// RUN โ€” FITS

Can the AMD Instinct MI325X run HiFi - NN?

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

HiFi - NN on the AMD Instinct MI325X

QuantFits?Max contextSpeedBeyond context
FP16 yes 3571k 630000-770000 t/s ~4200-5133.3 (slow)
Q8 yes 3571k 1260000-1540000 t/s ~8400-10266.7 (slow)
Q4 yes 3571k 2250000-2750000 t/s ~15000-18333.3 (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 AMD Instinct MI325X run HiFi - NN?โ–ถ

Yes. At Q4 it generates ~2250000-2750000 tok/s and fits up to 3571k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~15000-18333.3 tok/s.

How much context fits?โ–ถ

At Q4, up to about 3571k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.