RunLLMLocal
Open the checker

Compare · Models

Qwen3 30B-A3B (MoE) vs Qwen3 32B: which runs better locally?

Qwen3 30B-A3B (MoE) needs about 20.6 GB at Q4 with 8K context and fits fully on 4 of 40 GPUs. Qwen3 32B needs 23.3 GB and fits on 1. On cards that run both, Qwen3 30B-A3B (MoE) is about 3.4x faster because it is a mixture-of-experts model.

VerdictQwen3 30B-A3B (MoE) runs faster at a similar size

Qwen3 30B-A3B (MoE) needs about 20.6 GB at Q4 with 8K context and fits fully on 4 of 40 GPUs. Qwen3 32B needs 23.3 GB and fits on 1. On cards that run both, Qwen3 30B-A3B (MoE) is about 3.4x faster because it is a mixture-of-experts model.

Qwen3 30B-A3B (MoE)Chat, Coding, Reasoning
Parameters30.5B (3.3B active)
Memory at Q4, 8K20.6 GB
Memory at Q8, 8K36.3 GB
GPUs that run it fully4 / 40
Cheapest desktop cardRTX 3090 / 3090 Ti (24 GB)
Qwen3 32BChat, Coding, Reasoning
Parameters32.8B
Memory at Q4, 8K23.3 GB
Memory at Q8, 8K40.3 GB
GPUs that run it fully1 / 40
Cheapest desktop cardRTX 5090 (32 GB)

Each GPU at 8K context with the best quantization that fits. Offload assumes 32 GB of system RAM. Green marks the better result. Rows where the two differ come first.

GPUQwen3 30B-A3B (MoE)Qwen3 32B
RTX 4090 24 GBRuns~130 tok/sQ4_K_MOffload~4.8 tok/sQ4_K_M
RX 7900 XTX 24 GBRuns~130 tok/sQ4_K_MOffload~4.6 tok/sQ4_K_M
RTX 3090 / 3090 Ti 24 GBRuns~120 tok/sQ4_K_MOffload~4.5 tok/sQ4_K_M
RTX 5090 32 GBRuns~170 tok/sQ6_KRuns~50 tok/sQ5_K_M
RX 7900 XT 20 GBOffload~15 tok/sQ4_K_MOffload~3.5 tok/sQ4_K_M
RTX 5080 16 GBOffload~17 tok/sQ4_K_MOffload~3.7 tok/sQ4_K_M
RTX 5070 Ti 16 GBOffload~16 tok/sQ4_K_MOffload~3.5 tok/sQ4_K_M
RTX 4080 / 4080 Super 16 GBOffload~14 tok/sQ4_K_MOffload~3.0 tok/sQ4_K_M
RTX 4070 Ti Super 16 GBOffload~13 tok/sQ4_K_MOffload~2.8 tok/sQ4_K_M
RX 9070 XT 16 GBOffload~13 tok/sQ4_K_MOffload~2.7 tok/sQ4_K_M
RX 7800 XT 16 GBOffload~13 tok/sQ4_K_MOffload~2.6 tok/sQ4_K_M
Arc A770 16 GBOffload~12 tok/sQ4_K_MOffload~2.4 tok/sQ4_K_M
RX 6800 XT 16 GBOffload~11 tok/sQ4_K_MOffload~2.3 tok/sQ4_K_M
RTX 5060 Ti 16 GBOffload~10 tok/sQ4_K_MOffload~2.1 tok/sQ4_K_M
RTX 4060 Ti 16 GBOffload~7.9 tok/sQ4_K_MOffload~1.6 tok/sQ4_K_M
RTX 5070 12 GBOffload~13 tok/sQ4_K_MOffload~2.5 tok/sQ4_K_M
RTX 4070 / 4070 Super 12 GBOffload~11 tok/sQ4_K_MOffload~2.1 tok/sQ4_K_M
Arc B580 12 GBOffload~11 tok/sQ4_K_MOffload~2.0 tok/sQ4_K_M
RX 7700 XT 12 GBOffload~11 tok/sQ4_K_MOffload~2.0 tok/sQ4_K_M
RX 6700 XT 12 GBOffload~10 tok/sQ4_K_MOffload~1.9 tok/sQ4_K_M
RTX 3060 12 GBOffload~10 tok/sQ4_K_MOffload~1.8 tok/sQ4_K_M
GTX 1080 Ti 11 GBOffload~11 tok/sQ4_K_MOffload~2.1 tok/sQ4_K_M
RTX 3080 10 GBOffload~14 tok/sQ4_K_MOffload~2.6 tok/sQ4_K_M
Arc A750 8 GBOffload~12 tok/sQ4_K_MOffload~2.0 tok/sQ4_K_M
RTX 5060 Ti 8 GBOffload~12 tok/sQ4_K_MOffload~1.9 tok/sQ4_K_M
RTX 5060 8 GBOffload~12 tok/sQ4_K_MOffload~1.9 tok/sQ4_K_M
RTX 3070 / 3070 Ti 8 GBOffload~12 tok/sQ4_K_MOffload~1.9 tok/sQ4_K_M
RTX 3060 Ti 8 GBOffload~12 tok/sQ4_K_MOffload~1.9 tok/sQ4_K_M
RTX 4060 Ti 8 GBOffload~11 tok/sQ4_K_MOffload~1.7 tok/sQ4_K_M
RX 7600 8 GBOffload~11 tok/sQ4_K_MOffload~1.7 tok/sQ4_K_M
RTX 4060 8 GBOffload~10 tok/sQ4_K_MOffload~1.7 tok/sQ4_K_M
RTX 3050 8 GBOffload~10 tok/sQ4_K_MOffload~1.6 tok/sQ4_K_M
GTX 1660 / 1660 Super 6 GBOffload~11 tok/sQ4_K_MOffload~1.8 tok/sQ4_K_M
GTX 1650 4 GBOffload~11 tok/sQ4_K_MOffload~2.0 tok/sQ3_K_M
RTX 4090 Laptop 16 GBOffload~12 tok/sQ4_K_MOffload~2.5 tok/sQ4_K_M
RTX 4080 Laptop 12 GBOffload~11 tok/sQ4_K_MOffload~2.0 tok/sQ4_K_M
RTX 4070 Laptop 8 GBOffload~10 tok/sQ4_K_MOffload~1.6 tok/sQ4_K_M
RTX 4060 Laptop 8 GBOffload~10 tok/sQ4_K_MOffload~1.6 tok/sQ4_K_M
RTX 3060 Laptop 6 GBOffload~11 tok/sQ4_K_MOffload~1.8 tok/sQ4_K_M
RTX 4050 Laptop 6 GBOffload~11 tok/sQ4_K_MOffload~1.6 tok/sQ4_K_M

Which should you run?

Qwen3 30B-A3B (MoE) runs faster at a similar size. Qwen3 30B-A3B (MoE) needs about 20.6 GB at Q4 with 8K context and fits fully on 4 of 40 GPUs. Qwen3 32B needs 23.3 GB and fits on 1. On cards that run both, Qwen3 30B-A3B (MoE) is about 3.4x faster because it is a mixture-of-experts model.

Only the Qwen3 30B-A3B (MoE) fits on RTX 4090, RTX 3090 / 3090 Ti and RX 7900 XTX.

See every result on its own page: Qwen3 30B-A3B (MoE) and Qwen3 32B.

Questions

Is Qwen3 30B-A3B (MoE) or Qwen3 32B better?

Qwen3 30B-A3B (MoE) runs faster at a similar size. Qwen3 30B-A3B (MoE) needs about 20.6 GB at Q4 with 8K context and fits fully on 4 of 40 GPUs. Qwen3 32B needs 23.3 GB and fits on 1. On cards that run both, Qwen3 30B-A3B (MoE) is about 3.4x faster because it is a mixture-of-experts model. The right pick depends on what you use it for: compare the strengths listed on each model's page.

Which needs less VRAM, Qwen3 30B-A3B (MoE) or Qwen3 32B?

Qwen3 30B-A3B (MoE) needs about 21 GB at Q4 with 8K context, Qwen3 32B about 23 GB. At Q8 it is 36 GB versus 40 GB.

Can I run Qwen3 30B-A3B (MoE) and Qwen3 32B on a 12 GB GPU?

On an RTX 3060 12 GB, Qwen3 30B-A3B (MoE) only runs with part of the model in system RAM, at a few tokens per second and Qwen3 32B only runs with part of the model in system RAM, at a few tokens per second.

More comparisons

gpt-oss 20B vs Qwen3 30B-A3B (MoE)Model comparisonQwen3.6 35B-A3B (MoE) vs Qwen3 30B-A3B (MoE)Model comparisonLlama 3.3 70B vs Qwen3 32BModel comparisonQwen3 14B vs Gemma 4 12BModel comparisonQwen3 8B vs Llama 3.1 8BModel comparisonGemma 4 31B vs Qwen3.6 27BModel comparison

Estimates from memory size and bandwidth at 8K context. Real speed varies with software, drivers and settings.

Copied