Home › GPUs & Macs › NVIDIA RTX 3090 vs NVIDIA RTX 4090

NVIDIA RTX 3090 vs NVIDIA RTX 4090 for local LLMs

Across 8 tests measured on both, the NVIDIA RTX 4090 is faster in 7 and generates a median 1.15× the tokens/s of the NVIDIA RTX 3090.

On Qwen3.8 27B (Q4_K_M, llama-server) the NVIDIA RTX 3090 generates 36.6 tokens/s and the NVIDIA RTX 4090 47.7 tokens/s (measured).

Same model, same quant, same test

ModelQuant / testRTX 3090
tok/s		RTX 4090
tok/s		FasterSources and run details
Qwen3.8 27BQ4_K_M
llama-server		36.6
CUDA		47.7
RTX 4090 +30%qwen38-mtp RTX 3090 · qwen38-mtp RTX 4090
Qwen3.8 27BUD-Q4_K_XL
llama-bench		36.0
CUDA		47.9
CUDA		RTX 4090 +33%changtimwu/vastai-exp issue #11 RTX 3090 · ComputingForGeeks Qwen3.8-27B test RTX 4090
Qwen3.8 27BUD-Q4_K_XL
llama-server		34.0
CUDA		40.4
CUDA		RTX 4090 +19%qwen38-mtp RTX 3090 · qwen38-mtp RTX 4090
Llama 2 7BQ4_0
llama-bench		172
Vulkan		190
Vulkan		RTX 4090 +11%llama.cpp #10879 RTX 3090 · llama.cpp #10879 RTX 4090
gpt-oss 20BMXFP4
llama-bench		162
CUDA		255
CUDA		RTX 4090 +57%llama.cpp #15396 RTX 3090 · llama.cpp #15396 RTX 4090
Llama 3.1 8B InstructQ4_K_M
LocalScore		96.3
llamafile		105
llamafile		RTX 4090 +9%LocalScore RTX 3090 · LocalScore RTX 4090
Llama 3.2 1B InstructQ4_K_M
LocalScore		270
llamafile		292
llamafile		RTX 4090 +8%LocalScore RTX 3090 · LocalScore RTX 4090
Qwen2.5 14B InstructQ4_K_M
LocalScore		56.2
llamafile		49.1
llamafile		RTX 3090 +14%LocalScore RTX 3090 · LocalScore RTX 4090

Generation speed in tokens/s, best measured backend for each chip (shown under the number). Backends can differ between the two chips; check them before drawing conclusions. Measured results only, nothing estimated.

Specs

NVIDIA RTX 3090NVIDIA RTX 4090
Memory24 GB24 GB
Memory bandwidth936 GB/s1008 GB/s

All results for the NVIDIA RTX 3090 · All results for the NVIDIA RTX 4090 · More comparisons