← AI Hashrate

Can RTX 5090 32GB run DeepSeek-R1-Distill-Llama-70B?

NVIDIA · 32 GB GDDR7 · 1792 GB/s bandwidth · DeepSeek · 70.6B · model ctx up to 128K

No — DeepSeek-R1-Distill-Llama-70B at Q4 needs ≈ 46.89 GB even at 4K context, beyond the RTX 5090 32GB's 32 GB GDDR7. It would only run with heavy CPU/disk offload. Decode at Q4/8K: ≈ 16.2 tok/s (estimated, batch 1). FP16 needs ≈ 156.32 GB — does not fit on this card. 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)
Q44K46.89 GBNo21.5 est.
Q48K default53.95 GBNo16.2 est.
Q432K96.31 GBNo5.1 est.
FP164K149.26 GBNo
FP168K default156.32 GBNo
FP1632K198.68 GBNo

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 neededRTX 5090 32GB VRAM
Q438.83 GB14.12 GB1 GB53.95 GB32 GB
FP16141.2 GB14.12 GB1 GB156.32 GB32 GB

Other GPUs that run DeepSeek-R1-Distill-Llama-70B

All GPUs for DeepSeek-R1-Distill-Llama-70B →

Other models for the RTX 5090 32GB

All models on the RTX 5090 32GB →