SERVCHIP
Interactive view
ARCHITECTURE
Hopper
SERIES
H800
Memory
80GB HBM3
Bandwidth
3.35 TB/s
TDP
700W
In StockH800·Best for: AI Acceleration in Regulated Markets
NVIDIA H800 GPU
Hopper
|Data Center GPUsOverview
The NVIDIA H800 GPU is a variant of the H100 Hopper architecture designed for markets with specific export control requirements. Featuring the same 80GB HBM3 memory and Tensor Core compute as the H100, the H800 has reduced NVLink interconnect bandwidth while maintaining full compute performance. It provides a viable option for AI training and inference workloads in regions where the standard H100 configuration is not available.
Key Features
- 80GB HBM3 memory with 3.35 TB/s bandwidth
- Hopper architecture with Transformer Engine
- FP8 Tensor Cores for accelerated AI
- Reduced NVLink interconnect for regulated markets
- Multi-Instance GPU (MIG) support
Ideal Use Cases
ai trainingai inferencehpc
Full Specifications
Complete technical specifications for the H800.
Memory
- Memory
- 80GB HBM3
- Memory Bandwidth
- 3.35 TB/s
Compute
- Tensor Cores
- 528
- CUDA Cores
- 18432
- FP8 TFLOPS
- 1979
- FP16 TFLOPS
- 989
- TF32 TFLOPS
- 494.5
- FP64 TFLOPS
- 60
Interconnect
- Interconnect
- NVLink 4.0, 400GB/s
Physical
- TDP
- 700W
- Form Factor
- SXM5
- Cooling
- Liquid Cooled
Manufacturing
- Launch Date
- Q3 2023
- Manufacturing Process
- 4nm
