Tyan has seen a resurgence recently, especially in the GPU compute space. At ISC 2017 in Germany, Tyan is showing off its family of GPU compute servers based on the next-generation of Intel Xeon Processor Scalable Family (known as Skylake-SP or Purley.) This is significant because the company’s 4U GPU compute servers are popular in not just the HPC segment, but also in an architecturally closely related field of deep learning clusters. Intel Xeon Gold 5218
Tyan Supporting Skylake-SP / Purley is Significant in the HPC Space
Although STH is still under embargo around Skylake-SP and Purley, Intel has publicly shared a few key details around its next-generation platform. Here is a slide from Supercomputing 2016:
AVX512 will help floating point performance of the new chips in a similar fashion as to how it was bolstered in the Intel Xeon Phi Knights Landing edition. Beyond this new HPC-centric compute mode, Intel is also bringing Omni-Path to the new chips. This integration will allow companies to use Intel Xeon Processor Scalable Family CPUs to utilize a 100Gbps fabric without taking up a rear I/O PCIe slot with a Mellanox Infiniband card. That gives more room for GPUs or Intel Xeon Phi co-processor cards.
Key Tyan HPC & Coprocessor Platforms at ISC 2017
Here are a few key platforms Tyan is showing off at ISC 2017:
- FT76-B7922: 4U quad-socket Intel Xeon processor E7-8800/4800 v4-based platform with support for up to 4 Intel Xeon Phi coprocessor modules, 96 DDR4 DIMM slots, and 8 2.5” hot-swap SAS 12Gb/s or SATA 6Gb/s devices
- FT77D-B7109: 4U dual-socket Intel Xeon Processor Scalable Family-based platform with support for up to 8 Intel Xeon Phi coprocessor modules, and 14 2.5” hot-swap SATA 6Gb/s devices, 4 of the bays can support NVMe U.2 drives
- FT48B-B7100: 4U dual-socket Intel Xeon Processor Scalable Family-based platform with support for up to 4 Intel Xeon Phi coprocessor modules, and 10 2.5” hot-swap SAS 12Gb/s or SATA 6Gb/s devices
The FT77D-B7109 and FT48B-7100 both utilize the next generation Intel Xeon Processor Scalable Family and are some of the first GPU compute systems on the market based on the new Purley platform.