Latest / Oxide and Friends / The Oxide 3D Explorer

Transcript

The Oxide 3D Explorer · Oxide and Friends

• 0:00
  Adam Leventhal: Ben, can you hear and be heard?
• 0:02
  Ben Leonard: How about now?
• 0:04
  Bryan Cantrill: We can hear you now.
• 0:05
  Ben Leonard: There we go. I think push to push to talk was turned on and I, and I was not push pushing.
• 0:10
  Bryan Cantrill: You go. God, you've, And we know it is it's late in the evening there. I Rain was Rain was saying it was my understanding this was gonna be a Europe friendly time because you pointed out like, look, the kids have gotta get down. You've got def you definitely have the, like, after hours jazz DJ voice there.
• 0:30
  Adam Leventhal: Like you've been screaming at kids Yeah. All
• 0:34
  Ben Leonard: There is there oh, man. I I wish I had, like, a cold or flu, like, five days ago. I'd be I'd be really in the pocket. But it's yeah. It's it's a little bit late.
• 0:45
  It was somewhere in between my daughter's bedtime and my my my and my bedtime.
• 0:50
  Bryan Cantrill: And well, that's good. That's a that's a relief. Presumably, we are after your daughter's bedtime and before your bedtime, but it encourages me that
• 0:57
  Ben Leonard: we should ask at least ask
• 0:57
  Bryan Cantrill: the question. Ben, is there any tradition of father's day in the, in the old country there? Do you guys do that thing? Is that strictly in America?
• 1:05
  Ben Leonard: It's the the same American one actually.
• 1:11
  Bryan Cantrill: Different, but father's day is the same. That's weird.
• 1:13
  Ben Leonard: Yeah. That trip it trips me up every year because I you see everyone talking about it, you're like, oh, no. I've I've missed it. I've forgotten it. And then it's the US one.
• 1:23
  Bryan Cantrill: Well, happy Father's Day. You are before the age where you you will have a brief age where your kids will, like, make breakfast for you. I in my experience, like, definitely like that goes by and that's a very brief window in time. Are you, Adam, did you ever hit that window? Did you, I mean, you've got the-
• 1:45
  Adam Leventhal: Yes, but I wouldn't have like recommend eating it. That's right. No, sorry. Yeah. The phase where
• 1:53
  Bryan Cantrill: you get breakfast made for you and you should eat it like that. If that lasts at all, is if you have a year where that's the case. I am actually at the age where the kids don't say anything to me, but post very nice Father's Day things on Instagram.
• 2:11
  Adam Leventhal: Presumably, do you follow them on Instagram? I do now.
• 2:15
  Bryan Cantrill: I mean, it's like if I Yeah,
• 2:18
  I do. It's this was very it was actually very sweet. So and then you get, you know, the usual, like, who but who had the best memories that they posted with, you know, of the you get the usual civil civil rivalry. But I don't know if that's a
• 2:30
  Ben Leonard: That that is using you for content.
• 2:32
  Bryan Cantrill: They are you that's exactly right. That they are exactly right. And you know what? Chip off the old block, I gotta say. Like, I I I've got I've got absolute professional respect for using me for content.
• 2:41
  Adam Leventhal: So welcome to the podcast, everybody.
• 2:43
  Bryan Cantrill: Podcast, everybody. Exactly. Exactly that.
• 2:46
  Ben Leonard: Have you done a a a and Family podcast episode?
• 2:52
  Adam Leventhal: No, but if we were to do one, experienced the moment that his family would like to come and describe, which was last week.
• 3:03
  Bryan Cantrill: Last week on Thursday.
• 3:05
  Adam Leventhal: Oxide at the Oakland Ballers. You were wondering how long it was gonna take for us to mention baseball.
• 3:10
  Bryan Cantrill: Yeah. On Big O card, think I assume some people are checking that. So
• 3:15
  Adam Leventhal: at the Oakland Ballers, the Ballers have a tradition of something called knocker ball, where you get in this gigantic plastic hamster ball thing. Yeah, orb, like inflated orb and kind of battle it out by bumping into each other. That's right. And because Oxide was sponsoring Little League night at the ballers that night, I guess Brian parlayed that into a father versus son knocker ball competition like after the fifth inning or whatever
• 3:43
  Bryan Cantrill: Yes. It And so it should also be said that my oldest plays baseball and Alexander and I, my middle have in my 19 year old, he and I have watched a lot of baseball together. And this is actually a pretty common inner inning thing. This is not unique to the ballers.
• 3:59
  Adam Leventhal: Okay.
• 4:00
  Bryan Cantrill: So we have seen knocker ball knockout quite a bit. And whenever we watch knocker ball knockout, we immediately there's some speculation about who would prevail in knocker ball knockout. And so this was an opportunity to when the baller said like, look, know, you're sponsoring this part of your entitlement, you get to pick the the the first pitch, which is exciting. So Steve's a 10 year old baseball pitcher throughout the first pitch is great. And then you also get to pick knocker ball knockout.
• 4:29
  I'm like, well, I know. I mean, is it is it too gauche? Can I pick myself for knocker ball knockout? And so Alexander and I were we're gonna go at it. Knocker ball knockout.
• 4:40
  There've been a lot of talk. I felt like I was like the obvious underdog. So I was playing with house money.
• 4:45
  Adam Leventhal: This turns
• 4:46
  Bryan Cantrill: out if you lose, as I learned, if you lose all your house money
• 4:49
  Adam Leventhal: So you still have nothing left.
• 4:50
  Bryan Cantrill: You still have nothing left. Like, on
• 4:53
  the one hand, you're playing with house money. So, Ben, you should know that, like, when also so we get back so we're kind of, like, in the the the the backstage area where the, you know, you wait counting down the outs until you're gonna run out there and do this. As you're trying these things on
• 5:06
  Adam Leventhal: because you only have like a minute and a half to
• 5:08
  Bryan Cantrill: get it done. Minute and a half. It's it's very brief. And you you only have as this is very brief time, so you're kinda getting accustomed to these things. And we start bumping into one another a little bit backstage.
• 5:19
  And I what Alexander described afterwards is like, oh, the first time I bumped into you just a little bit with this thing, I got a rush of confidence. Like, I and and similarly, I can report the first time you bumped into me, I'm like, I'm in very deep trouble. And I am now like, these injury I I'm suddenly actually like, I got a whole bunch of other injury risks that are coming to mind. And so I that was really the priority was not to get injured, which I did not. Did not get injured.
• 5:44
  I did not get injured.
• 5:46
  Ben Leonard: Is this the that you did is this the reason you didn't go up against Steve? Because the the risk was too great to the
• 5:53
  Bryan Cantrill: The the risk was yeah. Yeah. Our insurer won't let us do that. That it's actually a violation. Yeah.
• 5:56
  That's right. The the well, and Steve and I are probably although actually, I think Alexander's actually even weighs a little bit less than I do, it's he's just he's he's quite a bit stronger. I think we we can safely say. So, yeah, we went at it, Ben. And I think we can say that Alexander prevailed.
• 6:13
  Adam Leventhal: Yeah. It was not like a split decision.
• 6:15
  Bryan Cantrill: It was not a split decision. Adam had a video that has my little orb kind of sprawling across the field, kind of spinning. I and so when I came back to the stands, I don't know that I've ever seen you and Bridget and Steve together laughing so hard, laughing from the belly as I did. And it was it really brought me joy that I had brought such joy to three people that are so important to me. So that was as far as I'm concerned, it was a victory.
• 6:49
  It was victory. The look that Steve gave me, I have never seen exactly that look. It was a look that said, I can't believe you came back to the stands after that. I can't believe if I were you, I would have walked home and you didn't. And I and I admire that, but I also feel very, very bad for you.
• 7:08
  So Alright. That's complicated. Yeah. It was a complicated look. Like I said, not never seen before.
• 7:13
  Hopefully hopefully not again, but it was it was it was a lot of fun. So, yeah, that was that was our Knockerball knockout. And it was so we yeah. We can talk about that on our our
• 7:23
  Adam Leventhal: On the family show.
• 7:24
  Bryan Cantrill: On the family show. Yeah. Which would include b Cantrill number one fan. So Yes. B Cantrill number one fan is the one who I famously Alexander created a Twitter account, b Cantrill number one fan, and proceeded to like all of what he thought were my cringiest tweets, which is back when.
• 7:39
  So anyway. But that is not why we're here, bud. We yeah. Exactly. So we are here so, Ben, we and we released this three d Rack Explorer.
• 7:52
  I wanna say it was, like, in May. Right? Maybe a month ago, something like that.
• 7:56
  Ben Leonard: Yeah. Something like that. And
• 7:59
  Bryan Cantrill: if folks have not checked this out, you definitely should. And I don't you want to we know what the images will look like, I think, for this episode, but I think we'll we'll allow people to go check it out on their own because no amount of static image can give this thing credit. This is a allows you to explore the rack in three d and allows you to explore each of its constituent parts. I got to tell you, Ben, it feels like this is a dream that we have always had. Adam, like we've always been like, wouldn't it be cool if you could do this on the website?
• 8:31
  Adam Leventhal: Totally.
• 8:31
  Bryan Cantrill: Right? Mean, I feel like we did this like from earliest days.
• 8:33
  Adam Leventhal: From like, yeah, like January 2020 when I joined, I feel like this is a subject of conversation.
• 8:38
  Bryan Cantrill: A subject of conversation, right? I think I'm not imagining that like, wow, would
• 8:41
  be really cool if you could
• 8:42
  do this. Then you're kind of like, what am I wait a minute. What am I saying? I'm describing like, is like yeah. Like, that that's not possible.
• 8:49
  So, Ben, do you wanna describe kind of the origins of this? And then and then how you built this thing because it's really pretty extraordinary.
• 8:57
  Ben Leonard: Yeah. I think I mean, I've been at Oxide five years now. So the benefit of being at Oxide for so long is there's so many of these long tail projects. I've wanted to do one of these for a while, but I've been kind of acutely aware of the challenges. And I think the thing that was delaying it for me wasn't the production of the app itself.
• 9:23
  It was the preparation of the assets.
• 9:26
  Bryan Cantrill: Interesting.
• 9:30
  Ben Leonard: CAD models are pretty chunky, especially the Oxide Rack is pretty chunky because you have I mean, you got 32 sleds each with a PCBA, each with kind of thousands of if not I know. That's a stab of components, And the model is really complex. Even working in CAD is slow. I don't know if you've ever seen you've ever you've, like, kind of got kind of dug one of the Emmys to screen share, and they're doing something in in the CAD software. And they like yeah.
• 10:02
  They'll yeah. They'll orbit around the model, and then you'll you'll you'll wait the extra kinda few seconds for the the camera to catch up and to do what they actually what they actually kind of have, like, commanded it to do. So it's yeah. Even working in CAD with these things is slow. There's millions and millions of of polygons.
• 10:24
  So and there isn't really a good way to kind of automatically remesh. I mean, that's the current thing. Who knows? Maybe in a year or two, we'll have all of these AI powered remeshing tools. But I don't think it's an easy one to remesh because you have the nature of the way that detail is spread across the model.
• 10:51
  I should really shout out Kennen, who's a 3Diast that I worked with on this. Essentially, you have to go through the CAD model and recreate it all.
• 11:03
  Bryan Cantrill: Oh,
• 11:04
  Ben Leonard: wow. There's a ton of manual work preparing the model. And then even within that, you have kind of your different levels of detail. So we we ended up with model as a fraction of a fraction of the existing of the of of the CAD X bot.
• 11:22
  Bryan Cantrill: What is the CAD export? Because, mean, I just feel I know nothing about mechanical CAD. Like, what is the content of that export?
• 11:29
  Ben Leonard: You'll you'll see the kind of gaps of my understanding. I think, like, the the actual CAD representation isn't a isn't kind of a mesh, but the export produces a mesh that's made up of triangles. That's the thing that you work with. Then we have meshes of varying detail. So our our latest compute sled, Cosmo, we have a lot more we have a lot more detail in the PCB the PCBs.
• 12:04
  I think we'd switched you know better than I, we'd switched CADs we'd switched PCB software. That's right.
• 12:11
  Bryan Cantrill: Yeah. And
• 12:12
  Ben Leonard: that and the kind of side effect of that is we had these we had these really detailed PCBA models that we could use and get a good representation of the inside of COSMO. So if click if you're in the rack the rack is broad, you double click on COSMO and you compare it to Sidecar, there is you can see the the difference of, of of detail that we get on those, on those things. But we we in fact, what we're doing is we're taking the exported mesh, and we're really reducing the detail as much as possible. So we're we're actually like, we're kind deleting things in places, remeshing in places. It's, yeah, it's a real kind of it's a real balance.
• 12:53
  I think we can kind of get into it little bit later. The the the like, the the big struggle here is is getting this thing to kind of run performantly, because there is so much detail. And I think if you remove, it's it's it's hard to justify removing elements, because they have they have, like, they have meaning. They're real elements that how that do things in the in in the rack. So there's yeah.
• 13:16
  You're kind of you're you're trying to strike the the balance of, of detail and and visual fidelity and performance.
• 13:24
  Bryan Cantrill: Well, there's definitely a lot of detail in there. I mean, I think one of the things that I think is cool is you can actually see the art, you can see the silk that's on the PCB. So, I mean, you've got all of these elements that are So you can see, I mean, if you look at the, for example, the Cosmo compute sled, one of the things you can actually see is the compass that we have deliberately put on the PCB to indicate that that we've got a directionality in the rack. And this allows you to see where north is in the rack, north and south, east, west, and so on. So there's like a there's it's an in addition to all the other component detail, there's you've really captured all of the detail in the sled.
• 14:09
  Adam Leventhal: And just on the North South thing, like that's particularly important because we, software, we talk about the different temperature sensors according to these cardinal directions. So depending on what you If you don't know what north is, it can you can be confused.
• 14:23
  Bryan Cantrill: You can be confused. Yes. Yes. That that that's exactly right. So
• 14:29
  Ben Leonard: yeah. There's yeah. There's tons of detail in there. I mean, even outside the Rec Explorer, we did did some sort of render experiments with a with CG artist. And the the great thing is is you can export the PCB as SVGs in its in its distinct layers.
• 14:49
  You can use those as kind of material inputs. You could take kind of one layer and you kind of put it in, you you drop it in as a as a as a normal map to add some depth. So you can kind of in whereas like, you you could drop another one in that kind of controls the specularity kind of how reflective it is. So there's a yeah. For even for kind of the CG, we can kind of we can take those outputs straight from the PCB software and the fact that we can kind of divide it into its the PCB layers.
• 15:19
  We can use those layers to drive the material rendering. We don't do that in the Explorer. I think we just have a limited use of textures. Think there's a couple of approaches you could do. Could either go, well, not ivy.
• 15:33
  You can go high detail geometry. You can go high detail in textures or you could do both. This is fairly high detail in geometry, we're barely using textures. We're using textures for the for the perforations. But, yeah, you're kind of you're the that's that's another thing that you're kind of balancing is is is load time.
• 15:57
  So if you if you're if you got all this geometry detail and you've got all this texture detail and you're throwing in, then then you've got something which is yeah. Which is which which which is gonna take a bit of time to load, which you don't necessarily have has an issue with game development because all of these things are sort of usually, already on your disk. You don't necessarily need to worry so much about load time. But we can actually of the places this came from was we added an interactive Cosmo to the marketing site as part of the homepage redesign. And that was a test bed for some of this technology.
• 16:43
  And that's a place where you're especially concerned about things like load time and performance. Because distinct from the Rack Explorer, the Rack Explorer is this self contained web app. So someone is interacting almost exclusively with the web application. Whereas on the marketing site, it's incidental. It's just on the page.
• 17:07
  So your budget for doing things is lower because you want someone to get kind of halfway down the page and suddenly their browser's locking up and it's getting chunky just because you've got kind of a three d model directly on there. So that was the kind of, that was the, one of the kind of little entry points into the larger Rack Explorer project.
• 17:31
  Bryan Cantrill: And so can then you describe like how, so how did you do it? Like, so you've got, I mean, you've got this huge problem of eliminating detail. You're trying to do that. So there's some manual kind of elements there. Yeah.
• 17:45
  How does this like technically rendered? I mean, I feel this is, and how are you able to do it on, to be able to support different browsers and so on? Mean, how does this thing work?
• 17:55
  Ben Leonard: So we're using Three. Js, which I think like essentially every web experience that uses three d is using Three. Js. It's kind of a high level library for for, yeah, for kind of doing three d work on the web. And so some of that comes some of that comes for free.
• 18:17
  And then so that's kind of we we that's the kind of base library, and then I'm using something called React three Fiber, which is a wrapper for Three. Js, which essentially lets you use Three. Js within the React ecosystem. It's more declarative. Rather than React three fiber, you're adding what looks like a HTML tag, but it might represent a box or a model or a texture.
• 18:48
  It looks much more like a web page that would whereas three. Js, you're kind of doing that through through JavaScript functions. But then the the reason why that's kind of a really great way to work is that you're building this thing just like you would any other kind of web application. Here I am using a I'm using like a separate state library, which kind of contains all the global state. And I think there's there's there's some other context here, which is in the past, I'd worked on this this kind of slightly eccentric three d website that my when my my last my last my last job, I was working for a a branding agency and we had a a client come in and they just they their their request was they just wanted like a like a weird site.
• 19:36
  They just wanted something that was
• 19:39
  Bryan Cantrill: think they for something and I I don't know what and I don't know whether I should ask the follow-up question.
• 19:43
  Adam Leventhal: I don't know if it's like if it's a a designer euphemism or a British euphemism or a British designer euphemism.
• 19:50
  Bryan Cantrill: That that's a very brave question. Believe
• 19:52
  Ben Leonard: it or not, they were they were very inclined. They were they were kind of art directors, and they just wanted something that kind of captured the culture of their company. And they were really they were really kind of they were really clear about it not needing to be a functional, experience. And so there's to describe some some some elements of it, like, you are you're playing with actually, not too dissimilar to the, to the Rec Explorer. You're playing with scale, so you're, you're looking into what is essentially like a we can link in the shows.
• 20:24
  You we you could you could I I wouldn't open it now. I think your fans will start kicking up, but, you're looking into what is essentially like a some, like, like, a museum glass, and then you kinda change camera, and you jump in, and you switch. You're suddenly, you're inside the exhibit, and then you switch, and then suddenly, you're inside a, a noodle cup, and you're a noodle flying through the air. It was it has some it has some interesting interesting interesting ideas. But that
• 20:51
  Bryan Cantrill: was Sounds my whimsical.
• 20:52
  Ben Leonard: Whim am I Whimsical. Whimsical. Yeah. I think it's I think it's safe to describe it as whimsical. This is functional,
• 21:01
  Bryan Cantrill: I think, was your other The non No.
• 21:02
  Ben Leonard: No. No. But I was gonna go make it I couldn't help but make it semi functional. There was sort of, like, a CMS back end, which would, like, drop their portfolio into this into this into this scene. But that I think that opened up the the, my experience with Web three d.
• 21:21
  It taught me that Web three d is there's so much latitude for interesting exploration. Three d is kind of painful. Performance is hard. Right. But the the the thing that really stands out is that this sort of like this idea of like shared state shared UI and canvas state.
• 21:42
  And that's the thing that we're doing in the Rack Explorer. We have this sort of like interoperability with the canvas and the layout outliner. Rather than you interacting exclusively with a canvas or exclusively with this surrounding UI, you can do both. We have this global state that controls, what element am I focused on? Are the specifications that apply to the kind of selected component, all of that stuff?
• 22:09
  And I think you get this kind of yeah. These really interesting really interesting user experience when you have this this kind of shared this, like, shared state between, like, frame UI and and the canvas, just like, say, three d software or creative software that kind of you have that sort of layout liner on the side and you have this canvas that is kind of inter it's kind of reacting to to that that you're interacting with.
• 22:36
  Bryan Cantrill: And then what is your how do you tell Three. Js what what the model actually looks like? I mean, what how is that actually expressed?
• 22:44
  Ben Leonard: So there's a there's the kind of base model which is exported in in a format GLB or GLTF. And then includes the geometry, includes material information. So there's an exporter on the side of the three d software. Then Similarly, there's an importer in three. Js, which is then taking that and recreating it as geometry at load time.
• 23:13
  And there's other stuff like this uses Draco compression, which would just be just be like any kind of other like any image compression that it uses to, like, kind of press and decompress and so that you're not kind of sending these huge these huge model files. It's actually funny enough, like, you get these you get these formats which are kind of which are are not human readable. They're just sort of like machine code. But, not to, not to ring the kind of LLM bell too early, but, they use this stuff you can poke, you can feed into Claude and be like, okay, trying to do this and it's not working. Can peruse the model format and say, okay.
• 23:55
  Your your your material naming is wrong or the way that this is this export setup is wrong. So yeah. It's it can kind of
• 24:03
  Adam Leventhal: yeah.
• 24:05
  Bryan Cantrill: Yeah. I was gonna ask him in terms of, so it sounds like Claude was useful to actually, or I guess what was the journey in this? So you want to experiment first with the website and then did you get, I mean, so obviously performance very, very important. Load time extremely important for that. How did you get that working?
• 24:24
  What was it and what was it like getting it working on the website?
• 24:27
  Ben Leonard: So I should say the, like, the the the initial journeys, as much as this is kind of a project that we've wanted to do for a while, like, the the the kind of impetus was partly Kevin on our marketing team had said that he'd found an agency that do this as a This is the thing they produce, these three d Explorers for kind of hardware companies. And that is kind of like that's that's not just mine. Think it's maybe an Oxide thing, but that's my Kryptonite is is someone coming someone coming along and saying like, they say they can build this. And it's, like and it's only gonna cost $200,000. Like, oh, okay.
• 25:09
  Okay.
• 25:09
  Adam Leventhal: I feel like that's, like, 90% of what you do here, Ben.
• 25:12
  Bryan Cantrill: The yes. Is is, like, is is refused to allow people to outsource. Yeah. No. Absolutely.
• 25:18
  Ben Leonard: Absolutely. Exactly. The
• 25:21
  Adam Leventhal: new swag that I won't mention in detail here was like basically me threatening to actually even better than outsourcing it is if I just say like how hard could Photoshop be.
• 25:30
  Bryan Cantrill: How hard could Photoshop
• 25:31
  Ben Leonard: Don't worry,
• 25:31
  Bryan Cantrill: I went to, I did I I could just do it in Zaggle. I wouldn't need to buy Zaggle? Is that my may I make that up? Zaggle a thing?
• 25:37
  Adam Leventhal: I think you're making it up,
• 25:38
  Ben Leonard: but I'm also using that. I think you may have. I think you may have.
• 25:42
  Bryan Cantrill: The t shirt making site? Is that not the Zazzle. Zazzle should be Zaggle, though.
• 25:48
  Adam Leventhal: It should be.
• 25:49
  Bryan Cantrill: But but it also isn't. And another that was the That that was the three d site. That was the with the whimsical three d site that that Ben did not wanna mention.
• 25:57
  Ben Leonard: Zaggle.com. I would say I'm not alone at this. I think Oxide, I think this is a reflection that affects us all. What's interesting though is that real Oxide value is that we, for better or worse, sometimes we love to build all of our own tools. So the RFD site, the docs, careers, all of this stuff, it makes sense.
• 26:21
  It's like Oxide fundamentally for so long was almost exclusively engineers. That's what engineers do. And I've kind of I've maybe inherited some of that.
• 26:31
  Bryan Cantrill: It's I just mean, I think on all of that stuff no. I know. Honestly, on all that stuff, I think that actually part of the Oxide experience has been some of these things that you think you can't do, you actually can do. And you ultimately, you can do it not just less expensively, although certainly that cheaply, a little cheaper, but it ends up being much better because it's much, I mean, like the RFD site has been, was so important for the way we use and think of RFDs.
• 27:02
  Adam Leventhal: Or just more like load bearing and useful. Like one could imagine this Explorer site in some uncertain future making its way into the product console.
• 27:11
  Bryan Cantrill: Oh, for sure. My God.
• 27:12
  Adam Leventhal: Yeah. And if we had handed this off to a third party, we would have been paying someone else to learn how to do it in the best way possible. We would have not really owned the assets, the ways we would have owned them, but not in the craftsmanship and ownership perspective. And this gives us a latitude to do other stuff with it.
• 27:30
  Bryan Cantrill: Yeah. And I got to think, I think we were ahead of the curve on that one. I think LLMs have now democratized a lot of that kind of stuff, where people do have the confidence to go build some of the stuff their own because Quad can
• 27:38
  Adam Leventhal: go do it.
• 27:39
  Bryan Cantrill: Yeah. You know?
• 27:40
  Ben Leonard: Yeah. The economics there has have have changed a bit. Yeah. It's it's so much it's so it's so much it's so much easier. I also think, like, within kind of software engineering, you have the ability to create your own tools, whereas it's much less prevalent within design because the technology that you might use to create your own tools is different from the technology that you're using day to day.
• 28:06
  Like, you can't you're you're you're not typically designing your own tools. You'd have to if you if you wanna make something like, let's say, like a a a a generator, an asset generator, you'd have to program it, which is different from I mean, most most designers aren't software engineers. So I think there's a wherein it was it was kinda it's something that's kind of deeply cultural to engineers, which is creating your own tools. I think that's starting to happen more with designers because suddenly they have the ability to produce their own tooling.
• 28:39
  Bryan Cantrill: That's really interesting. Because I feel that that has been so important, Ben. That's so much of what you have been able to do is to live at that confluence of design and software engineering where you're actually you you do create your own tools and have. And it has really shown in the results from a design perspective. I mean, the results are just so extraordinary.
• 28:59
  Ben Leonard: There's there's something there's something great about, being able to both design and code that opens all these opportunity. I was thinking about it earlier, which is, if you're if all you can do is design, especially if you're working within the kind of field of UI design, you design these things which look great but maybe really complicated to implement. And then they fall flat in implementation because you pass it over to an engineer, and they're like, this doesn't make any sense. You've you've ignored all the all all the rules. You've ignored, like, the, I don't know, the the box model or responsiveness.
• 29:32
  Then I think if you're a designer who also codes, at least you have maybe some awareness of the limitations and the frameworks, you can design in some sort of intersection of what looks good and what was implementable. I think my niche has been kind of coming through that to the other side again, which is designing things which are really difficult to make and then taking like, relishing the experience of of of implementing them. So, like, one example is the the new kind of the new Oxide homepage. I designed the the above the fold. And as I did it, I thought, okay, this works well for this one specific aspect ratio.
• 30:21
  And I and I just I have no idea how this is gonna work, kind of different aspect ratios. And it wasn't until I think maybe a few weeks before this site went live that I thought, oh, I can just have this. I can I can switch out the number of racks based on the screen size? And so I think like I've come back again. I've come back out the other way, which is like I'm but, yeah, I mean, I'm enjoying the experience of of making impractical things and then just trying to figure out a way of of, yeah, getting myself out of the problem that I've designed myself into.
• 30:53
  Bryan Cantrill: Does this mean that if I'm on a a screen with a very different aspect ratio, I can have more racks on the
• 30:59
  Adam Leventhal: Did you not notice?
• 31:02
  Ben Leonard: Bryan, have you been on our website?
• 31:04
  Bryan Cantrill: Yeah, I've been on our website. It's not always three racks?
• 31:07
  Adam Leventhal: If you're doing it on your phone now, that's like the dumbest way possible. You can't control the asterisk. No, I know. You looking up the website?
• 31:13
  Bryan Cantrill: No. I'm when I'm on the website, I've always thought it was three racks. I've never gotten
• 31:16
  Adam Leventhal: a different You can narrow it down. Can't grow it. Three is the max.
• 31:20
  Ben Leonard: Yeah. Three is the max. Three. I the from the, like, the ultrawide.
• 31:25
  Bryan Cantrill: Oh, look at that. Now there
• 31:26
  Adam Leventhal: are only two of them.
• 31:27
  Bryan Cantrill: Yeah. One of the racks is gone. And now it's gonna oh, now it's just under one rack. I didn't notice that. It's delightful.
• 31:35
  I no. This is a little little oxide dot computer DYN that I did not pick up on. I yeah. I apparently I'm the last to know. This is like, you know, I was the last person doing RFDs and markdowns.
• 31:45
  So we we know that I can be extremely oblivious to my surroundings. Yeah. That's what we need. Okay. So I'm I'm so sorry.
• 31:52
  I I I We should adjust my I'm gonna stop adjusting my browser size now. We we could go back
• 31:57
  Ben Leonard: to I think I know there's a really good point about because we own this thing, we can kind of we can we can adapt it to all sorts of different situations and contexts like the console. I think we should say maybe year or so ago, we were working on a thermal explorer. The discussion was, okay, we have all of this information which is unique to Oxide. We have visibility on thermal information throughout the rack. We want some way of representing that, even if it's not necessarily the most intuitive way that you interact with this data, I think, like, there's there's a there's something to be said about doing something that kind of is interesting and unique and cool just just to do it.
• 32:40
  And I and we I think we, like, we we we kinda started making yeah. We started making decent progress on it, think. And then we got kind of stuck on this this idea of flattening. We had this, like, this thermal information, which sort of exists in three d space. And we're sort of flattening it in one perspective.
• 32:59
  And then also we're flattening it in terms of time. Because what we're we're not just interested in, like, the thermal information. We're interested in how that relates to, like, fan speed. And they're just like there were a few I don't know. There are a few kind of, like, information design problems there, which are really tricky, which we'd I think we we we started on, but we just we didn't we I think we didn't figure out good resolution for.
• 33:21
  But I there's I can see some of that work in
• 33:25
  Bryan Cantrill: Yeah. That's very tantalizing. Yes.
• 33:28
  Adam Leventhal: No. Ben and I have been talking about we talked about that for
• 33:31
  Bryan Cantrill: I think we did this is another one of these like longtime fantasy Yeah. Kind of
• 33:34
  Adam Leventhal: I mean, in particular, I think that the thing that was so exciting, I think is so exciting about doing that is that it combines all kinds of data sources that we uniquely have. And I mean, not to get too excited about the thing that doesn't exist yet, but you show people that and it demonstrates this hardware software co design that literally would not be replicable in any other commercial system.
• 34:01
  Bryan Cantrill: That's right.
• 34:02
  Adam Leventhal: And the utility of it, I'll just set aside because I'm not sure knowing how hot this DIMM is versus that DIMM has it, but there's all kinds of related-
• 34:11
  Bryan Cantrill: Super interesting though.
• 34:12
  Adam Leventhal: Super interesting and I think there's a bunch of related information that one could infer from that.
• 34:16
  Bryan Cantrill: Yeah, that has never felt so real, so possible.
• 34:21
  Ben Leonard: Yeah. In this new three this new three d world.
• 34:24
  Adam Leventhal: Yeah. This third dimension we've discovered.
• 34:27
  Bryan Cantrill: Tell me more.
• 34:29
  Ben Leonard: It's easy to I mean, it's easy to forget that there's a there's a rack on the there's a rack on the other end. I think for especially for developers in that, I'd I think it's which is which is the, I guess, the beauty of Oxide. This developer doesn't need to think about the machine just like the public cloud. And maybe that's why, like, it it doesn't it doesn't exist in the in in the in the console at the moment. But I think, like, it's something that's always in the background that we could we could do a better job of representing the hardware within the within the web console, within the within the software.
• 35:03
  Bryan Cantrill: Yeah. Yeah. I think it is a great idea. I think it's great idea. I also think that it would almost certainly prompt some interesting discoveries for sure.
• 35:11
  For one thing or another.
• 35:14
  Ben Leonard: I think a more a like, a kind of more, like, near term practical use of specifically the Rack Explorer is support. I think you could imagine the tool as a method by which you explain, okay, the disc three on sled 16 is out. You've never done this operation before. You you can imagine a world where you you're kind of scanning the you're scanning something or you're going it through through the web console, and it takes you to the exact point in your rack. And it's and it's walking you through the operation of replacing replacing something within it.
• 35:50
  So because it's completely dynamic. In
• 35:53
  Adam Leventhal: the Explorer, you've got a little aspect of this. I love that when you click on the sled, it doesn't just pop out. It actually shows you the handle animation of how one would remove it. So there's all kinds of opportunities to actually have those kind of repair procedures be interactive and show you like, what is the manipulation of the handle or of the disc latch to remove the thing.
• 36:19
  Bryan Cantrill: Yeah, it is frequent. I also just love the fact that this is not like an artistic rendering. This is actually coming from the CAD. This is like the
• 36:28
  Adam Leventhal: That's true. But there has been some artistic choices. One that I need to quibble with Ben. So on the compute sled, there is like, there's great visibility of the PCB and the silk screen and so forth. What aspect of
• 36:43
  Bryan Cantrill: the computer that Adam feels strongly about has been popping
• 36:46
  Adam Leventhal: up. On the switch, place your
• 36:48
  Bryan Cantrill: bets down.
• 36:49
  Adam Leventhal: On the switch, on the PCB. Yeah. And I'll wait here while listening to the podcast, go to their Oxide computer, remove the switch, take off the lid and go and inspect the PCB. There is an image of a cat on the PCB.
• 37:04
  Bryan Cantrill: But he explained why that's not possible though.
• 37:07
  Adam Leventhal: Why is that not possible?
• 37:08
  Bryan Cantrill: Because this is the Altium thing. Oh, got it. Yeah. This is like your grievance, sir, is with kids. The Okay.
• 37:20
  The but it is it is I I honor the grievance because you're right. There is a it it that you cannot see the silk on the switch. And I assume that is because so we switched from Cadence OrCAD to Altium, and Altium has got a much better integration with SolidWorks. But we haven't ported the switch over to because why we're The switch will never be yeah. In Yeah.
• 37:47
  It will switch will only ever be
• 37:48
  in Cadence
• 37:48
  Ben Leonard: We
• 37:48
  probably could get the text I think the it's the model which is hard to get. We probably could get the texture, but then we I think we need to we need someone with a license to open the file.
• 37:59
  Bryan Cantrill: That's right.
• 38:00
  Adam Leventhal: Yes. Well, or maybe we would use it. No one would notice. I'm I'm just kidding. Employers.
• 38:06
  Bryan Cantrill: It's The Internet will never discover what you just said because we'll be, deleting it. The the, but what what you should describe what you're referring to because I do feel that, like, the cat that you're referring to is an important part of Oxide lore that I don't think we have described.
• 38:22
  Adam Leventhal: No. So, yeah. So I'm not sure I have the genesis of this correct, but I think that there was a we had like some sort of proto board that had, That had a, was it around the battery? The battery. We thought that the battery connector kind of looked, had like a cat like appearance.
• 38:41
  Bryan Cantrill: It looks like
• 38:42
  Adam Leventhal: a cat. And then we had like a silk of a cat on that proto board.
• 38:47
  Bryan Cantrill: We, Cliff put the battery holder, which is labeled literally nobody wants this battery. Right. Because we, when I show that board to like middle scores and you ask them like, what's the problem with the battery? They're like, they die. Yes, they die.
• 39:06
  That is the problem with the battery. The problem is the battery that it can be dead. And if you have to deal with the battery being dead, then you have to deal with not having a battery. So why have battery at all?
• 39:15
  Adam Leventhal: And actually, is like very early Oxide kind of point of principle too, where like many, most computers have a battery.
• 39:24
  Bryan Cantrill: Have a battery and as a result, they know what time it is.
• 39:26
  Adam Leventhal: Right. And we didn't wanna know it. We don't wanna know what time
• 39:30
  Bryan Cantrill: it We're not interested. So don't trouble me with the time.
• 39:32
  Adam Leventhal: Exactly.
• 39:34
  Bryan Cantrill: The So on that board, there was the battery holder looked like a cat. And so Cliff thought this battery holder looks like a cat. So he put the grumpy cat meme next to it.
• 39:47
  Adam Leventhal: That's right. And our founding board member Well,
• 39:50
  Bryan Cantrill: so we presented the photo at a board meeting. And the legendary Pierre Lamont looked at the photo of the board he said, That cat, I want you to change it. And Pierre is singular. And I'm like 98% sure he's joking. Like there's a twinkle in the eye, but like no one wants to know what's that 2% looks like.
• 40:20
  So Pierre was famously the first check-in the YouTube among other things. Was at Sequoia for many years. And the YouTube founders famously terrified of him. And I can only imagine. So Pierre has got a fearsome reputation.
• 40:38
  He's just very direct. So he said, want you to change the cap. And I'm like, do I what the okay. What do I
• 40:45
  Adam Leventhal: do with On this board that literally like is not a customer board.
• 40:48
  Bryan Cantrill: It's not a customer board. It was just our first board. It was a Perto board. Yeah. Exactly.
• 40:51
  Yeah. It's not gonna be anywhere. And so I told Cliff, I'm like, look, Cliff, I'm just telling you. I'm I'm not even sure what to do with this information, but I you know, we showed Pierre the board. He said he wants us to change the cat.
• 41:02
  I've got no idea what to Yeah.
• 41:03
  Adam Leventhal: You're just the middle manager. You're not making decisions, Brian. You're just relaying decisions from one to the other.
• 41:08
  Bryan Cantrill: That's more to think. Listen. I'm just we're just having a brainstorming meeting and I just wanna get everyone together. And also if you give me your status reports, the TPS reports, that'd
• 41:14
  Adam Leventhal: be great.
• 41:15
  Bryan Cantrill: And Cliff's like, well, if we're gonna I think we should change the cat. I think we should change it from grumpy cat to grumpy Pierre. And I'm like, that is a very good idea, but I absolutely need Steve and I need to hold hands on that one.
• 41:28
  Adam Leventhal: Oh, I thought it'd be like, never put this in writing. We never talked about it.
• 41:31
  Bryan Cantrill: You do
• 41:32
  Adam Leventhal: what you think is right. But
• 41:33
  Bryan Cantrill: No. No. No. Exactly. No.
• 41:34
  No. No. We were like, okay. So I I need to get Steve Steve and I need to hold hands and then I'll come back to you. And I'm like, alright.
• 41:40
  I told Steve, I think Cliff's got this great idea. We should make grumpy cat, grumpy pear. He's like, oh, that is a good idea. So the we turn the grumpy cat and the grumpy Pierre. Arian, who designed the Switch, loved the grumpy Pierre story.
• 41:51
  And so on the Switch and this is what Adam is right to point out because cadence or not, it is a it it on the Switch, on the silk on the Switch is a galactic grumpy Pierre for for for the size of silk
• 42:05
  Ben Leonard: on
• 42:05
  Adam Leventhal: a You would not need to zoom in very
• 42:07
  Bryan Cantrill: hard to zoom in to see the the the grumpy Pierre. So the, and I you know, it was would be kind of delightful. We get the first switch back for yeah. Exactly. The Pierre LeMond signature edition.
• 42:20
  We get the the the the the first, switch back, and, Pierre has shown up maybe ten minutes early for a board meeting. And Steve says, hey, Pierre, I gotta go take a quick call. Brian will show you sidecar in the meantime.
• 42:36
  And I'm
• 42:37
  like I'm like, fuck you, pal. I I so I Right into the wood chipper. Right into the wood chipper. And again, I'm like, I am virtually certain that this is gonna be fine. I again, maybe up to 99% certain, but that 1% just feels like a very high consequence 1%.
• 42:53
  So I show him the board, but I am I am pointing things out by putting my hand over the grumpy Pierre. And I am not moving my hand as it is pointing to other elements. And I'm like, I'm just trying to redirect his attention elsewhere and bore him with mind numbing technical detail. And that works. Like, he
• 43:19
  Adam Leventhal: What's the model number of this capacity?
• 43:21
  Bryan Cantrill: Absolutely. And and I usually like, Pierre and I have, I think it's fair to say, a bit of a Dennis the Menace, mister Wilson kind of dynamic. So I think after after maybe two minutes of this, of me having a sustained fight or flight reaction trying to bore him with the details of the VSC seventy four forty eight or whatever else is on that board. He's like Pierre's like, where's Steve? Let's start the board meeting.
• 43:47
  I'm like, where is
• 43:48
  Steve? Let's go get
• 43:49
  that guy.
• 43:50
  Adam Leventhal: Who's that guy?
• 43:50
  Bryan Cantrill: Where is that guy? So, yeah, he's never seen a group appear. And if he goes to the Rack Explorer, he's not gonna see it there either. So, you know, maybe
• 43:58
  Adam Leventhal: I also like that there's you've calculated the odds of him listening to this at zero point zero zero zero zero.
• 44:05
  Bryan Cantrill: I have the odds of here listening to this, I do think are actually at 0%. Yeah. In part because Peter, actually, one of the the actually, at the first board meeting. So when we were in an actual garage, Peter, at some point, I'm you know what? It actually no.
• 44:23
  I mentioned the podcast. I mentioned the On The Metal podcast. Right. And Pierre's like, terrible idea. You shouldn't be doing it.
• 44:30
  And I'm like, okay.
• 44:32
  Ben Leonard: I could do it.
• 44:33
  Bryan Cantrill: Okay. I don't know if I And I'm like, I don't know what to do with that. Because I'm like, I'm definitely I'm actually I'm thinking to myself, like, I'm gonna ignore that because this is very it's it's low cost. It it's clearly, like, I know that this is actually and Seth took me aside after the board meeting and just said, hey, I think the podcast is great. You guys need to keep at it.
• 44:53
  I I was like, okay. This is actually helpful to know that, like, Pierre Pierre's word is not always law here. Right? That that that I can I I can take the bits that I obviously agree with and on on some of the bits that are, maybe I disagree with with?
• 45:06
  Adam Leventhal: I noticed that Seth didn't say that in front of Pierre.
• 45:09
  Bryan Cantrill: Seth did not say that in front of Pierre. Do you think Pierre
• 45:11
  Ben Leonard: listens to any podcasts?
• 45:14
  Bryan Cantrill: No. No. Nope. I no. I I know.
• 45:17
  I I think he doesn't because I and again, I I I really do, like, think the absolute world of Pierre. But he I I almost know he doesn't because despite being the first check into YouTube, one of the conditions for Eclipse investing was that Pierre wanted to speak to each of us individually. The first Well, the first I told Pierre the first thing I said is like, it's an honor to meet a legend. And he just said, Legends are dead. And I'm like, this is not off to the right foot here.
• 45:51
  But it was a he was very he he was was very direct with me. But at one point, I asked him about, you know, hey, I know you like did I ask him a question about when he left Kosla? But he said, how do you know that? I'm like, you described it in the oral history that you did for the computer history museum. He says, yes.
• 46:13
  But how have you seen that? I'm like, it's on YouTube. Like, I don't like, my
• 46:17
  Adam Leventhal: Who do think it was for?
• 46:18
  Bryan Cantrill: I don't he I think he thought that was gonna
• 46:20
  Adam Leventhal: be private For like For the community history museum. Like what like how how private? Like, I've I've think he thought
• 46:28
  Bryan Cantrill: it was gonna be oh, he so in particular, the oral history with Pierre is extraordinary if you haven't watched it. It's like a four hour oral history.
• 46:34
  Adam Leventhal: It's great. Mean, this is like literally this is the reason I became like a YouTube, whatever it is prime
• 46:41
  Bryan Cantrill: subscriber You're or just sick of getting ads through the Pierre Lamont
• 46:44
  Adam Leventhal: or whatever. No, no, no. It was so that I could watch it like as a podcast. As podcast. Could like plug it into just like turn off the screen and like listen to it in my car.
• 46:53
  Bryan Cantrill: Yeah, it is very, very good. That series is very good. His oral history is very good. And especially knowing Pierre now, he said things that absolutely he thought were for history and for a kind of that he did not think were gonna be public. He thought maybe, like, researchers would have access to it, but not Who
• 47:11
  Ben Leonard: gave you that VHS, Bryan?
• 47:13
  Bryan Cantrill: Yeah. Well, so so so in particular, he says something so, you know, Khoesla has got a very bad reputation in VC.
• 47:20
  Adam Leventhal: Yeah. Khoesla, cofounder of Sun.
• 47:22
  Bryan Cantrill: Cofounder of Sun, fired after, like, eighteen months. Yeah. Which I always think is, like, kind of crazy that you're dining out on Sun Microsystems even though, like, you're basically shit canned, like moments after the company has started. But he talks about being at Coastal Ventures and realizing pretty early on that he doesn't wanna stay because Vinod is a very bad collaborator. And So he knew he wanted to leave, but he didn't leave immediately because he says if I left, he would not have been able to raise his next fund.
• 47:54
  Adam Leventhal: So
• 47:55
  Bryan Cantrill: I knew that I needed to wait long enough for him to raise his next fund. And that's definitely the kind of detail that that's nuts. That's very candid. Yes. You don't really get that.
• 48:05
  So, yeah, he did not think it was yes. Not Cole slaw Cole slaw ventures. You know, maybe it should be Cole slaw ventures. No. This is Vinod.
• 48:16
  Vinod a good old Vinod Cole slaw. I feel what is that like the that's that is that like the Muppets version of Silicon Valley? I just that. I just
• 48:32
  Adam Leventhal: Anyway.
• 48:34
  Bryan Cantrill: So where were we? Anyway, grumpy grumpy grumpy Pierre. And and my absolute certainty that that Pierre does not listen to this to the podcast. I'm sure I'm somewhere in the hour. This is Pierre's playing this back for me, and he has now just hit the stop button.
• 48:48
  Adam Leventhal: Not listen to the podcast. Would we? Exactly.
• 48:52
  Ben Leonard: I think we know Pierre's Pierre's a big fan of mine, so I think he will be he'll be listening specifically to
• 48:57
  Bryan Cantrill: Oh, that's true. He does mention that.
• 48:59
  Adam Leventhal: If he's gonna listen to an episode, it'd be this one. Fair enough.
• 49:01
  Bryan Cantrill: It'd be this one. Absolutely.
• 49:04
  Ben Leonard: But So the Adam Yeah. If you if you if if you do have any objections issues, it's a good time to say that the the Rack Explorer repo is open source. So Oh.
• 49:16
  Adam Leventhal: We go. PR is welcome. I get I understand what you're saying.
• 49:19
  Bryan Cantrill: I will. Okay. So the one thing so here I do have one question for you, Ben. You are missing our boss's favorite element of the rack.
• 49:36
  Adam Leventhal: I'm I'm I'm I'm Second favorite. Not drumming No
• 49:39
  Ben Leonard: no pun intended brain.
• 49:41
  Bryan Cantrill: No. He loves the PowerShelf controller. The PowerShelf, but not the PowerShelf controller.
• 49:49
  Adam Leventhal: Not the controller.
• 49:51
  Bryan Cantrill: You can see it. You just can't click
• 49:52
  Adam Leventhal: We on had a customer in
• 49:54
  Ben Leonard: the office today and
• 49:55
  Adam Leventhal: he was literally talking about the PowerShelf controller.
• 49:57
  Bryan Cantrill: Yeah. I'm not making that up.
• 50:00
  Adam Leventhal: Rounds of applause. Two rounds
• 50:02
  Bryan Cantrill: of applause. No. He it's a good know, I I have I I've I've seen it many times and it never gets old. Steve, on the power shelf controller. But there is you can see if you go to the back of the power shelf, you can see that little goober off to the left.
• 50:16
  That is the PowerShelf controller.
• 50:19
  Ben Leonard: It's not it's not directly selectable.
• 50:22
  Bryan Cantrill: Is this a test to see if he actually has used it? Because I can tell you he has. He loves this thing. I've actually watched Steve when we were we were with someone who's very interested in Oxide and we he and I intended to bring a sled and have forgotten to. So we pulled up the Explorer and it was great to be able to watch them mosey around at it.
• 50:41
  But not to
• 50:41
  Ben Leonard: be able I mean, because there's there's the great thing is there's so there's two modes. There's the kind of free explore mode where you can you can just poke around and click click on click on whatever. If you're on your phone, you'll be kind of ushered into the guy like, guided mode. The great thing is the way that this is kinda constructed, the there's like a componentry, which is represented as a as a as JSON file, which lists all the different components and then the models and their locations. Then similarly, there's kind of a separate file that has all the tours in.
• 51:19
  And all that's happening is when you're moving between points on the guided tour, whether it's the video one or the text one, it changes the current selected ID and state. And then the camera automatically interpolates between those waypoints. So it's really easy to to kind of produce to to to have it to have it kind of move from place to place. Like, I did I did the kind of text version first, and then I I saw I saw your kind of the FAQ Friday on the PowerShelf.
• 51:52
  Bryan Cantrill: Yeah. I
• 51:52
  Ben Leonard: thought, oh, this could be kind of a nice addition to work this directly into the Rack Explorer. That implementation is mega simple. It just there's a within the TOS file, a JSON file, has an array of steps, each of which describes a location on the rack and a timestamp. And then it will automatically rotate, get the camera in the right position, fit to the to the to the screen. Yeah.
• 52:25
  That that all happens automatically regardless of whether you're kind of clicking on something or whether you're kind of programmatically changing what is, yeah, what is selected.
• 52:35
  Bryan Cantrill: That is really cool. And it makes it feel very fluid. Oh, and there's look at the you got the video for the FAQ Friday. Look at this.
• 52:42
  Ben Leonard: Yeah. So this and it's and it's it's really trivial to extend. It's really trivial to add kind of new, yeah, new new new pieces of content. One one interesting thing as as we're talking about, like, the way the kind of the the camera movement, the camera interpolation is, I had I had a kind of a Claude skill installed that applied specifically to web animation principles. Yeah.
• 53:11
  And and bear in mind, this is designed for, I don't like animating a modal or a drop down or something like that. I But was interested to see how deal it with applying those same principles to a three d canvas. So I used that and it had some really good insights. So one of them, for example, is it's like the the camera the camera linear linear interpolation was was frame rate dependent. So so kind of had that, that it was that it was something that it was kind of part of the skill, but just for kind of regular animations.
• 53:51
  And similarly, it had this insight around the duration of the animation between states could scale with travel distance. So it'd be a slightly slower animation if you kind of moved further. And that's designed for, yeah, I don't know, like a dropdown should animate on quicker than a kind of page model, but kind of helpfully thought that that might also be relevant to the way that the camera was and it was kind of, yeah, it was absolutely right. So I think there were really interesting moments when you apply these kind of traditional UI principles to non traditional user experiences.
• 54:34
  Bryan Cantrill: Yeah. And I just, I mean, this thing is so pedagogically valuable. One question I've got for you is how difficult is it to add like explanations for different components? Because I feel like there's so many components that you can see on here that we could offer like a more detailed explanation for. Is that how how hard was that to add what you've got and how hard is it to add to it?
• 54:56
  Ben Leonard: There's there's there's two things. One is the meshes themselves. So, like, the meshes the meshes are sort of grouped together into, like, distinct assets because you need some way of, you need some way of targeting those. So you'll see the kind of green outline. Yeah.
• 55:16
  And so, like, the over the whole, like, rack, like, metal, like, core is one mesh, and then a the the chassis of the Cosmo is one mesh, and then the internal PCB is one mesh. But then when you start needing to kind of target and animate some things, discreetly, then you you start you start needing to prepare the the assets slightly different. So you mentioned the kind of PowerShelf as as one example. That's just one single mesh. There isn't necessarily a good way of say say outlining just the PowerShell controller unless you then isolated that, exported that separately, and then kind of had those rendering independently, which isn't necessarily an issue, but there are broader implications.
• 56:04
  So within the componentry, I have these kind of waypoints. Actually, tours, the tour waypoints, can either target a component ID or I can target a point in space and approximate scale. So there's a way around having to really kind of finally prepare the assets. So in your kind of FAQ Friday where you're talking about the PowerShelf, there's a bit when we're talking about the bus bar and that isn't a separate mesh. But within the guided tool, I'm specifying like a direction, a target and a scale without necessarily needing to be able to needing to split that up into its own its own mesh because that has that has a a performance implications.
• 57:01
  Every time you're kind of breaking
• 57:02
  Bryan Cantrill: these things
• 57:02
  Ben Leonard: into things that that yeah. That that you're adding extra extra kind of draw calls.
• 57:12
  Bryan Cantrill: Right. Right. Because I but it would be that there's because there's so much you can, I think I mean, I I like being able to use the sled to show people physically, you know, here's where the service processor is? You know, here's where the temperature sensors are. Here's where the IBC is.
• 57:26
  Here's where the NIC is. I mean, it just it helps for people to and we've got you can see all these things on here. It's really neat. And you can I also love that if you are on, like you zoom in on some of these views and you can see some of the interior parts that you can't otherwise see? So, you know, if you if you are if you only look at the cooling, for example, it will highlight the fans.
• 57:50
  And then you can really see though the shark fence, for example, because the rest of the storage has been kinda clamped off so you can actually see the shark fence.
• 57:56
  Ben Leonard: But Yeah. We kinda we we lock off, for example, the at at the moment, we don't have, like, the full, like, network switch modeled. But, like, the the underneath there is is, like, is is interesting and important. So, yeah, I need I need to once once we have those bottles prepared, we can kind of, yeah, let people, yeah, peruse the underside of the of of the switches.
• 58:20
  Bryan Cantrill: Yeah. It's very cool. And then what did you have to do in terms of like making this thing perform? Because it does perform really, really well. What are the things that you needed to do or was it mainly in this kind of curation and elimination of detail from the mechanical CAD?
• 58:35
  Ben Leonard: Yeah. There's there's there's tons of stuff, I think, of, like, varying impact. So makes heavy use of instancing. So what that means is also we have 32 compute sleds, for example. And if you kind of if you instance them, you're reusing some of the computational exercise from one sled across all 32, and it's a lot cheaper to render than 32 copies of the same sled.
• 59:04
  Or when inside a Cosmo, for example, we're instancing reused components. Know we might have a bunch of transistors in there and then we're reusing that one model however many times. And that's a feature of the export that you can do. We're also culling the things that aren't in view. So if you're looking at the top half of the rack generally, there's a frustrum that it uses render just the things that are being viewed.
• 59:35
  We do a GPU tier check. So there's a library that we're using that does, it's kind of, I mean, it's relatively blunt, but you're kind of using, I guess it's a media query to figure out what device someone is using or what graphics card they have. And then it will then put you in three buckets. And based on the tier it puts you in, will then adjust the variance to It will kind of reduce the quality if you're on a lower tier device. So there's there's there's like an initial guess it will do.
• 1:00:11
  So think I don't know. If you're on like a silicon like an Apple silicon device, it'll assume that maybe you're in the top tier, and then it will give you the kind of full version. If you're in like, yeah, lower tier, you'll kind of have, it will turn off. There's like a ambient occlusion, is used to kind of do shadows. It will kind of give you reduced materials.
• 1:00:33
  It will like reduce resolution, all of these things. So that's something that it does when you land on the webpage. It will also monitor the frame rate. So if your frame rate drops below a certain threshold a certain amount of time, it will then drop your quality level. And then it'll do it maybe again, it'll do it again.
• 1:00:55
  And then there's some stuff to stop oscillating so that when you go down your frame rate jumping up, it's not suddenly bumping you back up. But there's yeah. There's some stuff there that if you're on a if you're on a on a an older machine that it's not gonna completely yeah. It's not gonna kinda get completely fried. But, yeah, that's a bit of kind of there's a bit of experimentation there.
• 1:01:18
  Funny enough, so I got Claude to write a performance harness. And what it does is you add like a and this is the thing because I have all sorts of theories about ways in which to make it faster and Claude does too. And like the way in which we test this is there's a, you
• 1:01:37
  Adam Leventhal: kind
• 1:01:37
  Ben Leonard: of add these URL prompts and what it will do is you'll kind of, you'll just reload and it will run a bunch of tests for a certain amount of time. It will then save a JSON file of like the outcome of that test. And then you might do it again. You might tweak something to do it again. You throw both of like of those like test outputs into Claude and you say, Yeah, what do you think?
• 1:02:01
  And it will say, Okay, I thought this was gonna be more performant, but actually it's not. It will say this did have a meaningful impact. I think that's good not only for optimization, but that's the way in which I can kind of avoid performance regressions. And interestingly, I was trying on my my wife's laptop, I was getting, like, less than one frame per second. Like, it just was just was not working at all.
• 1:02:32
  I ran the I ran the kind of performance harness, saved the file, dropped it in Claude, and it told me that she didn't have that her browser didn't have hardware acceleration turned on. I was like, wow. I turned it on and suddenly suddenly was working. Bear in mind, is a is a is a problem that I've we've noticed on on her laptop for a couple years now. I was like, why
• 1:02:56
  Adam Leventhal: is this laptop
• 1:02:56
  Bryan Cantrill: so slow? Right. Turns out, you've powered off an important component of it. Yeah. Yeah.
• 1:03:03
  Yeah. Why is my audio not working? I think it's often because, like, well, as it turns out, you kicked out the plug. So Yeah. That that's that's If really
• 1:03:13
  Ben Leonard: you if you look at there's a there's a perf dot m d file in in that in that repo, and it describes it describes determinations, it describes things that we tried that didn't actually make a difference. So yeah, there's loads of stuff. That was really useful because I think you start a new session and suddenly you're kind of the LLM is, is I don't know, is trying the same things it tried this this time last time. But, yeah, it goes through and it sees it can see what it's tried already and it can it can it can continue where it where it where it left off.
• 1:03:49
  Bryan Cantrill: Yeah. That is really cool. And then meaning and I mean, we've been saying this on for any number of things, but that kind of performance harness is an example of the software that's just like, no one's writing that. It's it's not like Claude is putting someone out of business doing that. It's just like we that is just software we literally would not have written.
• 1:04:07
  It would not have been written to use the past space without without the I
• 1:04:11
  Ben Leonard: I think I started this when when the, like, last gen or I suppose two generations of models when kind of Sonnet was was was around. I I really did not find it. And the same for the kind of ASCII Explorer that we've that I that I worked on. The kind of ASCII tool that worked on in both cases, really didn't find GLOD particularly useful when working these things. And then yeah, then Opus came out and suddenly it was a it was like a kind of a force multiplier.
• 1:04:43
  So the kind of beginning of of both of those repos were like handwritten and then increasingly less so as as as time went on.
• 1:04:55
  Bryan Cantrill: Well, is, it's really cool. And what is the reaction to it been? Because I mean, I know we we threw it out there and, I didn't know that you had folks in the designer community certainly looking at it. And obviously, we we at Oxide looked at it, but with what have folks, said about it?
• 1:05:11
  Ben Leonard: Yeah. I think I I think peer I think people really like working people they're kinda working looking kind of real physical things, especially in, like, yeah, the current tech landscape. It's we have a we have a, like, a a unique proposition. Like, we make something that's that's that's that exists in the physical world and yeah. But I think people are always really, really excited when they get to when they, yeah, when they get to engage with that.
• 1:05:38
  Bryan Cantrill: I think so too. And I think that like the the physicality is obviously really important. But then I think to have an intuition for that physicality, you often need to have it physically in your hands. So I do think that like what you've done here is a very important thing to allow people to really appreciate what's been done. It's like, just am kind of like, there's so many things that I think we can go add to it as we're explaining why we did, you know, there's so many different design choices that you can better appreciate with the the Explorer than you can just kind of on your own or listen to a podcast or listen to us describe it to actually kinda see it in front of you.
• 1:06:17
  You can definitely appreciate some of these design decisions.
• 1:06:20
  Ben Leonard: I mean, even even the addition of There's
• 1:06:22
  Bryan Cantrill: a lot of potential.
• 1:06:23
  Ben Leonard: Of Cosmo to the to the homepage. I mean, you you've you've probably run into some version of this, which is, like, the people saying, you you do hardware and software. And and it and it sometimes it feels like no matter what you say, people will come away thinking you just do
• 1:06:38
  Bryan Cantrill: one of the other things. Oh, absolutely. Absolutely. You know, it reminds me of, know, when I, and you appreciate this Ben as a parent and coming on Father's Day, we're after it. When my now 21 year old was an infant, I was walking him in Noi Valley and the and a man stopped me and said, that baby is gonna freeze to death.
• 1:07:03
  You've gotta bundle that baby up. It's gonna freeze to death. I'm like, freeze to death. Okay. Okay.
• 1:07:07
  I'm like, okay. I think it's fine. Everything looks fine. But thank you for your unsolicited feedback. I walked not a half a block and an older woman stopped me and said that baby is gonna overheat.
• 1:07:21
  That baby is gonna and I
• 1:07:23
  Adam Leventhal: That sounds like San Francisco.
• 1:07:24
  Bryan Cantrill: Oh, totally. And I literally had changed nothing. And I I I said to her, do you see that man up there half a block ahead? He thinks this baby is gonna freeze to death. And I think you two should get together and then like, let me know.
• 1:07:36
  I think the conclusion for that may be that the baby's gonna be fine. I think that actually the baby And I just feel it's like, this is like with the website, like the baby is like too bundled up or not bundled up. With the website, I always find that people and I admittedly I could get it like there's a lot to understand about Oxide, but whenever the light goes on for someone, they immediately blame the website for not having turned I mean, I feel bet you must get a lot of this where it's just like, wait a minute, you do software? Like, why don't you talk about that on the website? It's like, we kind of do.
• 1:08:05
  We definitely talk about it
• 1:08:06
  on the website.
• 1:08:07
  Adam Leventhal: I mean, only if you read it.
• 1:08:08
  Bryan Cantrill: Only if you read It's a lot of words.
• 1:08:10
  Adam Leventhal: It's like that's fair.
• 1:08:11
  Ben Leonard: People are bringing their own, yeah, pre con preconceptions. If I I could start a therapy group for designers that are working on that have worked on marketing sites. Regular, I'll talk to people, and he's like, is the this is the the most, yeah, stressful thing to work on. The homepage specifically, the other pages people don't people people don't really care about, but home pages are a they're a
• 1:08:34
  Bryan Cantrill: They're yeah. Interesting. When I think it's in especially for for us where there's like, there's just a lot. There's just a lot to talk about. And it's not trying to express that.
• 1:08:47
  I mean, one point, actually, as long as we've been talking about Pierre a lot this episode, at one point, Pierre was just said, Bryan, what is the differentiator for Oxide? The one differentiator. And I'm like, Pierre, you're not gonna like this, but there are several differentiators. It depends on who you are. Took a clean sheet of paper, and as a result, there are a bunch of different things that we've done.
• 1:09:11
  We've innovated on a bunch of different axes. And if you find For any one of those axes, you will find someone from like, this is like, I don't know why you guys talk about any of that other stuff. And you guys should only talk about the security. You should only talk about the energy efficiency. You should only talk about the operability.
• 1:09:28
  You should only talk about the cabling. I waited one customer. They're like, we like The thing that's interesting about this is the cable backplane. Like you guys shouldn't be talking about anything else. You're like, okay.
• 1:09:37
  All right. Well, guess what? I think we're gonna But this is one of the the Rack Explorer is so helpful, Ben, because it allows people to go kind of explore it all and get some But it's tough because there's a lot to explain.
• 1:10:05
  Ben Leonard: Sip listen. It's late. The my cross to bear is is to figure out better ways of explaining what we are
• 1:10:14
  Bryan Cantrill: and Yeah. And we Yeah. Yeah. The feedback I get was like, Do guys sell one rack or two racks or three racks? Every time I go to the website, it's a
• 1:10:21
  Adam Leventhal: different answer.
• 1:10:21
  Bryan Cantrill: It's a different It's a different number. And I don't know how many racks you sell. And I don't even like, you guys can't even get that story straight. That that delightful story. Ben, is Ben, this is really awesome and really great work.
• 1:10:35
  I feel like it's been a again, this is like a long time dream of wouldn't it be neat if and I just love the fact that this is the that we are taking I mean, yes, curating the detail and so on, but, like, it is ultimately the ground truth of the mechanical CAD that we are showing here, which is makes it really cool. I mean, it is this it is like the actual thing we're building that people can go explore on their own. So this is really terrific work and mesmerizing stuff. There's a lot that we can go learn, lot we can go do with this. I think that now we've got this thing.
• 1:11:08
  I think we're seeing the power of it as a platform. I think there's a lot we can go do to explain what we've done and why in terms of like the actual thing. And Ben, I will also say, and I do try to pass this on to you whenever I hear it. But frankly, hear it so frequently and so frequently out of time zone that I don't do a very good job. I would say oftentimes people say, hey, whoever your designer is, they do terrific, terrific work.
• 1:11:38
  And anytime anyone has seen any aspect of Oxide's design, Ben, that is really due to you. So on really every aspect, thank you very much for your just terrific design sense and your design ethos very much permeates the company. So
• 1:11:59
  Ben Leonard: Thank you so
• 1:11:59
  Bryan Cantrill: just trying to butter you up to to make some of these historic logo t shirts that I'm that that that I'm pining for. So we're gonna get we're gonna get more swag.
• 1:12:07
  Ben Leonard: I'm getting straight on it.
• 1:12:10
  Bryan Cantrill: Yes. That's right. Very good. Or don't make me go to Zaggle or Zinkle or whatever it is. Yeah.
• 1:12:19
  Exactly. Don't make me Be
• 1:12:20
  Ben Leonard: from Lion King.
• 1:12:22
  Adam Leventhal: Yeah. That's right. Or or
• 1:12:24
  Bryan Cantrill: maybe maybe I should have been maybe I should have been like, look. We'll bring in a a t shirt house for $200,000. Maybe, like, I was, like, just going to No.
• 1:12:32
  Adam Leventhal: That's right. That's right. Like, I get that you you have more important things.
• 1:12:35
  Bryan Cantrill: You have important things to do. Exactly. Is
• 1:12:37
  Ben Leonard: this is This is No. No. Give me in a in a costume. Yeah. $200,000.
• 1:12:43
  Bryan Cantrill: Benin a costume. Exactly. But Ben, thank you very much. Thanks for joining us. I know it's late over there, and I hope those emergency vehicles got to their destination.
• 1:12:56
  I I assuming that that was on assuming that was on your end. I wasn't sure if they were lost or if they Just circling the block. Circling the block. I also feel that like, and maybe like every World War II era vintage movie has lied to me, but I thought European sirens had them, had a different sound.
• 1:13:15
  Adam Leventhal: What do they sound like? Do you think Brian?
• 1:13:18
  Bryan Cantrill: I'll be getting to the side. That's a little bit of a. I think that's like that's like
• 1:13:23
  Ben Leonard: the I
• 1:13:25
  Bryan Cantrill: feel like that's like You the
• 1:13:25
  Adam Leventhal: should've supported me. Yeah. Did. I feel like you had a gendarme racing in a narrow a narrow cobblestone street. Yeah. Let's get out of here before I need to write an apology letter. This time to all of Europe.
• 1:13:37
  I feel like I'm in a Dan Brown novel.
• 1:13:37
  Bryan Cantrill: brother? Apology Another letter to this time of all
• 1:13:42
  Adam Leventhal: That You have to read aloud.
• 1:13:44
  Bryan Cantrill: I've gotta read aloud.
• 1:13:46
  Oh, god. I'm so sorry. It's come to this. Dear Europe. Well, Ben, thank you, Exactly.
• 1:13:53
  I the I I you know what? I I'm I I won't be tricked into insulting our our our French neighbors. The oh, I think we were out next week, Adam.
• 1:14:03
  Adam Leventhal: That's
• 1:14:03
  Bryan Cantrill: right. But back the next week. So looking forward to And we'll come back to you, not in this European somewhat friendly time. Less European hostile time. That's right.
• 1:14:14
  This is what we call this. All right. Thanks, Ben. Thanks, everyone. Talk to you next time.