can-i-run-this-llm

Can I run this LLM?

Can a NVIDIA L40S (48GB) run Gemma 4 26B-A4B (MoE)?

Gemma · Apache 2.0 · estimated for a 48.0 GB machine at 4K context.

Yes — it runs great.

Best fit: Gemma 4 26B-A4B (MoE) at Q8_0 · 26.9 GB · 101–152 tok/s on a NVIDIA L40S (48GB).

Every quantization on this machine

QuantSizeSpeedFit
Q3_K_M12.7 GB158–236 tok/sRuns great
Q4_K_M17.0 GB135–203 tok/sRuns great
Q5_K_M21.2 GB119–178 tok/sRuns great
Q6_K23.2 GB112–168 tok/sRuns great
Q8_026.9 GB101–152 tok/sRuns great
FP1650.5 GB64–95 tok/sWon't fit

Ran Gemma 4 26B-A4B (MoE) on a NVIDIA L40S (48GB)? Add your real numbers →