// RUN โ€” FITS

Can the Huawei Ascend 950DT run ByT5-XXL?

Yes โ€” here's how fast and up to what context, with real measured numbers.

ByT5-XXL on the Huawei Ascend 950DT

QuantFits?Max contextSpeedBeyond context
FP16 yes 642k 100.8-122.5 t/s ~1-1.2 (slow)
Q8 yes 721k 201.6-245 t/s ~2-2.4 (slow)
Q4 yes 755k 359.9-437.4 t/s ~3.6-4.4 (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 ByT5-XXL?โ–ถ

Yes. At Q4 it generates ~359.9-437.4 tok/s and fits up to 755k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~3.6-4.4 tok/s.

How much context fits?โ–ถ

At Q4, up to about 755k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.