Product
NVIDIA A100 80GB SXM
Part number NVIDIA A100 80GB SXM
- Price
- USD $100
- Stock
- 1
Ampere-architecture SXM accelerator with 80 GB of HBM2e memory at 2,039 GB/s, NVLink at 600 GB/s, and a standard 400 W TDP, still widely used for training and inference on HGX A100 systems.
- 80 GB HBM2e memory at 2,039 GB/s
- SXM form factor for NVIDIA HGX A100 systems
- NVLink at 600 GB/s plus PCIe Gen4
- Standard TDP of 400 W; CTS SKUs can support up to 500 W
- Up to seven MIG instances at 10 GB each
Specifications
| GPU Model | NVIDIA A100 80GB SXM |
|---|---|
| Architecture | NVIDIA Ampere |
| Form Factor | SXM |
| GPU Count | 1 |
| GPU Memory | 80 GB HBM2e |
| Memory Bandwidth | 2,039 GB/s |
| Interconnect | NVLink 600 GB/s; PCIe Gen4 64 GB/s |
| Max TDP | 400 W standard (CTS SKU up to 500 W) |
| Cooling | Host-cooled SXM module |
| FP64 | 9.7 TFLOPS |
| FP64 Tensor Core | 19.5 TFLOPS |
| FP32 | 19.5 TFLOPS |
| TF32 Tensor Core | 156 / 312 TFLOPS (with sparsity) |
| FP16 / BFLOAT16 Tensor Core | 312 / 624 TFLOPS (with sparsity) |
| INT8 Tensor Core | 624 / 1,248 TOPS (with sparsity) |
| Multi-Instance GPU | Up to 7 MIGs at 10 GB each |
| Typical Host | NVIDIA HGX A100 (4, 8, or 16 GPUs) and NVIDIA DGX A100 |
Description
The NVIDIA A100 80GB SXM is the Ampere-architecture Tensor Core GPU used on NVIDIA HGX A100 boards and in DGX A100 systems. It remains a widely deployed training and inference accelerator, even as Hopper and Blackwell systems take the top of the catalog.
NVIDIA’s official table lists 80 GB of HBM2e memory with 2,039 GB/s of bandwidth, NVLink at 600 GB/s, and PCIe Gen4 at 64 GB/s. Peak figures include 9.7 TFLOPS of FP64, 19.5 TFLOPS of FP32, and, with sparsity, 312 TFLOPS of TF32 Tensor performance and 624 TFLOPS of FP16/BF16 Tensor performance. Multi-Instance GPU can split the device into as many as seven 10 GB instances. Standard thermal design power is 400 W; NVIDIA notes that the HGX A100 80GB custom thermal solution (CTS) SKU can support TDPs up to 500 W.
In an HGX A100 server, NVSwitch can connect four, eight, or — in some partner designs — sixteen A100 SXM GPUs at the full 600 GB/s NVLink rate. The module is host-cooled. Compared with H100 SXM, A100 offers less memory bandwidth and no FP8 Transformer Engine, but it is still a capable node for established training pipelines and production inference.
If you are planning a new eight-GPU fabric, also consider H100 or H200 SXM. Request a quote if you need A100 80GB SXM modules or a compatible HGX A100-class host.
Related products
NVIDIA H100 80GB SXM5
P/N NVIDIA H100 80GB SXM5
Price USD $100 Stock 1
Hopper-architecture SXM accelerator with 80 GB of HBM3 memory, fourth-generation NVLink, and a configurable TDP up to 700 W, built for dense HGX training and inference nodes.
- GPU Model
- NVIDIA H100 80GB SXM5
- Form Factor
- SXM5
- GPU Count
- 1
- GPU Memory
- 80 GB HBM3
Ready to specify a configuration?
Share the use case, the target build, quantity, destination country, and timeline. We will take it from the inquiry.