// RUN โ€” FITS

Can the Intel Gaudi 3 run phi-3-mini 3.8B?

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

phi-3-mini 3.8B on the Intel Gaudi 3

QuantFits?Max contextSpeedBeyond context
FP16 yes 1658k 282.4-350.5 t/s ~3.1-3.8 (slow)
Q8 yes 1716k 564.7-701.1 t/s ~6.1-7.6 (slow)
Q4 yes 1742k 1008.5-1251.9 t/s ~10.9-13.5 (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 Intel Gaudi 3 run phi-3-mini 3.8B?โ–ถ

Yes. At Q4 it generates ~1008.5-1251.9 tok/s and fits up to 1742k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~10.9-13.5 tok/s.

How much context fits?โ–ถ

At Q4, up to about 1742k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.