SERVCHIP
Interactive view
ARCHITECTURE
Ampere
SERIES
A800
Memory
80GB HBM2e
Bandwidth
2 TB/s
TDP
400W
In StockA800·Best for: AI Acceleration for Export-Controlled Markets
NVIDIA A800 GPU
Ampere
|Data Center GPUsOverview
The NVIDIA A800 GPU is a variant of the A100 designed for availability in markets with export control regulations. Featuring the same 80GB HBM2e memory and compute capabilities as the A100, the A800 has reduced NVLink interconnect bandwidth. It delivers equivalent per-GPU performance for AI training, HPC, and data analytics workloads while meeting regulatory requirements for deployment in specific regions.
Key Features
- 80GB HBM2e memory with 2 TB/s bandwidth
- Ampere architecture with 3rd-gen Tensor Cores
- Multi-Instance GPU (MIG) support
- Reduced NVLink for export compliance
- PCIe 4.0 and SXM form factors
Ideal Use Cases
ai trainingai inferencehpc
Full Specifications
Complete technical specifications for the A800.
Memory
- Memory
- 80GB HBM2e
- Memory Bandwidth
- 2 TB/s
Compute
- Tensor Cores
- 432 (3rd-gen)
- CUDA Cores
- 6912
- FP8 TFLOPS
- 624
- FP16 TFLOPS
- 312
- TF32 TFLOPS
- 156
- FP64 TFLOPS
- 19.5
Interconnect
- Interconnect
- NVLink 3.0, 400GB/s; PCIe 4.0
Physical
- TDP
- 400W
- Form Factor
- SXM4
- Cooling
- Liquid Cooled
Manufacturing
- Launch Date
- Q4 2022
- Manufacturing Process
- 7nm
