← AI Hashrate

Can L40S 48GB run DeepSeek-R1-Distill-Qwen-7B?

NVIDIA · 48 GB GDDR6 · 864 GB/s bandwidth · DeepSeek · 7.6B · model ctx up to 128K

Yes — the L40S 48GB (48 GB GDDR6) can run DeepSeek-R1-Distill-Qwen-7B. At Q4 with the default 8K context it needs ≈ 6.7 GB VRAM (weights 4.18 GB + KV cache 1.52 GB + 1 GB runtime overhead), which fits within 48 GB. Decode at Q4/8K: ≈ 72.3 tok/s (estimated, batch 1). FP16 (≈ 17.72 GB) also fits, at ≈ 19.9 tok/s (estimated). Estimates come from memory-bandwidth math; rows tagged measured override estimates. Relative ranking is more reliable than absolute tok/s. Methodology.

Fit & speed by quant and context

QuantContextVRAM neededFits?tok/s (decode)
Q44K5.94 GBYes72.3 est.
Q48K default6.7 GBYes72.3 est.
Q432K11.26 GBYes72.3 est.
FP164K16.96 GBYes19.9 est.
FP168K default17.72 GBYes19.9 est.
FP1632K22.28 GBYes19.9 est.

Fits = weights + KV(ctx) + 1 GB overhead ≤ 95% of VRAM. Measured anchors are context-agnostic; the fit verdict is recomputed per context. A missing tok/s means the model is far beyond this card (offload-only territory).

VRAM breakdown at 8K context

QuantWeightsKV cacheOverheadTotal neededL40S 48GB VRAM
Q44.18 GB1.52 GB1 GB6.7 GB48 GB
FP1615.2 GB1.52 GB1 GB17.72 GB48 GB

Other GPUs that run DeepSeek-R1-Distill-Qwen-7B

All GPUs for DeepSeek-R1-Distill-Qwen-7B →

Other models for the L40S 48GB

All models on the L40S 48GB →