NVIDIA · Ampere · A100
80 GB of HBM2e at 2.0 TB/s, built for training, inference, fine-tuning.
From $1.19/hracross 8 providers
Memory
80 GB
HBM2e
Bandwidth
2.04 TB/s
FP16 tensor
312 TFLOPS
FP8 tensor
—
no FP8 path
Interconnect
PCIe/NVLink
TDP
400 W
CUDA cores
6,912
Tensor cores
432
| Provider | Type | Spot | Details |
|---|---|---|---|
 RunPod | neocloud | Yes | View |
 Vast.ai | marketplace | Yes | View |
 Lambda | neocloud | — | View |
 Crusoe | neocloud | — | View |
 Amazon AWS | hyperscaler | Yes | View |
 Google Cloud | hyperscaler | Yes | View |
 Microsoft Azure | hyperscaler | Yes | View |
 Oracle Cloud | hyperscaler | Yes | View |
