GPUs & accelerators / HopperNVIDIA / SERVER
NVIDIA H200
SXM / HGX platform
Communication-intensive distributed AI training, large-model inference, and scientific HPC.
Eight-GPU HGX H200 baseboard shown; memory specifications are per GPU · NVIDIA141 GBHBM3e
4.8 TB/sMemory bandwidth
Up to 700 WGPU board power
HopperGPU architecture
TECHNICAL SPECIFICATIONS
Know what you’re ordering.
| Manufacturer | NVIDIA |
|---|
| Product | NVIDIA H200 |
|---|
| Edition | SXM / HGX platform |
|---|
| Architecture | Hopper |
|---|
| Memory | 141 GB HBM3e |
|---|
| Memory bandwidth | 4.8 TB/s |
|---|
| Board power | Up to 700 W |
|---|
| Cooling | Platform-integrated air or liquid |
|---|
| Host interface | SXM module / HGX baseboard |
|---|
| Form factor | Platform component · not a PCIe card |
|---|
Reviewed September 11, 2026. Peak bandwidth and board power are manufacturer specifications, not measured NexTec application results.
Platform & integration notes.
01SXM modules require a compatible HGX server. Specify the full supported baseboard and system configuration.
02Eight H200 SXM GPUs provide 1,128 GB aggregate memory and an NVLink/NVSwitch fabric in supported HGX designs.
03Order the complete system with host CPUs, memory, storage, scale-out networking, cooling, and support coverage.
LET’S BUILD WHAT’S NEXT
Your workload. Our starting point.
Tell us what you’re building. We’ll help scope the infrastructure.
