Side by side datacenter GPU specifications from the Flopper.io database.
| Huawei Ascend 910A | NVIDIA RTX PRO 5000 Blackwell 48GB | |
|---|---|---|
| FP32 TFLOPS | 16.0 | 65.0 |
| Memory | 32 GB | 48 GB |
| Memory Bandwidth | 1.2 TB/s | 1.3 TB/s |
| TDP | 310 W | 300 W |
| Precision | Huawei Ascend 910A peak TFLOPS (dense) | NVIDIA RTX PRO 5000 Blackwell 48GB peak TFLOPS (dense) |
|---|---|---|
| FP64 | — | — |
| FP32 | 16.0 | 65.0 |
| TF32 | — | — |
| BF16 | — | — |
| FP16 | 256.0 | — |
| FP8 | — | — |
| FP6 | — | — |
| FP4 | — | 1032.0 |
| INT8 | 512.0 | — |
| Huawei Ascend 910A | NVIDIA RTX PRO 5000 Blackwell 48GB | |
|---|---|---|
| Architecture | Da Vinci | Blackwell |
| Process Node | TSMC 7nm | 4nm |
| Launch Year | 2019 | 2025 |
| Form Factor | OAM | PCIe |
| Memory Type | HBM2 | GDDR7 |
| Interface Width | — | 512-bit |
| Interconnect | HCCS | — |
| TDP | 310 W | 300 W |
| FP32 Efficiency | 0.05 TFLOPS/W | 0.22 TFLOPS/W |