SC20 is on again this week, as a fully virtual, online event. The event remains a key gathering of the supercomputing and HPC industry, and this year is no different. There were a range of announcements from the key vendors in the field relevant to XENON customers – including NVIDIA, Intel, AMD, DDN and Lenovo.
Contact the XENON team for more information on these new products and to discuss how you can leverage these developments in your infrastructure.Talk to a Solutions Architect
NVIDIA Updates A100 GPU and DGX Line
NVIDIA has used SC20 to announce new products and enhancements to the A100 Tensor Core GPU line which just arrived in May, as well as the new NVIDIA Mellanox 400 Gbps InfiniBand.
- A100 Tensor Core GPU memory increase. The A100 Tensor Core GPU is now available with 80GB memory, in NVLink form and as an option in the DGX systems. Read the new A100 Tensor Core GPU datasheet. Please note that the increased memory is not yet available on the A100 PCIe GPUs.
- DGX Station A100. A complete redesign of the popular workstation form factor solution, the new DGX Station A100 uses the new HGX motherboard and 4 x A100 Tensor Core GPUs to deliver a powerful AI appliance for workgroups that fits under a desk and doesn’t need data centre power and cooling. The 4 x A100 GPUs with either 40Gb or 80GB memory each delivers over 2.5 petaFLOPS of performance. Mutli-Instance GPU (MIG) enables the 4 x A100’s to be divided and combined into up to 28 separate GPU instances to provide a flexible agile workstation for use across data analytics, training and inference. See full details of the new DGX Station A100.
- DGX A100. NVIDIA announced an expanded the DGX A100 product line now available with two memory configurations leveraging the new 80GB A100s. The “original” DGX A100 with 320GB memory across the 8 x A100 GPU’s and another with 640GB memory utilising the 80GB A100 Tensor Core GPUs. Customers with the 320GB model have an option to upgrade to the newer higher memory A100 Tensor Core GPUs – contact us for upgrade options. Read the full updated DGX A100 specifications or download the new DGX A100 datasheet.
- NDR 400Gbps NVIDIA Mellanox InfiniBand. This new InfiniBand solution family from NVIDIA Mellanox provides the fastest networking performance available for researchers and scientists. This fully in-network acceleration platform is integrated from NIC to switch. The new line up includes adapters, data processing units (DPU’s or smart NICs), switches and cables. These new 400Gbps units are expected to ship in Q2 of 2021 delivering lower power use and better TCO, these are expected to be the standard for exascale infrastructure.
A great feature of the virtual SC20 is that the material is available online – see the full NVIDIA program at the NVIDIA Website.
Intel Reveals 3rd Generation Xeon Processors – Ice Lake
At SC20 Intel revealed the third generation Xeon Scalable Processors – Ice Lake-SP. The Ice Lake-SP is squarely aimed at intensive HPC and AI workloads, and the range of improvements across the platform are designed to deliver improvements in power and speed in these demanding environments. The Ice Lake-SP range is expected to ship in Q1, 2021.
Ice Lake-SP is Intel’s first server processor with 10nm technology, and combined with the Whitley 2-socket server platform the Ice Lake-SP delivers an increase of DDR4 memory channels from 4 to 6 and introduces PCIe Gen4. With Optane persistent memory, the platform supports up to 6TB per channel. The Ice Lake-SP platform is performing 18% higher instructions per clock verses the previous generation Cascade Lake, which results in overall boost to performance with the same core count.
AMD’s new MI100 GPU Accelerator
AMD was also active at SC20, releasing the new AMD Instinct MI100 accelerator. AMD claims this is the fastest accelerator for scientific workloads and the first GPU to exceed 10 teraFLOPS (FP64) performance. The MI100 is built on a new CDNA architecture and supported by ROCm 4.0 open source ecosystem software.
AMD also announced they are on target for volume shipments of the EPYC Zen3 core processors to select HPC and cloud customers this quarter, and wider public availability of these CPUs in Q1, 2021.
DDN Enhances Products, HPC focus
DDN announced a range of product enhancements at SC20 across their range, including EXAScaler 5, DataFlow and IntelliFlash. DDN has called out the features relevant to enterprise, cloud and AI implementations, helping to reinforce DDN’s position as storage of choice for high performance storage.
EXAScaler 5.2 adds enterprise data management capabilities for AI and analytics customers, and has the potential to make Lustre more customer friendly and easier to deploy in enterprise as well as HPC environments. DataFlow delivers increased efficiency for large volume data management and cloud mobility, with backup and archive capabilities. Finally, IntelliFlash is the high performance hybrid SSD/HDD storage solution from Tintri that is now available under the DDN At Scale offering.
The DDN positioning in market is sharper, with DDN At Scale brand focused on the requirements of large HPC customers, while their Tintri brand goes to market focused on enterprise customers after hybrid performance for structured data sets.
Lenovo’s new ThinkSystems
Lenovo announced two new ThinkSystem servers, which both use Lenovo’s Neptune liquid to air cooling technology.
The new ThinkSystem SD650-N V2 is a compact 1RU server, with 4 x NVIDIA A100 Tensor Core GPUs connected with NVLink. Lenovo claims peak performance metrics for the system at 2.8 petaFLOPS (HPC) and 45 petaFLOPS (AI). Read the full details on the Lenovo website here.
The new ThinkSystem SR670 V2 is a 3RU server, with a modular design that can accommodate up to 8 double width GPUs with NVLink, or 8 single GPUs with 8 drives. The ThinkSystem SR670 V2 is designed to support the new NVIDIA Ampere Data Centre GPUs – the A100 and the A10. Read the full specs and details on the ThinkSystem SR670 V2 at the Lenovo website here.
More News From SC20
This post has highlighted news from XENON’s key vendors. Contact XENON today to set-up a briefing on up-dates relevant to your organisation.Talk to a Solutions Architect