WEKA Helps Applied Digital Serve Continuous Data to the Fastest GPUs on the Planet
Hi, my name’s Mike Maniscalco I am the CTO of Applied Digital. We provide digital infrastructure as well as services and technology to support people who are doing AI training applications and AI inference applications. As part of my role, I support all the technology for Applied Digital and some of the key services that we’re providing to our customers. The first is proprietary design, specially designed data center space for large scale AI training clusters.
The other services we offer, we offer GPU as compute services for individuals or small teams who are looking to run smaller jobs. One of our other offerings that I personally love and it’s really exciting for us is supercompute as-a-service. And what that means is we’re delivering large scale supercomputer clusters typically for training purposes. So a few years ago we looked at how do we diversify our business, Where is the market going, where are the customers looking for with GPUs and started investing more in GPU as-a-service and ultimately supercompute as-a-service as I think the world really woke up to what was possible with AI.
I think the best way to view it is, what are people deploying today at scale? If you’re looking at what OpenAI did with GPT-4, or you’re looking at what some of the other LLM companies are doing, you’re putting that in perspective and saying what’s going to come next. So we’re looking at some of the bigger clusters right now, 15,000 to 20,000 plus CPUs in that cluster and you’re building a building for say, 20 years of lifetime.
Where are we going to be in 20 years from now? It’s pretty clear that we’re going to go from 20,000 to 30,000 to 40,000 to 50,000 to 60,000 GPUs in a single application cluster when the teams have the ability to do that. If you’re going out looking for a data center that could accommodate 50,000 H100 GPUs today, you’d have a really, really hard time finding a facility that can accommodate that. And that’s something we’re really excited about.
When I was thinking about storage solutions and how we should evaluate them and we were talking to energy companies, oil and gas companies, pharmaceutical companies, and what you quickly realize, especially as a cloud service provider, is, you know, I might know my customer today, but when that contract is up in 18, 24, 36 months from now, I really don’t know what kind of customer we’re going to be servicing or what type of workloads they’re going to be needing and running. We’re using WEKA to basically serve the data up to these GPUs, the fastest GPUs on the market today. When we’re talking about performance and storage, we wanted a solution that would basically hang with the fastest GPUs in the world, something that would deliver the speed that those GPUs need.
They’re they’re hungry for data. They want to process data. They want to process information. And there are only a few solutions available that can do that and offer the flexibility that we need for the future.
AI infrastructure can certainly be power hungry, and that’s not a trend that’s going backwards. We see the future generations of chips just getting more and more power hungry, requiring more and more watts per GPU. That shows in our data center design. We were looking at power efficiency and building our data centers in North Dakota is one of the most power efficient ways to develop a data center without going to the Arctic Circle.
How can we leverage the outside climate and the outside air and to chill the systems that chill the water, to make these systems as efficient as possible? Looking at the GPUs and how quickly they compute and how quickly they move, you don’t want them sitting, waiting, being idle, wasting power. You want to utilize those as much as you can to make your training as efficient as possible. And that’s one of the things I think WEKA excels at, is they deliver really, really quick data to those GPUs so that they can be utilized to the max.
We’re just opening the world’s eyes to the possibilities of AI. And now that that’s happened, I think that this is going to do nothing but grow very, very quickly, which for us means more customers who need compute more opportunities to help those customers train and serve those applications to their customers. I think the future is bright and it’s great to be aligned, as a fast growing startup, it’s great to be aligned with another company that’s so focused on innovation and growth and shares the same ethos. So really happy to be working with WEKA.