설명

32 A100 GPUs 896 CPU cores 4,672GB RAM Each 48U rack includes: • 4 ZT / Microsoft WCS GPU servers, each with 8 NVIDIA A100 40GB SXM4 GPUs — 32 GPUs and 1,280GB aggregate GPU memory per rack • 3 utility servers • 14 AMD Rome 64-core processors — 896 physical CPU cores per rack • 4,672GB DDR4-2933 system memory • 32 NVIDIA ConnectX-6 200GbE PCIe adapters • 5.76TB aggregate NVMe storage across the utility servers • Arista DCS-7060CX-32S 100GbE switch, Ingrasys networking, rack management hardware, power distribution, and rack accessories

환경 설정

Line Equipment Qty/rack Configuration 01 GPU compute servers 4 ZT / Microsoft WCS MSF-053000 02 Utility servers 3 ZT / Microsoft WCS MSF-052999 03 Arista network switch 1 DCS-7060CX-32S; 32 x 100GbE 04 Ingrasys management switches 2 Management networking 05 Ingrasys network switch 1 S6200-50ST; 1U 06 WCS management hardware Included Rack/row managers; accessory board 07 ZT rack cabinet 1 48U 08 Geist power distribution 4 Rack power assemblies 09 Rack accessories Included Rails, cabling, blanking panels; spare switch fans and PSUs GPU SERVER CONFIGURATION 8 x A100 40GB SXM4 (HGX-8) | 320GB GPU memory | 2 x AMD Rome 64-core CPUs 16 x 64GB DDR4-2933 (1,024GB) | 8 x ConnectX-6 200GbE PCIe NICs A2010 FPGA | M.2 carriers | PCIe switch | 6 x Lite-On 3,000W / 54V PSUs UTILITY SERVER CONFIGURATION 2 x AMD Rome 64-core CPUs | 192GB DDR4-2933 | 2 x PM983 960GB NVMe A2010 FPGA | PCIe riser | Flextronics power assembly | System fans

OEM 모델 설명

NVIDIA A100 Tensor Core GPU delivers unprecedented acceleration at every scale to power the world’s highest-performing elastic data centers for AI, data analytics, and HPC. Powered by the NVIDIA Ampere Architecture, A100 is the engine of the NVIDIA data center platform. A100 provides up to 20X higher performance over the prior generation and can be partitioned into seven GPU instances to dynamically adjust to shifting demands. The A100 80GB debuts the world’s fastest memory bandwidth at over 2 terabytes per second (TB/s) to run the largest models and datasets.