Side by side datacenter GPU specifications from the Flopper.io database.
| Huawei Ascend 910B | Huawei Atlas 350 | |
|---|---|---|
| FP32 TFLOPS | — | — |
| Memory | 64 GB | 112 GB |
| Memory Bandwidth | 1.6 TB/s | 1.4 TB/s |
| TDP | 400 W | 600 W |
| Precision | Huawei Ascend 910B peak TFLOPS (dense) | Huawei Atlas 350 peak TFLOPS (dense) |
|---|---|---|
| FP64 | — | — |
| FP32 | — | — |
| TF32 | — | — |
| BF16 | 400.0 | 425.0 |
| FP16 | 400.0 | 425.0 |
| FP8 | — | 804.0 |
| FP6 | — | 804.0 |
| FP4 | — | 1561.0 |
| INT8 | 800.0 | 804.0 |
| Huawei Ascend 910B | Huawei Atlas 350 | |
|---|---|---|
| Architecture | Da Vinci v2 | Da Vinci v3 |
| Process Node | SMIC 7nm | — |
| Launch Year | 2024 | 2026 |
| Form Factor | OAM | Custom |
| Memory Type | HBM2e | HiBL 1.0 |
| Interface Width | — | — |
| Interconnect | HCCS | Unified Bus 2.0 |
| TDP | 400 W | 600 W |
| FP32 Efficiency | — | — |