Latest / Full Tech Ahead / Off-Grid Data Centers

Transcript

Off-Grid Data Centers · Full Tech Ahead

• 0:19
  SPEAKER_00: Hello and welcome to Full Tech Ahead.
• 0:23
  SPEAKER_00: This is season two, and I'm so excited to be here today with the CEO of Ferveret, Reza Azizian.
• 0:30
  SPEAKER_00: How are you doing today?
• 0:32
  SPEAKER_01: Good, Amanda.
• 0:33
  SPEAKER_01: Nice to see you.
• 0:34
  SPEAKER_00: Nice to see you as well.
• 0:36
  SPEAKER_00: Can you first off share with our audience what is fervirate and what does it mean?
• 0:43
  SPEAKER_01: Sure.
• 0:43
  SPEAKER_01: Ferverate is a Latin word for boiling.
• 0:47
  SPEAKER_01: And that's a kind of fundamental uh process that we use in our technology, and that's how we came up with the name Ferverate.
• 0:56
  SPEAKER_01: Uh by the way, I would be curious to know what you would have thought about the name and like uh what you were thinking the name means when you when you heard it first.
• 1:06
  SPEAKER_00: I really wasn't sure.
• 1:08
  SPEAKER_00: That's why I asked.
• 1:09
  SPEAKER_00: What I can you share a little bit about what your company does then based on that name?
• 1:15
  SPEAKER_01: Absolutely.
• 1:16
  SPEAKER_01: Um, so we developed liquid cooling technologies for data centers and AR factories, which is inspired by um nuclear reactor cooling, interestingly enough.
• 1:27
  SPEAKER_01: And I can kind of always give you the background, the story of how we got to started and uh and how we are here.
• 1:34
  SPEAKER_01: But in a nutshell, that's what that's what FairWet does.
• 1:37
  SPEAKER_00: Okay, great.
• 1:38
  SPEAKER_00: Well, I would love to hear more about a recent study that y'all did, I believe, with UCLA, I think.
• 1:46
  SPEAKER_00: Can you tell me a little bit more about that study and what it involved?
• 1:50
  SPEAKER_01: Yeah, so um we have done a benchmark study in collaboration with uh Department of Computer Science and Engineering at UCLA.
• 2:00
  SPEAKER_01: Um, we have tested basically H200 GPUs, which is kind of one of the latest NVIDIA GPUs.
• 2:08
  SPEAKER_01: It's like one generation before the Blackwell GPUs.
• 2:12
  SPEAKER_01: And the idea there was to basically benchmark Ferret's adaptive face cooling solution in comparison to latest and greatest directorship cooling solution that that industry currently uses.
• 2:27
  SPEAKER_01: And and then the and the results were um really interesting because uh we find out that with Ferret's adaptive face cooling solution, you would be able to get 15% more teraflop per watt out of each server.
• 2:44
  SPEAKER_01: That means you would be able to get 15% more compute for each watt of power that goes to your server, which is quite significant, right?
• 2:53
  SPEAKER_01: Because right now people are fighting to get more and more compute, and accessing to more power is not that easy.
• 2:59
  SPEAKER_01: So you need to basically squeeze as much power as you have to get compute out of it.
• 3:05
  SPEAKER_01: And we have seen that there was 15% improvement at the server level, and on top of that, we have seen significant improvement on the PUE at the facility level, which, if you combine that improvement with these 15% in total, in whole kind of data center and AI factory, we would be able to deliver 35% more compute out of the same power envelope.
• 3:31
  SPEAKER_01: Which, if I give you an example of, you know, in in 100 megawatt data center, that can translate into almost 230 million of more revenue per year, which is quite significant.
• 3:43
  SPEAKER_00: Oh wow.
• 3:45
  SPEAKER_00: And how do you plan on applying this knowledge that you gained moving forward?
• 3:52
  SPEAKER_01: Yeah, so so we have different partners that that that we are working with on the industry, both from the OEM and O and ODM side.
• 4:01
  SPEAKER_01: And we have some potential customers that we are doing pilot right now, and then we are planning to scale our solution with them.
• 4:08
  SPEAKER_01: The idea again is is is very simple.
• 4:11
  SPEAKER_01: Um, you know, all these data center operators um have access to uh constant amount of power, right?
• 4:18
  SPEAKER_01: Constant power envelope.
• 4:20
  SPEAKER_01: And the idea is basically to get as much compute as you can uh from that constant power envelope.
• 4:27
  SPEAKER_01: And and we help them to do that.
• 4:28
  SPEAKER_01: Like you understand, 35% is is quite a bit of more compute that you can get from the same power contract that you have.
• 4:36
  SPEAKER_01: So we are working with different partners and with different customers and try to bring your solution to the market.
• 4:42
  SPEAKER_00: Yeah, absolutely.
• 4:44
  SPEAKER_00: Well, we keep hearing about more and more data centers.
• 4:47
  SPEAKER_00: The need for data centers is growing exponentially.
• 4:50
  SPEAKER_00: Uh, we actually have some coming into my city and uh neighboring cities.
• 4:57
  SPEAKER_00: And of course, there are concerns that come with data centers moving into moving into cities.
• 5:04
  SPEAKER_00: Um, a lot of people are worried about the water usage.
• 5:08
  SPEAKER_00: Uh, from your experience, um, can you just share what knowledge you have around that?
• 5:12
  SPEAKER_00: Can you compare that water usage to anything else that might be in a city?
• 5:18
  SPEAKER_01: Yeah, I mean, I I think a good example there would be to compare the water consumption of a typical data center to like kind of volume-wise to a swimming pool that you guys have in the city, right?
• 5:32
  SPEAKER_01: Olympic-sized swimming pool.
• 5:34
  SPEAKER_01: And just to give you an idea, a typical 100 megawatt data center can consume up to 4,500 Olympic swimming pool uh of water per year for cooling, which is quite significant, right?
• 5:50
  SPEAKER_01: The example that I always give to kind of people at my team is that we already went through the industrial revolution, you know, and and and we did that kind of unresponsible way, because we went through the industrial revolution, all the GDPs of different countries went up, you know, the life became more prosperous, but no one was thinking about the environmental consequences that that we were dealing with.
• 6:16
  SPEAKER_01: And and now till today, we are still dealing with those consequences, right?
• 6:21
  SPEAKER_01: All the global warming issues that we have is still kind of the tale of that industrial revolution that we went through.
• 6:29
  SPEAKER_01: And I truly believe that now we are going through AI revolution or or digital revolution, if if you wish.
• 6:35
  SPEAKER_01: And we need to do it in a sustainable way, because if we don't do it, we similar to the industrial revolution, we would need need to deal with all these consequences for the years to come.
• 6:48
  SPEAKER_01: So we truly believe that, and that's the mission of Ferret, is to enable AI revolution in a sustainable way.
• 6:57
  SPEAKER_01: Uh, and as a matter of fact, our solution basically works with the closed loop.
• 7:02
  SPEAKER_01: So it the water consumption is basically almost zero with with our solution.
• 7:07
  SPEAKER_01: So we kind of do the AI revolution in a very, very sustainable way.
• 7:13
  SPEAKER_00: Yeah, that was my next question, as I know that was for a regular data center with the original cooling methods.
• 7:19
  SPEAKER_00: So now uh we are hearing about closed loop systems.
• 7:22
  SPEAKER_00: We're hearing more about that.
• 7:24
  SPEAKER_00: So tell me more about the closed loop system and your solution and how it works better.
• 7:30
  SPEAKER_01: Yeah, absolutely.
• 7:32
  SPEAKER_01: So a typical um data center, right?
• 7:36
  SPEAKER_01: Uh use evaporative cooling uh both for air cooling and also for the director chip.
• 7:42
  SPEAKER_01: The evaporative cooling is as the name implies, they basically evaporate huge amounts of water to take advantage of evaporative cooling and and cool down the air that they are using in their data center.
• 7:55
  SPEAKER_01: The way that we do it, I mean, our system doesn't have any airflow, right?
• 7:59
  SPEAKER_01: The whole server is kind of contained in a in a in a sealed chassis, and and our system works with the closed loop, which is just the dry cooler outside.
• 8:08
  SPEAKER_01: Our system can work at um liquid inlet temperature of up to 45 degrees C, which kind of make it suitable for most environments without using any evaporative cooling.
• 8:20
  SPEAKER_01: So the water consumption with our solution is is basically zero.
• 8:24
  SPEAKER_00: So that so once you've filled up with the amount of water needed, you that stays in there and doesn't evaporate.
• 8:31
  SPEAKER_01: Correct.
• 8:32
  SPEAKER_00: Correct.
• 8:32
  SPEAKER_00: Okay.
• 8:33
  SPEAKER_00: Okay.
• 8:34
  SPEAKER_00: Does it ever have to be flushed out for any reasons?
• 8:37
  SPEAKER_00: Um, or is it pretty much um that's the water always used from then on out?
• 8:42
  SPEAKER_01: No, that's the water always used, unless you have some sort of failure or or leak in the system.
• 8:47
  SPEAKER_01: But the typically the water that you fill in your closed loop, that's the one that stays in the system.
• 8:54
  SPEAKER_00: So that's a huge deal then compared to you said, all, you know, the Olympic-sized pools example.
• 9:00
  SPEAKER_00: That's a huge savings in water.
• 9:02
  SPEAKER_00: Um, big picture.
• 9:03
  SPEAKER_00: How important is this to the environment?
• 9:05
  SPEAKER_00: With the need of all these data centers, how big of an environmental impact do you think that this would make?
• 9:12
  SPEAKER_00: And are you working with other companies that are also sort of on the same page of reducing the water in the data centers?
• 9:19
  SPEAKER_01: Yeah, so so majority of the water consumption comes from the cooling, right?
• 9:24
  SPEAKER_01: So data centers typically consume water in two ways.
• 9:28
  SPEAKER_01: One is direct, that's basically directly on the cooling tower that they have at the side of their data center.
• 9:34
  SPEAKER_01: Um, and the other way that they consume water is basically indirect, and that's basically the cooling tower and the power plant that generates power for the data center.
• 9:44
  SPEAKER_01: Which in most cases we help them save that water, right?
• 9:47
  SPEAKER_01: So on the cooling side, we completely eliminate the cooling tower on the data center side.
• 9:53
  SPEAKER_01: And uh we also help them save energy.
• 9:55
  SPEAKER_01: So, in a way, we also help them to save water from the power plant size as well.
• 10:01
  SPEAKER_01: And the concern is here, especially for the areas that you are dealing with the dry out, you are dealing with the regulatory environments, right?
• 10:08
  SPEAKER_01: For example, Arizona, Nevada, right?
• 10:11
  SPEAKER_01: The areas that's like extremely hot, you would need even more water for cooling.
• 10:16
  SPEAKER_01: And and our solution is kind of revolutionary because it it enables you to build data centers in those regions without any water consumption, which is quite significant.
• 10:25
  SPEAKER_01: And if you think of it in a in a kind of like a big picture, right, it would also enable you in a way to combine your data center with solar and deploy it off-the-grid.
• 10:37
  SPEAKER_01: Because right now, whenever you have solar energy, it's too hot, and the data center is not efficient, and the ratio doesn't make it financially viable to have completely off-the-grid data center.
• 10:49
  SPEAKER_01: But with our solution, uh, your data center is efficient, it doesn't need any water, so you can easily deploy kind of an off-the-grid data center in the areas that it was not possible in the past, which is like a huge advantage and consequence of the solution like FevRet that enable you to build data centers in those regions, which was not possible before.
• 11:11
  SPEAKER_00: So, are how quickly do you think that this new technology can be deployed and used throughout all the upcoming data centers?
• 11:22
  SPEAKER_01: Yeah, well, we have our solution.
• 11:24
  SPEAKER_01: We are piloting our solution.
• 11:26
  SPEAKER_01: Um, if anyone is listening and they are interested, we would be more than happy to pilot the solution with them and with some of our customers we are planning to scale.
• 11:35
  SPEAKER_01: Uh, one thing is very clear the problem is becoming bigger and bigger as we move things forward because you know people have a problem of access to power, a problem of getting permits for water consumption, right?
• 11:49
  SPEAKER_01: And and the example that I always give is that you cannot have a leaky bucket and pull more water to it.
• 11:55
  SPEAKER_01: So you cannot just keep getting more and more permits for power if you don't use your power in an efficient way.
• 12:01
  SPEAKER_01: So the problem is just getting bigger and bigger, and we are here to help.
• 12:04
  SPEAKER_01: So we're here to help any customer or any potential partners that's out there and they are building data centers today.
• 12:12
  SPEAKER_01: We are ready to work with them to bring this solution to the market.
• 12:16
  SPEAKER_00: Are you hitting any walls when you go to talk to data center builders or providers?
• 12:20
  SPEAKER_00: Are you hitting any walls when it comes to changing their current cooling methods?
• 12:26
  SPEAKER_01: Yeah, well, I mean, I think most of the data centers, right, currently use this kind of a liquid cooling solution they call direct to chip.
• 12:35
  SPEAKER_01: And that's what the majority of them use.
• 12:37
  SPEAKER_01: But even that one is still dealing with the water consumption uh at the data center level and doesn't provide the performances that our solution does provide.
• 12:47
  SPEAKER_01: So again, as as the problem becomes bigger and bigger, and and these data centers are feeling the pain more and more, I think then there's a chance for the companies like ours to you know to grow and come to the market and offer the solution that they don't have.
• 13:03
  SPEAKER_00: Absolutely.
• 13:04
  SPEAKER_00: Well, if there was one key takeaway you could leave our audience with today, what would that be?
• 13:09
  SPEAKER_01: Yeah, well, we are here to help to make the AI revolution sustainable and uh we enable you to be data centers in the areas that you were not, you know, able to build data centers in the past.
• 13:23
  SPEAKER_01: Just come and talk to us, and we are here to help.
• 13:26
  SPEAKER_00: Wonderful.
• 13:27
  SPEAKER_00: Thank you so much.
• 13:28
  SPEAKER_00: And um, where where can they find more information about this study if they wanted to look at the full study?
• 13:35
  SPEAKER_01: Absolutely.
• 13:36
  SPEAKER_01: So we have the press release and we have the white paper in our website, which is on www.febread.com.
• 13:44
  SPEAKER_00: All right.
• 13:45
  SPEAKER_00: Thank you so much.
• 13:46
  SPEAKER_00: And thank you to our audience.
• 13:48
  SPEAKER_00: If you have any questions about this, leave them in the comments and we'll try to get them answered for you.
• 13:53
  SPEAKER_00: Have a great day.