// RUN โ€” FITS

Can the NVIDIA B200 run mBART-50?

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

mBART-50 on the NVIDIA B200

QuantFits?Max contextSpeedBeyond context
FP16 yes 2654k 4786.9-5704.9 t/s ~23.9-28.5 (slow)
Q8 yes 2663k 9573.8-11409.8 t/s ~47.9-57 (slow)
Q4 yes 2667k 17096-20374.7 t/s ~85.5-101.9 (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 NVIDIA B200 run mBART-50?โ–ถ

Yes. At Q4 it generates ~17096-20374.7 tok/s and fits up to 2667k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~85.5-101.9 tok/s.

How much context fits?โ–ถ

At Q4, up to about 2667k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.