Side by side datacenter GPU specifications from the Flopper.io database.
| AWS Trainium2 | NVIDIA H100 SXM5 80GB | |
|---|---|---|
| FP32 TFLOPS | 181.0 | 67.0 |
| Memory | 96 GB | 80 GB |
| Memory Bandwidth | 2.9 TB/s | 3.4 TB/s |
| TDP | — | 700 W |
| Precision | AWS Trainium2 peak TFLOPS (dense) | NVIDIA H100 SXM5 80GB peak TFLOPS (dense) |
|---|---|---|
| FP64 | — | 34.0 |
| FP64_TC Tensor Core | — | 67.0 |
| FP32 | 181.0 | 67.0 |
| TF32 | 667.0 | 494.0 |
| BF16 | 667.0 | 990.0 |
| FP16 | 667.0 | 990.0 |
| FP8 | 1299.0 | 1979.0 |
| FP6 | — | — |
| FP4 | — | — |
| INT8 | — | 1979.0 |
| AWS Trainium2 | NVIDIA H100 SXM5 80GB | |
|---|---|---|
| Architecture | NeuronCore-v3 | Hopper |
| Process Node | — | 4nm |
| Launch Year | 2024 | 2022 |
| Form Factor | — | SXM |
| Memory Type | — | — |
| Interface Width | — | — |
| Interconnect | NeuronLink-v3 | — |
| TDP | — | 700 W |
| FP32 Efficiency | — | 0.10 TFLOPS/W |