 1:13:44
  Big Thing.
• 1:13:44
  Bryan Cantrill: Was it?
• 1:13:45
  Adam Leventhal: Next Big Thing. Right.
• 1:13:45
  Bryan Cantrill: Big Jobs and the Next Big Thing. That is such a good book.
• 1:13:48
  Adam Leventhal: It was a great book. I mean, really dispelled a lot of the mythology around Steve Jobs.
• 1:13:53
  Bryan Cantrill: Well, it
• 1:13:53
  it Well, it it Because it's written at the trough.
• 1:13:58
  Adam Leventhal: Yes.
• 1:13:58
  Bryan Cantrill: You know, I I I think it it just shows you and I I because I do think that, like, again, we love the acquired episodes, and and there are I think there are there are 3 or 4 episodes that acquired it about NVIDIA. They did a multi parter on NVIDIA. Yeah. And that but that have you listened to the last one? It's either the 3rd or the 4th one.
• 1:14:17
  The last one listen
• 1:14:18
  Adam Leventhal: to is the one with Jensen.
• 1:14:22
  Bryan Cantrill: Oh, yeah. The last one is like, okay. God. Someone, like, turn on the lights in here. These guys someone hose these guys down.
• 1:14:27
  This is just like, this is this is just too much. This I mean, come on. Guys, this is not I mean, because they're saying things like, you know, CUDA's got, like, a 10,000 year lead. I mean and, like, when you start thinking about that, that's like, come on. It no techno.
• 1:14:42
  They like, now you're beginning to to sow the seeds, plant the seeds of actually the erosion of dominance is like Mhmm. If, like, god forbid, anyone in NVIDIA hear this and get that kind of inherited arrogance because you don't have that. Like, there that kind of a lead doesn't exist. And it's easy to denigrate, you know, AMD's. So I'd rock them or what have you.
• 1:15:02
  But the idea that, like, it's a it is teralogically impossible for AMD to surpass NVIDIA is not correct. And, that would be, in in the chat, there's a bit of debate about whether to lose faith in Gelsinger or not, returning to Intel. I I would say that, I I the the Gelsinger was the heralded return of the engineer to Intel, and let's just say the results have been under well banking. Yeah. Right.
• 1:15:34
  But, but it it it it goes to show you, like, how how these say and I think it is also important. And I don't know if you if your perspective on this has kinda changed over the years, but it it you know, when you're especially when you're new to the industry, it feels like everything feels so intransigent at some point. And then you begin to learn, like, actually, everything is actually, nothing is is totally immutable, and things absolutely do change over time. It's what what makes this industry interesting, and giants do actually fall.
• 1:16:06
  Adam Leventhal: Yeah, absolutely. I mean, the fact that, you know, in our careers, we saw, like, the rise of, of Facebook, of Meta. Or, like, I I don't know. Just like the these companies that I was using Google as an Internet Microsoft in 1999 as this, like, weird search engine or whatever. To To think it was not that long ago that these things that we think of as just pillars were absolutely nothing.
• 1:16:31
  Were nothing. And yeah.
• 1:16:34
  Bryan Cantrill: And then also some of the things that that we thought of as being absolutely dominant, and, you you know, impossible, like, certainly, I felt that way with Microsoft and Windows. It felt that way with proprietary systems. It felt that way at so many junctures. It felt like this thing is so dominant. Nothing will ever erode it.
• 1:16:52
  And then all of a sudden, you wake up, and it's like, wait a minute. Like, that we ever thought that was dominant? And, you know, I think it's that, you know, one of the things that kinda came up online a little bit, but, the, you know, we've got our logos in the, in the Oxide office that are, are, echoes of past logos that I love. And, you know, we've got a a couple of kinda trivia questions in there. I won't spoil out in case you're in the Oxide office.
• 1:17:20
  But one that so we we have one, for example, one that is pretty clear is an oxide logo in the form of a Commodore,
• 1:17:28
  which
• 1:17:28
  I think is pretty great. The you know, but have you and I I don't know. Have you had the pleasure of kinda showing this wall to people?
• 1:17:34
  Adam Leventhal: No more just like watching you do it, but that's sort of
• 1:17:36
  Bryan Cantrill: pleasurable as well. Okay. So maybe you got your own little bingo on Lee. But it's amazing to me. So we've got an oxide logo in the style of the deck logo, and that one's like a generational marker.
• 1:17:50
  And if you are younger than a certain age, you do not recognize it at all. And which to me is kind of shocking because Deck was so dominant for for an extended period of time. You know, IBM was so dominant for for an extended period of time. And the I mean, it's so, these things these things do change over time. And, but it it and it ultimately, like, companies take and it's important.
• 1:18:24
  Right? I also think that feel that, like, that that is a service to innovation. We don't wanna be locked into these things forever, and it's a service innovation where these companies either like, I think it was a tremendous service innovation that Microsoft was forced to really change to research, right, where Microsoft was a real problem for a long period of time, and is now, like, not a problem at all. Are you uncomfortable with that, by the way? With, like, this new It's weird.
• 1:18:47
  Adam Leventhal: Right? It's weird, especially because I, you know, I grew up, like, you know, as, like, a, like, a 8 year old subscribing to Mac user magazine. So, like, feeling angry feelings about Microsoft, you know, just, like, innately.
• 1:18:59
  Bryan Cantrill: You you were like a wild warrior in the
• 1:19:02
  Adam Leventhal: Exactly. You
• 1:19:02
  Bryan Cantrill: were what they they taught you at a very young age to
• 1:19:05
  Adam Leventhal: Yeah. That's right. But yeah. Totally.
• 1:19:08
  But then to say, you
• 1:19:09
  know, there's, you know, I was at an API conference where I was hanging out with a bunch of, a bunch of, folks from Microsoft who were doing things that are, like, right up my alley. And we're doing it all in the open and, you know, wanting to talk about it and collaborate. It's weird. It's weird. It makes me feel uncomfortable.
• 1:19:27
  Bryan Cantrill: Yeah. It is weird. And Great. You know, it is the the the fact that well, so this has hit home for me when we in I wanna say, like, 2017. We had a candidate that was coming out of college and, was looking at, at the time, Giant, versus Microsoft, versus Facebook.
• 1:19:55
  And he decides to go to Facebook. And my attitude is, at least it wasn't Microsoft. Like at least I I I don't have to like I mean, I at least I can like respect him as a as a human being. And the younger engineers are like, what are you talking about?
• 1:20:12
  Adam Leventhal: You're out
• 1:20:13
  of your mind. It's the opposite.
• 1:20:14
  Bryan Cantrill: Your mind. Like, I am furious with him for going to I like Microsoft, like, what's wrong with Microsoft? You definitely should. Like, at least at Microsoft, you can, like, you you can at least, like, you know Yeah. You're not, like, undermining democracy.
• 1:20:27
  Right? You don't undermine democracy, you're not and I'm like, what is going on? And and and this would be this is like in early 2017. And because the that's yeah. It would've been the summer of 2017, and it was kind of a light going on for me that, like, oh, Facebook and the, you know, the younger engineers were, of course, like, much more on this than I was.
• 1:20:48
  And realizing that, like, oh, Facebook has kinda lost its way. Facebook has become that. And and for, you know, for Facebook had had kind of, taken on the mantle of villainy that that Microsoft had had. And now Microsoft just and and it I mean, both of them earned it. Like, Microsoft earned their much better reputation because they've done a lot that's been in service of innovation, and that has also helped the company.
• 1:21:16
  Right? Yeah. So, you know, I think it's it's gonna be really interesting to to watch this whole thing unfold. And we don't I mean, we're very curious. We I I think, you know, we I think people know this.
• 1:21:30
  I don't think we we we've kept it a secret that we are very interested in what an accelerated compute product looks like. But I think we also haven't kept it a secret that, like, we've gotta find a way we can do that in a way that delivers real oxide value. And this is where you get to the challenges with NVIDIA are around this proprietariness makes it really hard for us to deliver the kind of system that we wanna go deliver. And, you know, we I kinda believe that like Intel X86, the company could be really successful and be a lot more open. But, I don't know.
• 1:22:02
  You know? It's I but also, like, why would NVIDIA, like, care about what I think? It's like, I'm sorry. I'm I I oh, oh, yeah. Let's take your advice.
• 1:22:09
  You know? Yeah. Sorry. Did you mention 98% of the value of some Microsoft lost under your tenure? Like, why don't you I turned point on it.
• 1:22:16
  Adam Leventhal: I turned off after you said that. Right?
• 1:22:18
  Bryan Cantrill: Exactly. Like you sold for less than it's cash. Right? Like, okay. Yeah.
• 1:22:21
  Let's let's do what you say. This is like, now I'm channeling my 17 year old.
• 1:22:28
  Adam Leventhal: It it does seem like it's particularly tough to have that kind of insightfulness or that kind of soul searching when you're at your peak to say, hey, yeah, but how do we how do we prevent the next generate or how do we maintain that dominance
• 1:22:41
  Bryan Cantrill: That's right.
• 1:22:42
  Adam Leventhal: After the next generation? How do we keep, the the early upstarts from even getting started? Actually, this is something that Amazon has done incredibly well, despite, perhaps, if you're right, that, the the loss and dominance. But I I think that they they've always been focused on how they, yeah, how they kinda maintain that dominance generationally.
• 1:23:02
  Bryan Cantrill: For sure. For sure. And, you know, Amazon the other thing that Amazon has the other thing that I think is the the you get this loss of dominance when you also take your eye off of your customers, when you what you're delivering to your customers. And Amazon's always done a good job of delivering value to their customers and understanding their own customers. And that's what IBM lost track of their customers.
• 1:23:25
  Sun lost track of their customers. The the the the companies that have lost that dominance, Microsoft absolutely lost track of their customers because they were they were coercive with respect to their customers. The, so the and I think, you know, I I I NVIDIA seems to be pretty focused on their customers right now. It's part of their executing well. Right?
• 1:23:47
  They seem they're delivering things that people want. So
• 1:23:50
  Adam Leventhal: So agreed. But, you know, it seems like they're also focused on a model of renting, not owning, and a full system delivery. And will that continue to be what customers want? And will they be able to hear when it's not? Or will they be insulated from that in a way that I think Sun was, maybe, and now I'm making this tweet's point.
• 1:24:12
  But, Sun was insulated from hearing those the the the complaints from customers, and by the time we did, it was too late.
• 1:24:19
  Bryan Cantrill: Okay. Well, so okay. So this is, I think, gonna be the big challenge for NVIDIA is I do think this is where the valuation comes into conflict of this where you kind of need to ignore the valuation. And this is something that I always thought, like, Google and Amazon were very good at, especially in their earlier days of, like, no, like, we are going to do things that are that you, the market, may disagree with. They that are that we think are in our long term interests.
• 1:24:46
  And I think when the valuation gets to $2,900,000,000,000 it is harder to tell the market of, like, no. Look. Here are the things that we need. Like, I know what you want us to do is turn into a private equity shop and only rent the GPUs, but we will be we will be planting the seeds of of the the of our we'll be undermining our own dominance by doing that ultimately in the limit. We will be more extractive, but ultimately we will lose our dominance if we do that.
• 1:25:16
  And that's the knife edge that that they've gotta go walk. And they and and then they also need to, like, need need to stay focused, which I kinda get I get the sense that they are. I don't get the sense that they, because many companies, Sun included, just get high on their own farts and just, completely lose focus. I mean, god, that's I mean, any any SGI alum would be like, oh my god. I mean, SGI is, like, rise and fall was so quick.
• 1:25:45
  And that that like, it was a very brief period of of time of dominance, but they completely lost focus.
• 1:25:51
  Adam Leventhal: Yeah.
• 1:25:52
  Bryan Cantrill: And and then the wheels came off surprisingly quickly.
• 1:25:55
  Adam Leventhal: Well, and I think arguably hastened by NVIDIA. Right? Like, hastened by the commodity market around around graphics.
• 1:26:01
  Bryan Cantrill: But that that's for sure. Hey. So it wouldn't and also, like, what the the the kind of the rents that you were seeking, charging, being able to extract were not gonna happen going forward. Like, there were structural changes that were and that and that would be an interesting moment. I'll be in a bit.
• 1:26:16
  They just had so many brush brushes with death that, I think that they are in probably better shape to to do that, but it's gonna be interesting. It's gonna be interesting to watch. So I I mean, I I I think that it's reasonable to to believe that this level of dominance will not last forever, but it's also reasonable to believe that it's gonna, it's not gonna end tomorrow.
• 1:26:38
  Adam Leventhal: Yeah.
• 1:26:38
  I don't know. Yeah.
• 1:26:39
  Bryan Cantrill: How how's that for a hot take? It's yeah. Wait. Like, why why why I'm sorry.
• 1:26:47
  Adam Leventhal: Did you say the weather today would probably be the same as the weather tomorrow? I guess that's an impression.
• 1:26:51
  Bryan Cantrill: That's great. Well, this is fun to tune in too. That was that was really great. The, I I I I understand that that sounds like kind of a bit of a of a milk toast take, But I think, I don't know, less complicated. That's where we are.
• 1:27:06
  What do you want from us? You know?
• 1:27:08
  Adam Leventhal: There we go. We're simple people.
• 1:27:09
  Bryan Cantrill: We're simple people. Hey, you know what? We read the tweet, folks. You can't we we just
• 1:27:13
  Adam Leventhal: The whole thing.
• 1:27:14
  Bryan Cantrill: We read the whole thing. We did it without resorting the mocking voice. I mean, I feel like we we made some strides here. We've made some strides here.
• 1:27:22
  Adam Leventhal: It's good progress.
• 1:27:23
  Bryan Cantrill: And Todd had a question for us and then evaporated. So maybe Todd is just like, you know what? Forget these these 2 jackasses. I don't know if they're they're just gonna so sorry, Todd. But, well, it'll be interesting to know why this is a space that we're keeping, I think very, continue to keep a very keen eye on.
• 1:27:42
  Super important, it continues to be important, there's a lot of data there, so it's gonna be interesting to watch. Absolutely. And, I think that that and also this freedom of history is definitely bizarre and incorrect. That's be my other conclusion is that, this tweet did tremendous numbers from this guy, 7.2,000 likes.
• 1:28:04
  Adam Leventhal: Maybe that's the win. Maybe maybe he's just trying to launch a 1,000 podcast episodes about his tweet.
• 1:28:11
  Bryan Cantrill: Maybe that's it. Maybe that's it. I know it's bait. I'm sorry. We chomped down on the bait.
• 1:28:16
  I I I would say that I tried my best to resist, but then I'd be lying to us all.
• 1:28:21
  Adam Leventhal: And I
• 1:28:22
  think that was your best, actually.
• 1:28:26
  Bryan Cantrill: You know what? You you you think that I've got higher expectations. I don't have high expectations for you. I've got low expectations for you. Alright.
• 1:28:32
  Well, I've I'd like to think that I've got higher expectations for myself, but here we go. That's good. We are off next week. Adam, you've got a, a Out
• 1:28:40
  Adam Leventhal: of town.
• 1:28:41
  Bryan Cantrill: Yeah. So it's still sporadic here with respect to, various summer vacations and so on, but we will have everyone back in 2 weeks. Alright. Hey. Thanks, Adam.
• 1:28:52
  Thanks for that. Thanks for indulging me on No. This
• 1:28:55
  Adam Leventhal: is great. I enjoyed it.
• 1:28:56
  Bryan Cantrill: Alright. Take care, everyone.