← Back to catalog
NVIDIADatacenter
NVIDIA A100 80GB SXM4
NVLink-connected Ampere for real multi-GPU training pods.
$16,000
80GB HBM2e
Memory
NVLink 3.0, 600 GB/s
Interconnect
312 TFLOPS
FP16 Tensor
SXM4 module with full 3.0 bandwidth for 4- and 8-GPU HGX baseboards — the classic building block behind most pre-2023 large model training runs, still very capable today.
Full Specifications
Compute
- Architecture
- Ampere
- FP16 Tensor
- 312
Memory
- Memory
- 80GB HBM2e
- Memory Bandwidth
- 2.0 TB/s
Interconnect & Networking
- Interconnect
- 3.0, 600 GB/s
Power & Cooling
- 400W
General
- Form Factor
- SXM4 module
Compatible & Required
Pairs well with