All About Air Cooling: Overcoming Net-Zero Challenges
Jun 28, 2022 3:00 PM - 4:00 PM EDT
The Sustainable Enterprise Technology Council (SETC) is on a mission to create a more sustainable environment as it relates to enterprise technology. Today, they’ve brought in experts to facilitate a discussion on air cooling efficiencies and achieving net-zero emissions.
Although a great deal of innovation still needs to happen in this space, experts are calling for a shift in air cooling practices. Data centers need more efficient ways of cooling, especially as we adopt more power-intensive CPUs and GPUs. Experimenting with liquid cooling, experts are seeing hope for more sustainable, low-cost cooling methods.
In this virtual event, Greg Irwin is joined by Brendan Walsh, Principal of ESG at World Wide Technology, and Don Molaro, Chief Technical Advisor at World Wide Technology, to talk about air cooling innovation and best practices. Together, they discuss the different methods of cooling, why liquid cooling may be the most beneficial option, and the challenges of keeping costs and energy low in the enterprise technology space.
World Wide Technology (WWT), a global technology solution provider that designs, builds, demonstrates and deploys innovative technology products, integrated architectural solutions and transformational digital experiences for large public and private organizations around the globe.
Connect with World Wide TechnologySustainable Enterprise Technology Council is a community of professionals whose mission it is to increase the understanding, promote and share practical steps towards achieving more energy efficient enterprise technology.
Connect with Sustainable Enterprise Technology CouncilPrincipal - ESG at World Wide Technology
Brendan Walsh is the Principal of ESG (environmental, social, and governance) at World Wide Technology (WWT), a global technology services provider. In 2020, Brendan received a Sustainability and Climate Risk (SCR) Certification from the GARP SCR program. He’s the Founder of ESG Risk Guard and previously worked as the Executive Vice President of Global Corporate Payments at American Express. Brendan holds a bachelor’s degree in computer science from the University of Glasgow and a master’s in sustainability from Harvard.
Co-Founder, Co-CEO at BWG Strategy LLC
BWG Strategy is a research platform that provides market intelligence through Event Services, Business Development initiatives, and Market Research services. BWG hosts over 1,800 interactive executive strategy sessions (conference calls and in-person forums) annually that allow senior industry professionals across all sectors to debate fundamental business topics with peers, build brand awareness, gather market intelligence, network with customers/suppliers/partners, and pursue business development opportunities.
Chief Technology Advisor at World Wide Technology
Don Molaro is the Chief Technical Advisor at World Wide Technology, where his focus is to make data centers and related technologies carbon neutral. Don has deep technical expertise in many areas including cloud computing, embedded systems, virtualization and containers, network programming, and digital security. Don received both his bachelor’s and master’s degrees in computer science from the University of Calgary.
Principal - ESG at World Wide Technology
Brendan Walsh is the Principal of ESG (environmental, social, and governance) at World Wide Technology (WWT), a global technology services provider. In 2020, Brendan received a Sustainability and Climate Risk (SCR) Certification from the GARP SCR program. He’s the Founder of ESG Risk Guard and previously worked as the Executive Vice President of Global Corporate Payments at American Express. Brendan holds a bachelor’s degree in computer science from the University of Glasgow and a master’s in sustainability from Harvard.
Co-Founder, Co-CEO at BWG Strategy LLC
BWG Strategy is a research platform that provides market intelligence through Event Services, Business Development initiatives, and Market Research services. BWG hosts over 1,800 interactive executive strategy sessions (conference calls and in-person forums) annually that allow senior industry professionals across all sectors to debate fundamental business topics with peers, build brand awareness, gather market intelligence, network with customers/suppliers/partners, and pursue business development opportunities.
Chief Technology Advisor at World Wide Technology
Don Molaro is the Chief Technical Advisor at World Wide Technology, where his focus is to make data centers and related technologies carbon neutral. Don has deep technical expertise in many areas including cloud computing, embedded systems, virtualization and containers, network programming, and digital security. Don received both his bachelor’s and master’s degrees in computer science from the University of Calgary.
Greg Irwin 0:18
Good afternoon. Nice to see everybody. Again, I'm Greg at BWG. You know, we are partnered here with WWT is kind of the founding members of what we're calling the SETC, the sustainable enterprise technology council. And the mission is, is very clear to drive practical steps towards achieving a better sustainable environment emanating from or as a result of enterprise technology. And we know that there are many high level initiatives across various companies to improve the carbon footprint to improve the sustainability. And this group is a collection I'd say, some of which borrowed from BWG's community and growing above and beyond with an effort to spread best practices to spread and understanding and to build a network and community around these ideas. So today, we've highlighted topics around cooling, cooling systems, particularly for data centers, and Brendan going to ask you to be our, our emcee for this. And we will get everybody actively involved with questions and comments of your own. And, again, continue to roll forward here in terms of driving, driving good ideas. The council itself, right now, it's pretty simple. It's it's meetings and research. The initiative certainly has much broader and greater aspirations, and can go in many different directions. And you know what, we're going to fill them out as we go, we're still we're still very young in the journal very young in the journey. And we know, we still have big, big, wide eyes about what the possibilities are, but we're taking it one step at a time. So we do have a, you know, our council of advisors, and we ask everybody, if you're not on it, you are now and your involvement can basically be as involved deeply involved as you wish, or as lightly involved as you wish, you want to be a passive listener, that's fine. You want to be deeply involved and help drive the content and the agenda. Wonderful. And you want to help govern even better. We've got we've got rolls across the board.
Brendan, let me turn to you for a moment here. And you and I are going to start off our session on on cooling. But before we do that, tell us a little bit about give your intro and a little bit on WWT’s perspective on the SETC and what it means for your organization.
Brendan Walsh 3:18
Your thoughts? Thank you, Greg. And thank you, everyone, everyone. Good afternoon. Thanks for joining us, we are to cut to the chase on your last question. We are very, very excited about this, etc. Why? Because we believe what we can do here is bring together you know, people with the common interest, a common goal, and a big goal and that and that goal being, you know, identifying ways to reduce the carbon footprint of data centers, and the numerous different methods we can employ to do that. It's a very, very significant goal. We hear it every day from all of our clients. Certainly the larger clients have all issued Net Zero targets. And clearly the data center being one of the significant contributor, contributors to those emissions, they are being tasked with quite significant reductions. So we see that sitting as we do as a class significant systems integrator, and being able to cross between the physical and digital worlds. And we see ourselves being well placed in that regard. So we're hugely excited about this, etc. And everything we can do. But I think you touched on this, Greg, this is only going to be as successful as the participation the contribution the idea generation that we get from the group sort of benefiting so So today we're gonna talk about cooling.
Greg Irwin 5:04
And I know that these are initiatives that you and your team have put some time into, as have others around the group. So this is I'd like to basically cover both. One, what you've been speaking to your clients about where the challenges are, what the design parameters, what some of the solutions that you've been proposing, and driving the conversation here around it. So I'm good at asking questions, but I'll encourage in this forum, lots of people to ask questions. We have our chat window. Let's all use it throughout, please layer in your comments and questions. And let's again, make this a productive system or a productive session. Brendan? So let's let's get right at it, if you would, if we can. Tell us a little bit about the types of consulting and the types of ideas that you've been grappling with questions you've been grappling with around more efficient cooling for, for data centers.
Brendan Walsh 6:09
Sure. Just before we get to that, it just let me remind everyone of the WWT ESG. offerings such as similar to or done with the SETC, we have just been a recently launched an ESG consulting practice. And that practice was launched following quite a considerable amount of research and strategy development, looking at customer needs, what was out there, looking at our own assets. And given some of the unique differentiators that we have such as the etc, the Advanced Technology Center the lab, as it's knowing the known as some of the different practice areas, we have come up with three distinct areas of offering. One I'm very pertinent to today's discussion is unsustainable technology. And that really looks at a whole raft of different sustainable technology solutions, one of which is cooling, because I believe in those technicians out there, especially Don, the latter off, WWT team will correct me if I'm wrong. But I believe that some 40% of the energy in a data center is spent on cooling. So other beyond like CPU or GPU and needs. Cooling clearly is a significant target, and a significant opportunity. That's why we're talking about it today. But also beyond the sustainable technology. We're also another pillar we're offering is ESG consulting, which is everything from starting how to measure your footprint through to tcfd reporting and in the more advanced reporting, that goes along with typically what we see in clients that are more mature in their ESG journey. And then the last pillar is ESG data and AI. And that fits very nicely with one of the practice areas that WWF Qi has, which is obviously aim. But getting back to the scenario seven table, let's bring us back home to the cooling. The question of cooling, we talked about the 40% What would like to do is maybe have done done the lateral set out and set the scene for us and take us into some of the specific questions that we have been getting. Donor. Sure. Sure. Thank you, Brendan.
Don Molaro 9:02
As you as Brendan pointed out, you know roughly roughly in a typical modern data centre, you know, pu ease, which is the power use efficiency calculation comes out somewhere in the range of 1.4 which means for every watt you for you're putting 1.4 Watts into the data centre for every useful watt of work you're getting out of the computers and the majority and the vast majority of that is going into the cooling that goes into in into that data centre. There's roughly sort of three or one could argue four classes of cooling. One is one that I'm sure everybody's seen which is you take your computers you take all the fans out of it and you put it into a tank of what is essentially baby oil. least it was originally it's it's other it's other chemicals now but but it's it's it's full immersion computing you do it with a with a nonconductive would. And it works. It has all sorts of drawbacks, but we can talk about the one that we're particularly interested in. And because we think not only does it provide a lot of the same, the same efficiency, it's it's not as efficient. But it's, it's, it's getting better. But it also is much more deployable and much more maintainable. Because we're talking about enterprise data centres is what's called direct on chip liquid cooling, which is where they were where the you replace the heat sinks on various components of the system, typically the CPUs, and shortly it'll be the GPUs and the other sort of exhilarate compute units that produce a lot of the heat and use a use of a fluid. Sometimes water, sometimes not to draw the heat directly from those components, and radiate it out to other parts of the data center.
Don Molaro 10:56
We've seen some success with that. In the lab, we've seen some small deployments commercially. But the the sort of train coming down the track on this is that enterprise CPUs are currently dissipating around around 200 Watts 205 watts for the for the sort of top end Intel's and slightly higher than that for top end AMDs. Next year, we're going to see, you know, Sapphire rapids come in at 350 Watts, we're going to see 1000 Watt GPUs. And we're going to see these things used in very high density in data centers. And, frankly, air cooling, just simply blowing enough air is going to be a, it's going to be hard to blow that amount of air efficiently or at all right to cool those components. So we're a bit of a bit of a double whammy in the fact that air cooling sort of is an inefficient way to do it electrically. Because it just you have to convert electricity into air motion. But you also but you also running into, you know, the realities of it's simply going to be hard to cool. Those devices, there's still a lot of work to be done around this, they're still I'm still, we're still looking at some very innovative companies and companies doing what I would call hybrid immersion cooling systems, where it's, it's not a full tank, it's a tank, you know, it's suitable only for, for a single system, we're seeing some stuff, very innovative and interesting stuff coming out of the open 19 group around how they're going to take on liquid cooling and sort of high density computing. There's still a great deal of innovation in that space, I'd be really interested to find out sort of, from the people on this call sort of any new and innovative technologies, they're, they're they're seeing around that, you know, as Randy just pointed out, right, anything around heat reuse, part of the problem with data centre heat is for reuse is it's really not that hot, right? It's only mean in, you know, we're only running up around 50 degrees Celsius. And it's really hard to use that for anything other than building reheating or beautiful building heating. So I'm really interested to find out if there's people with new and innovative ideas around sort of what to do with with either in the cooling space or what to do with the heat once you've once we've
Brendan Walsh 13:34
captured it. I can you tell
Greg Irwin 13:38
us a little bit You said you've seen some limited production? Um, what what can you describe a production deployment of immersion cooling May the single system may be, you know, unrack that,
Don Molaro 13:54
I mean, so, so the places we're seeing immersion cooling go into production are very dense, very hot environments. And in my world, I spent a lot of time and global finance. That's trading houses. And that's the, you know, where they're looking at. You know, they're looking at, you know, as hot a system as they can possibly run. And the other thing is, they're only looking at, you know, an 18 month depreciate, you know, at, you know, effective lifespan for these systems. So they're not really dealing with the kinds of challenges that we see in data centres. We're seeing data centre operators, sort of look at these systems. And the, you know, well, from an engineering point of view, they might make good sense. They're very effective at what they do. But operationally, you know, the idea that you're going to somehow pull a computer out of what is essentially you know, a very large bathtub, let it drain out, and so that you can swap a disk drive in it. That's a really tough thing. For those people to wrap their heads around, from an operational point of view, and I can, I can see a reluctance there for that. It's not water, right? It's a it's a, it's a, it's a, it's essentially an oil, right. So it's a, it's a manufactured chemical. And the one I saw, and I'm not dissing on this one company, but the one I saw that we really took a look at, actually, it's one that we're going to bring into our lab is from a company called Summer. And it's quite popular in Europe, Orange is deploying them in, in Europe as as edge computing for for cell towers. But you know, you have a computer and the dielectric in the fluid. And if you want to change something, well, you have to pull it, you have to get a crane, because the computer weighs 100 pounds, right, you have to get a crane to pull it out, you have to let it drip for about 10 minutes vertically, and then you put it horizontally and let it drip for another 10 minutes. And then you dip it in a solution that dissolves the last bit of oil, and that and then that flashes off, which also takes about 10 minutes. And then you can do your work on the system. And just it works. And from an operational point of view, it's a nightmare. But for something like an edge deployment, like what Iran is doing, at the base of a cell phone tower, it makes perfect sense. Because you know, those computers once they go there, they're essentially unserviceable. And, you know, and if there's anything that needs to be changed, they're replaced the whole unit. Right, and they take it back and they you know, you know, they they do it, you know, they have a different sort of operational model. But for a data center, it's going to be a hard thing to do on a day to day basis.
Brendan Walsh 16:41
I guess the
Greg Irwin 16:43
at first, the idea of liquid cooling for me, seemed arduous and complex. And then I started thinking about air conditioning. And I kind of got, because basically, that's what we're using to cool most of the western world. And it seems a little less, a little less crazy. How do you feel about the uptake here, particularly as you have, you know, more power intensive CPUs and GPUs coming? Is it just a matter of time here? On? Yeah.
Don Molaro 17:17
Yeah, I mean, I mean, we're fighting physics, we're fighting physics on two levels, right? One is, CPUs are going to get hotter. Right? There is a startup that we're looking at CPU startup, but again, sort of really leaning into things, you know, their initial CPU that they're going to tackle this company called tak em, that, you know, their initial offering in 2023, or 2024, is going to come out at 950 watts. And there's that. The other side of it is, is, as I mentioned before, fans are terrible at at cooling devices, and especially small fans, right? The reason that turbine that wind turbines get bigger and bigger and bigger to get more efficient, is that big fans or big turbines are more efficient than small ones. And so as you get, you know, those little tiny one you fans that are in servers, as you try and spin them at 30 and 40,000 RPM, which is how fast they're run, they get really inefficient at moving air. And so we're sort of fighting physics on two levels. The CPUs are just by their nature, because the as the traces get smaller, you know, resistance goes up as a as a power factor of the size of the size of the wire. Right. And fans and fans get inefficient at a you know, you know, by cubic function, right, you're sort of between a rock and a hard place from from a physics point of view on this. So I personally think it's an element, I think it's inevitable. What is the
Greg Irwin 18:57
it's inevitable, does that include your thinking in terms of the the impact, as you said, the operational impact of now you need a hardware system design that can be repaired, operate, you know, plug and play all the interfaces, you know, resilient with all the same uptime and swappable drives? Yeah, we've come to expect. How about your thinking on that side of the equation? And how much of that can effectively been be addressed with these hybrid or fully immersive systems? Versus so?
Don Molaro 19:33
So, I mean, this is this is where I think one of the biggest challenges is, right, you know, you know, prototypes are easy, you know, doing stuff in the lab is easy, getting into production and making it work every single day. You know, that's a much harder thing to do. Certainly, most of the customers I deal with, they're already I mean, they don't consume individual computers, right, they write they consume, you know that there's certain Unit of consumption is a rack. And so that's I mean, that's the business that WWT is in, that's a business that a bunch of companies are in, I see very much, you know, enterprise computing, you know, where the unit of consumption, if it's not already, you know, it'll be a rack at a time. And the racks will come integrated, you know, with networking with power and with cooling, whether that's going into a full immersive environment, like you would see, I would think in an edge computing environment, or prep, you know, with a with a direct on chip or a DLC a direct liquid cooling solution, like we see what we're doing, you know, in our labs. I think that's, I think there's gonna I mean, for those enterprises that are buying the onesie, twosie computers, that's a much harder operational thing to deploy than rolling a rack out onto the floor, and, you know, wiring in the power wiring in the network, and Mayor, you know, and maybe, or maybe not wiring in the cold water loop, you know, to provide cooling. We're in the process right now of running and standing up a large scale direct on chip liquid cooling experiment for one of our banking customers, they're going to, we're going to stand up a whole rack fully populated, it's going to come out probably over 30 to 30 kilowatts in in the rack. And a really big question is, how do we measure this? How do we measure the power that goes in? How do you know, where do we put temperature probes? What's the quality of the temperature probes? What's, you know, what, you know, how, you know, how, you know, how is it how is the airflow measured, I've spent a bunch of time trying to find some standards, and I saw you posted some stuff, and I will go review those. But there's really, I mean, from a measurement and evaluation point of view, and this is something for the group to think about is, there's really not a lot sort of written down about how, you know, to make sure that, you know, the the test results you get out of one group are, are translatable to test results, you get out of another group, there's, I mean, there needs to be needs to be some standardization around that, at least a de facto standard around it.
Greg Irwin 22:29
I'd like to pull this up a level, because I feel like, while I'm while I understand the details are important. Fundamentally, what we want to do is reduce coming back to the 40% reduce the amount of power that we need to run the dataset. And I think we get it, you need to run faster to run sufficiently fast chips, and keep them cooled at a lower cost, basically, and less with less energy. And I guess my high level question here, Don, is, if that was our goal, and we were to liquid cool. Do you believe that it's fairly straightforward that you should be able to see an energy reduction of how much power you need to put in to run the same amount of MIPS, the same amount of processing?
Brendan Walsh 23:26
I believe it is,
Don Molaro 23:28
I believe, I mean, I mean, measuring electricity, and measuring the is in you know, you know, I mean, you I mean everybody gets a power bill. You know, I mean, it's, I mean this, I mean, it's a it's a relatively relatively simple thing for at a gross level right now at the building level. Yeah, right. Right. It's a it's a relatively straightforward thing, you look at your power bill, and, and, you know, and we're, we're in, you know, we're looking at companies that are getting, you know, you know, minute by minute to let you know, power telemetry from from their power companies. So, I mean, they can really look at it in some level of detail, and that goes into a lot of the analytics that we're thinking about in the space. But there's also, you know, this is, this is also going to be, you know, a game of inches, right, where, you know, a, you know, if you have a if you can, if you can squeeze an extra 345 6% out of your rack, but if you know it, but if you're deploying, you know, 10,000 racks in your data center, you know, that adds up. If you're doing you know, if you're doing workload placement that gets you, you know, another two or 3% On top of that, well, that adds up. I'm firmly of the opinion like I say, cooling is, this is the first initial and largest sort of rock in this optimization problem to move. But there There are hundreds of there appears to be hundreds of others of them. The, again, for me, it's not it's also going to come down to, you know, you know, absolutely we can see Pee Wee's in full immersion systems, the you know, getting very close to one where, you know, you know, the, the, because water is a great, you know, you do heat exchange to water and it's a great medium to do that to, you can get very close to very efficient systems. It's a question of whether that is really practical at all at scale. So to me, there has to be sort of the ying and the yang balance there.
Greg Irwin 25:43
Brendan any any guideposts here for.
Brendan Walsh 25:47
So first of all, I think this is a fabulous discussion, some real terrific nuggets there. It comes with the Astra question is a fair statement to say. And probably an overgeneralization that for the smaller and midsize enterprises. Priority One is optimize your workload. Forget about immersion, immersion really should be ring fenced to the the sizable enterprise type customers
are looking at being conscious of everyone's time, thanks for the fabulous conversation. Really very, very useful nuggets, in terms of you know how to go about this in that pragmatic, commercially minded perspective. And I think Alan, you know, that last point has given us all food for thought in terms of one possible approach. I think that I'd love to explore that further with here, Alan, and may will take the opportunity to do that quite soon, I can see the pros and cons and run big mark mid market businesses. And I know what's some of the, the opportunities and some of the concerns as well. But we should definitely pursue that.
Don Molaro 27:11
So this has been really enlightening. A lot of it comes down to, you know, you know, looking at optimization, looking at convincing organizations to take on these things, you know that from a solid engineering point of view, a lot of that crosses a lot of internal domains that have internal silos that need to be broken down, and that, that is going to be a hard thing to do. From, from an organizational point of view, there's a lot of, you know, these are my toys, and you can play with my toys, kinds of kinds of things that happen in these organizations. So I mean, those are something we as an industry sort of need to take to heart. You know, we need to, you know, and we need to, we need to make sure that the people running the applications understand, you know, what, you know, how much power that infrastructure is costing them, right. And so there's, there's lots of work to be done there. But there's some great insight, certainly, around cooling around sort of how to approach customers, how to how to talk about different ends of the market that come out of this, it's been a great conversation, thank you, everybody
Greg Irwin 28:22
want to build this community. I love this group. So I always look forward to our sessions. But I want to double our group next time. And I think we can do it. So continue to, if you have a colleague, a friend do you think would be, you know, additive to the community, either just as a participant or contributor, all good. We're looking to build it. With that, we're going to wrap it up, you'll all be in touch. We'll be in touch with our next sessions. Next topics, and we're taking suggestions. So we can do that offline. We have a planning, planning meeting coming up here. Not not content, but more governance and operational and welcome anyone here who wants to take on more of those kinds of roles? All right, let's do it. Thank you guys. Thank you guys. Thanks, everyone. Bye bye. Take care guys.