In this video from ISC 2018, Perry Hayes and Martin Galle from Supermicro describe the company’s latest innovations for HPC and AI workloads.
Supermicro delivers the industry’s fastest, most powerful selection of HPC solutions offering even higher density compute clusters to deliver maximum parallel computing performance for any science and engineering, simulation, modeling, or analytics applications,” said Charles Liang, president and CEO of Supermicro. “The new STAC-N1 world record shows our Hyper-Speed Ultra Servers improved the latency performance by 23% for financial applications to deliver the performance, scalability and reliability needed to quickly solve their computing challenges.”
As medium enterprises adopt critical applications to meet business needs, they turn to HPC solutions capable of managing compute-intensive workloads. Supermicro’s SuperBlade and FatTwin platforms enable them to get the most out of their computing systems. Supermicro’s Resource Saving Architecture found in these platforms provides added value with space-saving, low-power technology to refresh their systems over time while delivering maximum computing performance.
Supermicro’s HPC optimized systems feature NVIDIA Tesla GPU and Intel Xeon Scalable processor technology for a broad range of configurable solutions that meet compute requirements in applications for financial, deep learning, cluster, or hyper-scale workloads.
Supermicro’s exhibit will include the following product demonstrations featuring NVIDIA based products:
- New HGX-2 based system. Supermicro’s new Cloud servers based on the HGX-2 platform will deliver the highest compute performance and memory for rapid model training. This new HGX-2 based system combines 16 Tesla V100 32GB SXM3 GPUs connected via NVLink and NVSwitch to work as a unified 2 PetaFlop accelerator with half a terabyte of aggregate memory to deliver unmatched performance for deep learning and compute workloads.
- SuperServer 4029GP-TVRT. Supporting maximum GPU-to-GPU bandwidth for cluster and hyper-scale applications, this system, a part of the NVIDIA HGX-T1 class of GPU-Accelerated Server Platforms, features independent GPU and CPU thermal zones to ensure uncompromised performance and stability under the most demanding workloads.
- SuperServer 1029GQ-TVRT. A part of the NVIDIA SCX-E2 class of GPU-Accelerated Server Platforms, this system will be running a live demonstration for a facial detection application at the show.
Supermicro also demonstrated the following Intel based products:
- Hyperspeed ultra system. Supermicro, Red Hat and Solarflare announced on June 20, 2018 a new world record had been set for lowest latency on the well-known STAC-N1 benchmark from STAC – the Securities Technology Analysis Center – for the world’s leading financial institutions. This benchmark was performed on a pair of Supermicro SYS-1029UX-LL1-S16 servers, each with dual 8-core Intel Xeon Scalable 6144 (Gold) processors overclocked at 4.18GHz. The servers were also loaded with Red Hat’s Red Hat Enterprise Linux 7.5 operating system and Solarflare X2522 Adapters. Compared to all prior publicly released STAC-N1 results, the bare metal system demonstrated the lowest mean latency of 2.3 microseconds at both the base rate (100k messages per second) and the highest rate tested (1 million mps).
- High-density All-Flash NVMe 1U storage servers and JBOFs. For high-performance storage, Supermicro’s new all-flash NVMe 1U JBOF (Just a Bunch Of Flash) and 1U SuperServer with support for 32 hot-swap U.2 or NVMe Intel SSD “ruler” more than triples the all-flash storage density of previous 1U solutions and will provide Petabyte scale storage in a single 1U system in the near future. The new 1U all-NVMe Storage Servers and JBOF disaggregate storage into shared pools that are rapidly becoming the preferred hardware infrastructure for demanding Big Data analytics applications such as autonomous driving and real-time financial fraud detection.
- FatTwin. The Supermicro FatTwin represents a revolution in Green Computing and is highly efficient by design; this system supports customers’ critical applications while reducing datacenter TCO in order to help preserve the environment and extends the compute and storage capabilities to achieve increased performance and power efficiency. Due to its shared components, the FatTwin improves cost-effectiveness and reliability, while its modular architecture makes it flexible to configure and easy to maintain.
- SuperBlade. Supermicro is also showcasing the 8U X11 SuperBlade, a very high density and high-performance solution for HPC applications. This scalable, modular solution supports up to 20 Intel Xeon Scalable processor-based DP blade servers, ten Intel Xeon Scalable Processor based 4-socket blade servers, or 20 Intel XeonPhi (codename Knights Mill) based blade servers.