Skip to content

Can RTX 4090 run Qwen2.5 7B Instruct?

Comfortable: GGUF Q6_K needs 6.8 GB at 4K, batch 1; headroom 17.2 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_M1555.4

Quant memory at 4K

QuantTotal GBVerdictWeight sizing
GGUF Q4_K_M5.1ComfortableEstimated weights
GGUF Q6_K6.8ComfortableEstimated weights
AWQ INT44.2ComfortableEstimated weights
EXL2 4.65bpw4.9ComfortableEstimated weights

Context capacity · GGUF Q6_K

ContextTotal GBVerdict
4K6.8Comfortable
16K7.5Comfortable
32K8.4Comfortable
128K14.2Comfortable

Generation speed

Measured: 155 tok/s, GGUF Q4_K_M, llama.cpp

GGUF Q6_K: 113–149 tok/s · Derived range

Other model / GPU pairs