DUG Takes AI Inference to the Edge With Solidigm SSDs
Harry McHugh, CIO of DUG Technology, provides insight into how DUG is bringing high performance computing (HPC) to the edge in the harshest of environments using Solidigm SSDs. For example, DUG Nomad 10, a containerized modular data center, can place over 100 H200 GPUs in a 10-foot shipping container. And DUG Cool immersion technology means that it's very power efficient and thermally stable inside the tank, no matter how remote the environment.
But with so much compute power in such a constrained space, data storage must provide massive capacity in a small footprint. Solidigm SSDs meet that challenge with up to 122TB in a single drive, giving DUG the storage capacity it needs for AI inference and HPC at the edge.
Having the extremely high capacity and high performance of Solidigm NVMe drives, we can place a huge amount of data storage and performance in a small footprint, which is exactly what we need for AI inference at the edge.
The requirements for data processing are exploding. AI’s demand for both data ingestion and storage, but also the processing power, means we need more compute than ever. And that's no longer just true in the data center in big physical premises that you might traditionally have located, but also now to the edge.
My name's Harry McHugh, I'm CIO at DUG Technology.
So DUG Technology is a company focused on high performance computing. We traditionally focused in the oil and gas industry, but since then we've developed into high performance computing across all verticals.
DUG have been operating data centers for over ten years, all of which are based on DUG COOL’s immersion technology. What that immersion technology allows us to do is put a huge amount of compute power in a very small footprint. So instead of using air to cool computers like you might see in a movie in a traditional rack, what we do is submerge the entire computer in a bath of oil, or a tank of oil as we call it.
What that means is that you get dramatically more compute in the same footprint because it's much more efficient to cool your computers using oil or liquid than it is to use air. Industries and verticals have suffered from something known as data gravity, where it takes a huge amount of time to move large data sets from one location to another.
Traditionally, that happens over fiber networks. But even the fastest fiber networks; it can take days, weeks, or months to move the amount of data that AI and HPC applications are needing to use. And over 10 years of learning and adapting and implementing immersion technology in data centers, we've now taken that to the edge with DUG Nomad—a containerized modular data center. To give you an example, the DUG Nomad 10 product can place over 100 H200 GPUs in a 10-foot shipping container.
The shipping container allows the logistics to be simplified because the shipping container is the data center. It allows us to deploy it in very remote conditions and it's battle hardened and tested to ensure that it can withstand the harshest of climates. And then once deployed on site, the DUG Immersion Cool technology means that it's very power efficient and thermally stable inside the tank.
Whilst being incredibly mobile and rugged, we are limited on how much we can place within these tanks. Traditionally we used HDDs and a luster based storage system.
We found that had a lot of complications around failure of drives, of complexity, of managing that solution. So we transitioned a significant portion of the storage system to Solidigm NVMe-based drives. Having the extremely high capacity and high performance of Solidigm NVMe drives, we can place a huge amount of data storage and performance in a small footprint, which is exactly what we need for AI inference at the edge.
