…
Cloud rental price, compute per dollar and specs, side by side.
Open in the interactive comparer →| Spec / price | A100 SXM 80GB | A100 PCIe 80GB |
|---|---|---|
| Memory | 80 GB HBM2e | 80 GB HBM2e |
| Memory bandwidth | 2,039 GB/sbetter | 1,935 GB/s |
| FP16 / BF16 (dense) | 0.312 PF | 0.312 PF |
| FP8 (dense) | — | — |
| FP4 (dense) | — | — |
| INT8 (dense) | 0.624 PF | 0.624 PF |
| TF32 (dense) | 0.156 PF | 0.156 PF |
| FP32 (dense) | 0.0195 PF | 0.0195 PF |
| FP64 (dense) | 0.0195 PF | 0.0195 PF |
| Power | 400 W | 300 W |
| Architecture | Ampere (2020) | Ampere (2021) |
| Lowest on-demand | $0.81better | $1.13 |
| Median on-demand | $2.35 | $1.51better |
| Lowest spot | $0.90 | $0.21better |
| FP16/BF16 $ per PFLOP·h | $2.60better | $3.62 |
| Providers | 17 | 8 |
The A100 SXM 80GB and A100 PCIe 80GB have the same dense FP16/BF16 compute. The A100 SXM 80GB has 1.1× more memory bandwidth than the A100 PCIe 80GB. Real-world speedups depend on the workload: training and prefill scale with compute, while LLM decoding scales mostly with memory bandwidth.
Right now the A100 SXM 80GB is cheaper per unit of compute: $2.60 vs $3.62 per PFLOP-hour at the lowest on-demand prices.
Running one GPU around the clock at the lowest on-demand price costs about $591 per month for the A100 SXM 80GB and $824 for the A100 PCIe 80GB.
Both have 80 GB of VRAM; the A100 SXM 80GB has more bandwidth.