The GPU has been on the rise as a compute tool. They are incredibly powerful, especially for computing parallel processes. So why isn’t everyone using a ton of them? Aside from any cost factors, GPUs are large and they require a lot of power and cooling, which most systems are not equipped to handle. GPUs are typically added to a server but the amount of data that can be manipulated is dependent on the number of GPUs the computer can support. Many current servers provide 7 slots but only a few have enough bandwidth to fully support the latest GPUs. In most cases, the more GPUs available to manipulate data, the faster the data reaches the analyst. The most advanced computers hold multiple GPUs for this purpose but since GPUs are power hogs, most computers are not equipped to handle more than one or two GPU cards.

A way to solve this issue is to expand the system. Multiple GPUs can be added to any computer by expanding the PCI Express (PCIe) bus from the computer to a separate enclosure that houses multiple boards. Connecting to the computers PCIe bus directly eliminates the necessity for any software conversion back to the root complex, tremendously reducing latency and cost. Your system will think the GPUs are in your system without actually being in it. These enclosures CAN BE connected to one or more servers through (up to) PCIe Gen3 x16 cables with a theoretical bandwidth of 128Gb/s.

The One Stop Systems (OSS) High Density Compute Accelerator (HDCA) accommodates up to sixteen GPUs in a 3U chassis making it the densest solution on the market right now. We used CUDA Tools and performance tested it with 16 Titan Black GPUs (4U chassis) and got an actual bandwidth of 98 GB/s with only one PCIe 3.0 x16 uplink to a Dell PowerEdge R720. Our 3U works with multiple servers and various GPUs, including the NVIDIA Tesla K40s and K80s. The box packs quite a powerful punch and solves the problem of your system not having enough resources for multiple GPUs.

We realize 16 GPUs are a lot, which is why we also have various sized compute accelerators if you don’t need quite that many. In any case, consider GPU-based solutions for your high performance computing application needs.

To learn more about the products or download datasheets, visit the product page here.

Share this post: