Skip to content

Can RTX 4090 run DeepSeek-R1-Distill-Llama-8B?

Comfortable: GGUF Q5_K_M needs 6.5 GB at 4K, batch 1; headroom 17.5 GB.

Text input, FP16 KV cache, one card. Unified-memory capacity uses the GPU guide’s usable limit; negative headroom may require host offload. Speed source settings may differ from this 4K estimate.

Measured runs

FrameworkQuanttok/sVRAM used (GB)
llama.cppGGUF Q4_K_M1455.6

Quant memory at 4K

QuantTotal GBVerdictWeight sizing
GGUF Q4_K_M5.6ComfortableEstimated weights
GGUF Q5_K_M6.5ComfortableEstimated weights
EXL2 4.65bpw5.4ComfortableEstimated weights
AWQ INT44.8ComfortableEstimated weights

Context capacity · GGUF Q5_K_M

ContextTotal GBVerdict
4K6.5Comfortable
16K8.2Comfortable
32K10.4Comfortable
128K23.6Tight

Generation speed

Measured: 145 tok/s, GGUF Q4_K_M, llama.cpp

GGUF Q5_K_M: 124–164 tok/s · Derived range

Other model / GPU pairs