Key Architecture Highlights
- GPU Architecture: NVIDIA Ampere
- GPU memory specifications: GPU memory: 80GB HBM2e GPU memory ECC: Enabled
- GPU memory bandwidth: 1,935 GB/s
Applications
- AI training and inference
- High-performance computing
- Analysis of data
Engineering Parameters
| GPU Architecture | NVIDIA Ampere |
|---|---|
| GPU memory specifications | GPU memory: 80GB HBM2e GPU memory ECC: Enabled |
| GPU memory bandwidth | 1,935 GB/s |
| GPU memory bus width | 5120 bit |
| General-purpose compute performance | FP32: 19.5 TFLOPS FP64: 9.7 TFLOPS FP64 Tensor Core: 19.5 TFLOPS |
| Tensor Compute performance | TF32 Tensor Core: common:156 TFLOPS; sparse:312 TFLOPS FP16 Tensor Core: common:312 TFLOPS; sparse:624 TFLOPS BF16 Tensor Core: common:312 TFLOPS; sparse:624 TFLOPS |
| Integer compute performance | The first is the INT8 Tensor Core: normal: 624 TOPS; sparse: 1248 TOPS. |
| PCIe Interface | PCIe Gen4 x16 |
| NVLink / Interconnection | Connectivity: NVIDIA® NVLink® Bridge for 2 GPUs: 600 GB/s PCIe Gen4: 64 GB/s |
| MIG Division | The maximum number of instances is 7 and the maximum number of instances is 10GB. |
| Board power | 300W by default/maximum and minimum 150W |
| Power input Interface | 1 CPU 8-pin power input interface |
| Card form factor / cooling | card form factor: Full length (FHFL) 10.5 ton, double-slit cooling method: passive cooling (requires server wind tunnel) |
| GPU Frequency | Base: 1065 MHz Boost: 1410 MHz |
| Operating temperature | 0 °C to 50 °C |
Warranty
manufacturer's public product information does not specify the warranty period for this model in its supply channel; applicable region, channel and actual warranty service confirmed by the supply contract.
Manufacturer information
Manufacturer informationVerified Documentation & Datasheets
Official products page
Download DocumentPB-10577-001_v03 | March 2022 NVIDIA A100 80GB PCIe GPU Product Brief
Download Document