// RUN โ€” FITS

Can the NVIDIA GeForce RTX 3080 10GB run MoEFold2D?

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

MoEFold2D on the NVIDIA GeForce RTX 3080 10GB

QuantFits?Max contextSpeedBeyond context
FP16 yes 117k 288958.3-344375 t/s ~15208.3-18125 (slow)
Q8 yes 117k 577916.7-688750 t/s ~30416.7-36250 (slow)
Q4 yes 117k 1031994-1229910.7 t/s ~54315.5-64732.1 (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 GeForce RTX 3080 10GB run MoEFold2D?โ–ถ

Yes. At Q4 it generates ~1031994-1229910.7 tok/s and fits up to 117k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~54315.5-64732.1 tok/s.

How much context fits?โ–ถ

At Q4, up to about 117k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.