Paul McLeod of Supermicro discusses collaborating with Solidigm to boost power efficiency for AI workloads in the data center. The cost of running a GPU cluster, not just acquiring it, is a key element in modern data centers, making cooling them top on mind.
One of the major power draws comes from the fans used to expel heat, but fans just can’t keep up with the amount of heat produced by the GPUs and other components. Liquid cooling is the solution to maintaining appropriate temperatures, eliminating the need for power-hungry fans.
Working with Solidigm to lower power usage, Paul calls out the Solidigm D7-PS1010 liquid-cooled E1.S SSD, an innovative technology that can help Supermicro deliver the best technology to its customers for the best value.
Supermicro and Solidigm collaborate to provide the best cooling in the data center today.
In an AI data center the very first thing someone is looking at is “What is this going to cost me to do?” We already know that the acquisition cost of a GPU cluster is enormous. But once you have that GPU cluster how much is that going to cost every day that you run it?
My name is Paul McLeod and I'm a Product Director at Supermicro. We've specialized in bringing technology to our customers sooner than anybody else and at the best value.
So, we always been a champion of startups and the little guy who's trying to get into the bigger IT business. Our key value is what we call our building block solutions and we're able to help our partners get to a solution step-by-step or at the data center level. AI workloads need the most performance in the world. From a performance-per-watt perspective, power efficiency really ends up being one of the key elements to make sure that your data center stays inside of its power envelope and its thermal envelope.
GPU's from the start always were thermally hot devices and now that's increasing exponentially. As those temperatures go up, as that power draw goes up, there's almost no way that you could push air through them to cool them down. So, it's essential that you have some other method and liquid cooling is the most efficient way for removing heat from an electronic element.
With liquid cooling you have this tool now to dial in what the temperature of that device is and set it at a perfect temperature with the flow of liquid. The advantage of that of course is not only are we able to cool the processors and components, we're able to get that at a much better price because one of the biggest power consumers in a data center are all the fans that are blowing in there.
Solidigm and Supermicro have been great partners because we both are really into delivering the best technology to our customers at the best value. An example of this would be Solidigm’s liquid-cooling solution for E1.S devices. This is new and innovative technology that nobody else has so I'm looking forward as a system builder to be able to have that inside of my system. Cold-plate technology really is a plug and play type of technology because it integrates in the same footprint as a[n] air cooled environment and that really is excellent for our customers because again they want to be able to get this technology as soon as possible.
Supermicro and Solidigm collaborate to provide the best cooling in the data center today. And I think from a performance-per-watt perspective Supermicro is really leading the world in how you deploy AI and HPC compute today.
Subscribe for product updates, technical resources, and insights on accelerating AI workloads with Solidigm Storage