Side by side datacenter GPU specifications from the Flopper.io database.
| Huawei Ascend 950PR | NVIDIA RTX PRO 6000 Blackwell Server Edition | |
|---|---|---|
| FP32 TFLOPS | — | 120.0 |
| Memory | 128 GB | 96 GB |
| Memory Bandwidth | 1.6 TB/s | 1.6 TB/s |
| TDP | 900 W | 600 W |
| Precision | Huawei Ascend 950PR peak TFLOPS (dense) | NVIDIA RTX PRO 6000 Blackwell Server Edition peak TFLOPS (dense) |
|---|---|---|
| FP64 | — | — |
| FP32 | — | 120.0 |
| TF32 | — | 234.0 |
| BF16 | 500.0 | 500.0 |
| FP16 | 500.0 | 500.0 |
| FP8 | 1000.0 | 1000.0 |
| FP6 | — | — |
| FP4 | 2000.0 | 2000.0 |
| INT8 | — | — |
| Huawei Ascend 950PR | NVIDIA RTX PRO 6000 Blackwell Server Edition | |
|---|---|---|
| Architecture | Da Vinci v3 | Blackwell |
| Process Node | — | 4nm |
| Launch Year | 2026 | 2025 |
| Form Factor | OAM | PCIe |
| Memory Type | HiBL 1.0 | GDDR7 |
| Interface Width | — | 512-bit |
| Interconnect | Unified Bus 2.0 | — |
| TDP | 900 W | 600 W |
| FP32 Efficiency | — | 0.20 TFLOPS/W |