Qwen3 30B A3B
36 records across 10 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RTX 4090 | Qwen3 30B A3B | Q4_K_M | llama.cpp (CUDA) | 225.9 | 1130.8 | – | – | tg128 | 2026-04-01 | community | – | source |
| RTX 4090 | Qwen3 30B A3B | Q4_K_M | llama.cpp (CUDA) | 223.2 | 1132.8 | – | – | tg128 | 2026-04-01 | community | – | source |
| RTX 4090 | Qwen3 30B A3B | Q4_K_M | llama.cpp (CUDA) | 211.0 | 1087.8 | – | – | tg128 | 2026-04-01 | community | – | source |
| RTX 4090 | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 207.3 | 6159.4 | – | – | 4096 | – | sourced | – | source |
| RTX 6000 Ada | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 199.6 | 3888.0 | – | – | 4096 | – | sourced | – | source |
| RTX 4090 | Qwen3 30B A3B | Q4_K_XL | llama.cpp llama-bench | 195.78 | – | – | – | 4096 | 2026-08-17 | sourced | – | source |
| RTX Pro 5000 Blackwell | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 188.8 | 5266.1 | – | – | 4096 | – | sourced | – | source |
| RTX 3090 Ti | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 166.9 | 3441.0 | – | – | 4096 | – | sourced | – | source |
| RTX Pro 5000 Blackwell | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 148.5 | 4043.8 | – | – | 16384 | – | sourced | – | source |
| RTX 4090 | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 139.2 | 4027.3 | – | – | 16384 | – | sourced | – | source |
| RTX A6000 | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 131.2 | 2142.4 | – | – | 4096 | – | sourced | – | source |
| RTX Pro 5000 Blackwell | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 123.2 | 3032.7 | – | – | 32768 | – | sourced | – | source |
| RTX 3090 Ti | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 121.9 | 2205.6 | – | – | 16384 | – | sourced | – | source |
| RTX 6000 Ada | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 120.1 | 3156.5 | – | – | 16384 | – | sourced | – | source |
| RTX 4090 | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 105.1 | 2627.7 | – | – | 32768 | – | sourced | – | source |
| RTX 3090 Ti | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 92.0 | 1483.9 | – | – | 32768 | – | sourced | – | source |
| RTX Pro 5000 Blackwell | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 91.9 | 1708.9 | – | – | 65536 | – | sourced | – | source |
| RTX A6000 | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 75.4 | 1901.1 | – | – | 16384 | – | sourced | – | source |
| RTX 4090 | Qwen3 30B A3B | Q4_K_XL | llama.cpp llama-bench | 74.64 | – | – | – | 57344 | 2026-08-17 | sourced | – | source |
| RTX 6000 Ada | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 74.3 | 2010.6 | – | – | 32768 | – | sourced | – | source |
| RTX 4090 | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 68.2 | 1502.5 | – | – | 65536 | – | sourced | – | source |
| RTX Pro 5000 Blackwell | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 61.4 | 702.7 | – | – | 131072 | – | sourced | – | source |
| RTX A6000 | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 48.1 | 1275.6 | – | – | 32768 | – | sourced | – | source |
| RTX 6000 Ada | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 40.8 | 913.2 | – | – | 65536 | – | sourced | – | source |
| M3 Pro 36GB | Qwen3 30B A3B | Q6_K | llama.cpp (CUDA) | 38.1 | – | – | – | – | 2026-10-05 | community | – | source |
| RTX Pro 5000 Blackwell | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 37.0 | 364.1 | – | – | 262144 | – | sourced | – | source |
| RTX 5060 Ti | Qwen3 30B A3B 30B-A3B | Q4_K_M | llama.cpp (CUDA) | 35.4 | 149.6 | – | – | tg128 | 2026-10-05 | community | – | source |
| M4 | Qwen3 30B A3B | 4bit | oMLX | 31.3 | 426.4 | – | – | 4096 | 2026-10-14 | community | – | source |
| RTX A6000 | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 27.6 | 764.2 | – | – | 65536 | – | sourced | – | source |
| M3 Pro 36GB | Qwen3 30B A3B | Q6_K | llama.cpp (CUDA) | 27.5 | – | – | – | – | 2026-10-05 | community | – | source |
| M3 Pro 36GB | Qwen3 30B A3B | Q6_K | llama.cpp (CUDA) | 22.1 | – | – | – | – | 2026-10-05 | community | – | source |
| RTX 6000 Ada | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 20.7 | 448.6 | – | – | 131072 | – | sourced | – | source |
| RTX 5060 Ti | Qwen3 30B A3B 30B-A3B | Q4_K_M | llama.cpp (CUDA) | 19.7 | 64.0 | – | – | tg128 | 2026-10-05 | community | – | source |
| M4 Max (128 GB) | Qwen3 30B A3B | Q4_K_M | llama.cpp | 16.2 | – | – | – | 32768 | 2026-10-17 | sourced | – | source |
| RTX A6000 | Qwen3 30B A3B 30B | Q4_K | llama.cpp | 14.8 | 429.2 | – | – | 131072 | – | sourced | – | source |
| M1 Pro 16GB | Qwen3 30B A3B | Q6_K | llama.cpp (CUDA) | 13 | – | – | – | – | 2026-10-05 | community | – | source |