// RUN โ€” FITS

Can the AMD Instinct MI300X 192GB run ViT-G (model soup)?

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

ViT-G (model soup) on the AMD Instinct MI300X 192GB

QuantFits?Max contextSpeedBeyond context
FP16 yes 2616k 905.9-1107.2 t/s ~6.8-8.4 (slow)
Q8 yes 2644k 1811.7-2214.3 t/s ~13.7-16.7 (slow)
Q4 yes 2657k 3235.2-3954.2 t/s ~24.4-29.8 (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 MI300X 192GB run ViT-G (model soup)?โ–ถ

Yes. At Q4 it generates ~3235.2-3954.2 tok/s and fits up to 2657k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~24.4-29.8 tok/s.

How much context fits?โ–ถ

At Q4, up to about 2657k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.