May 29th, 2018 by Adam Armstrong
Supermicro Announces AI & HPC With NVIDIA At GTC
Today at the GPU Technology Conference (GTC) Taiwan 2018, Super Micro Computer, Inc. announced the adoption of NVIDIA’s HGX-2 cloud server platform for artificial intelligence (AI) and high-performance computing (HPC). Supermicro is also showcasing its new GPU-optimized platforms based off of NVIDIA Tesla V100 PCI-E and V100 SXM2 Tensor Core GPU accelerators in its Platinum Sponsor Booth. These new platforms also address the demands of AI as well as deep learning and big data analytics.
AI has the potential to rapidly transform many industries and is quickly becoming the focus of several. Another area of interest that can unlock insights is in HPC. Both of these technologies and their models are increasing in size and need more and more powerful platforms to fully leverage them. In order to give AI and HPC teams what they need, Supermicro is releasing its new HGX-2-based cloud server systems. These systems have up to 16 Tesla V100 GPUs and NVSwitch leverages over 80,000 Cuda Cores. The company states that its new HGX-2 SuperServer can hit 2 PetaFlops and refers to it as the world’s most powerful cloud server platform for AI and HPC.
Also on the GPU-based front, Supermicro is showcasing several platforms that leverage NVIDIA technology. The company has introduced the 4U, 4029GP-TVRT for maximum performance. The 4029GP-TVRT leverages the NVIDIA HGX-T1 and supports eight NVIDIA Tesla V100 32GB SXM2 GPU accelerators with maximum GPU-to-GPU bandwidth for cluster and hyper-scale applications. This new platform also incorporates the latest NVIDIA NVLink technology for bandwidth five times that of PCIe 3.0.
Another 4U system geared toward performance is the SuperServer 4029GR-TRT2. This sever supports up to ten PCIe NVIDIA Tesla V100 accelerators by leverageing the company’s GPU-optimized single-root complex PCIe design, which it claims dramatically improves GPU peer-to-peer communication performance. Supermicro is showcasing high GPU density in a smaller from with the 1U 1029GQ-TRT that supports four NVIDIA Tesla V100 PCIe GPU accelerators and the 1U 1029GQ-TVRT that supports four NVIDIA Tesla V100 SXM2 32GB GPU accelerators.
Supermicro’s GPU systems support the Tesla P4 that is ideal for scale-out servers and those that need to transcode multiple HD video streams in real-time. This also lends itself well to deep learning.