• Telecom & ICT Equipment Supplier
  • Multi-Vendor Sourcing
  • Worldwide Shipping
YT CIG Telecom & ICT Equipment
Get a Quote

NVIDIA GPU Baseboard — HGX H200 8-GPU

Part Number
HGX H200 8-GPU
Manufacturer
NVIDIA
Condition
New / Used — confirmed on quotation
On request

The NVIDIA HGX H200 8-GPU baseboard pairs eight Hopper-architecture NVIDIA H200 SXM GPUs for AI training, inference and HPC. Each GPU carries 141 GB of HBM3e at 4.8 TB/s, giving 1128 GB (1.1 TB) and 38.4 TB/s per board. Fourth-generation NVLink gives 900 GB/s GPU-to-GPU and third-generation NVSwitch 7.2 TB/s aggregate bandwidth. It exposes eight Gen5 x16 PCIe links plus one Gen4 x2 link and delivers 32 petaflops of processing power, though NVIDIA states no precision. FP8 Tensor Core throughput reaches 3,958 TFLOPS per GPU with sparsity; each GPU is configurable up to 700 W TDP.

What are the specifications of this equipment?

AI Research Sourced in part from publicly available data via AI research — contact us for details.
Part Type GPU Baseboard
Part Number NVTHGX-H200-SXM5-8141 (as listed on RFQ)
Configuration 8x NVIDIA H200 SXM GPUs on a single HGX H200 baseboard
GPU Architecture NVIDIA Hopper
GPU Memory 141 GB HBM3e per GPU
Total GPU Memory per Baseboard 1128 GB (1.1 TB) HBM3e per 8-GPU board
GPU Memory Bandwidth 4.8 TB/s per GPU; 38.4 TB/s aggregate per 8-GPU node
NVLink Fourth-generation NVLink, 900 GB/s GPU-to-GPU
NVSwitch Third-generation NVSwitch, 7.2 TB/s total aggregate bandwidth
PCI Express Eight Gen5 x16 links and one Gen4 x2 link per 8-GPU baseboard
AI Performance 32 petaflops of processing power (precision not stated by NVIDIA)
FP8 Tensor Core 3,958 TFLOPS per GPU (with sparsity)
Max Thermal Design Power Up to 700 W per GPU (configurable)
Multi-Instance GPU Up to 7 MIG instances at 18 GB each
Brand NVIDIA
Condition New / Used — confirmed on quotation
Availability Confirmed on quotation

Which similar equipment can replace this unit?

More NVIDIA H200 equipment we stock

How does it compare with related parts?

Parameters as recorded for each published part; a dash means the record does not carry that field. Reach values are per part and not interchangeable.

ModelPart typeData rate WavelengthReach / distanceConnector
HGX H200 8-GPU GPU Baseboard — — — —
H200 NVL PCIe GPU Accelerator — — — —
900-23946-0000-000 NVLink Bridge Board — — — —
SYS-821GE-TNHR GPU Server — — — —

What is this category of equipment used for?

Spare parts and accessories include line cards, power modules, fan trays and small parts across 20+ manufacturers' product lines.

Frequently Asked Questions