If you are at an AI startup in 2017, somewhere around the first or second round of funding you are probably building your own GPU compute cluster. We are seeing a boom of systems in the Silicon Valley data centers. One company we are starting to see more of in these GPU cluster build-outs is Tyan. Tyan has spent energy focusing on the HPC space and is showing off the fruits of those efforts at GTC 2017 with dual Intel Xeon E5-2600 V3/ V4 multi-GPU servers as well as quad Intel Xeon E7 V3/ V4 multi-GPU servers for those who need more RAM and CPU compute.
From the Tyan Press Release
Here is an excerpt about the systems from the Tyan press release:
TYAN’s FT76-B7922 is the industry’s only Quad Socket Xeon E7 4800/8800 v4 server platform with both 96 DIMM slots and support for 4x NVIDIA Tesla P100/P40 GPU accelerators. This high-performance platform is designed for both scale-up (fat node) and scale-out (many-core computing). With 1:1 CPU to GPU ratio, the FT76-B7922 provides a high price/performance and performance/watt for HPC community that both needs CPU and GPU intensive computing workloads.
The FT77C-B7079 supports up to 8x NVIDIA Tesla P100/P40/P4 GPU accelerators with 2x Intel® Xeon® E5-2600 v4 processors and is designed for large-scale GPU cluster deployments. With a 2:1 oversubscription of each x16 root port, the FT77C-B7079 architecture lends itself to high amounts of CPU to GPU bandwidth across pairs of GPU accelerators and is ideal for embarrassingly parallel workloads such as BLAST searches, brute force cryptography, parallel rendering, large scale facial recognition, and numerical weather prediction workloads.