// RUN โ€” FITS

Can the NVIDIA L40S 48GB run Routing Transformer (WT-103)?

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

Routing Transformer (WT-103) on the NVIDIA L40S 48GB

QuantFits?Max contextSpeedBeyond context
FP16 yes 649k 3966.8-4727.5 t/s ~183.6-218.9 (slow)
Q8 yes 650k 7933.6-9455.1 t/s ~367.3-437.7 (slow)
Q4 yes 650k 14167.1-16884.1 t/s ~655.9-781.7 (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 L40S 48GB run Routing Transformer (WT-103)?โ–ถ

Yes. At Q4 it generates ~14167.1-16884.1 tok/s and fits up to 650k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~655.9-781.7 tok/s.

How much context fits?โ–ถ

At Q4, up to about 650k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.