New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
Can the Huawei Ascend 950DT run ALBERT-xxlarge?
Yes — here's how fast and up to what context, with real measured numbers.
ALBERT-xxlarge on the Huawei Ascend 950DT
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 1991k | 5531.9-6723.4 t/s | ~55.3-67.2 (slow) |
| Q8 | yes | 1995k | 11063.8-13446.8 t/s | ~110.6-134.5 (slow) |
| Q4 | yes | 1997k | 19756.8-24012.2 t/s | ~197.6-240.1 (slow) |
Estimate (memory-bound). Beyond "max context" the KV cache spills to system RAM offload (depends on your system RAM) and speed drops to the "beyond" figure.
FAQ
Can the Huawei Ascend 950DT run ALBERT-xxlarge?▶
Yes. At Q4 it generates ~19756.8-24012.2 tok/s and fits up to 1997k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~197.6-240.1 tok/s.
How much context fits?▶
At Q4, up to about 1997k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.