Can an RTX 5090 run Qwen3-32B?
Yes: an RTX 5090 runs Qwen3 32B at 69.8 tok/s at Q4_K_M, measured on llama.cpp with a 110-token prompt in October 2026.
3 signed runs on llm-speed. Single-request decode speed in tokens per second; the short-chat workload unless noted. 1 of 3 runs were measured by llm-speed (@llmspeed); the other 2 are community-submitted (signed on the submitter's own machine, not measured by llm-speed, and not used for llm-speed verdicts). Runs measured by llm-speed are credited to @llmspeed. Verdicts and headline numbers use only llm-speed's own runs and fit tests.
Measured tok/s by quant and runtime
| Quant | Runtime | Build | Decode tok/s (best run) | Workload | Runs | Source |
|---|---|---|---|---|---|---|
| Q4_K_M | llama.cpp | commit 1268eda | 69.8 | chat-short | 1 | Measured by llm-speed@llmspeed |
| not reported | llama.cpp | not reported | 69.5 | chat-short | 2 | Community-submitted |
Fastest quant that fits: Q4_K_M on llama.cpp, 69.8 tok/s (measured by llm-speed; it fits because it ran on this hardware). Quants without a run are not listed.
Every measured row
| Run | Date (UTC) | Workload | Quant | Runtime / build | Prompt tokens | Output tokens | Decode tok/s | Prefill tok/s | TTFT ms | VRAM GiB | Source |
|---|---|---|---|---|---|---|---|---|---|---|---|
| r_vhbqyt1f_xr | 2026-10-17 | chat-short | Q4_K_M | llama.cpp · commit 1268eda | 110 | 256 | 69.8 | 1047.0 | 105 | – | Measured by llm-speed@llmspeed |
| r_phvxm9dcak0 | 2026-07-01 | chat-short | – | llama.cpp · not reported | 0 | 256 | 69.5 | – | 327 | – | Community-submitted |
| r_-txe_hiq44n | 2026-07-01 | chat-short | – | llama.cpp · not reported | 0 | 254 | 66.5 | – | 158 | – | Community-submitted |
| r_vhbqyt1f_xr | 2026-10-17 | chat-long | Q4_K_M | llama.cpp · commit 1268eda | 3147 | 1024 | 64.1 | 3139.6 | 1002 | – | Measured by llm-speed@llmspeed |
| r_-txe_hiq44n | 2026-07-01 | chat-long | – | llama.cpp · not reported | 0 | 1020 | 62.3 | – | 1067 | – | Community-submitted |
| r_vhbqyt1f_xr | 2026-10-17 | agent-trace | Q4_K_M | llama.cpp · commit 1268eda | 3247 | 840 | 66.5 | 8161.5 | 253 | – | Measured by llm-speed@llmspeed |
| r_-txe_hiq44n | 2026-07-01 | agent-trace | – | llama.cpp · not reported | 3362 | 840 | 64.3 | 8394.1 | 259 | – | Community-submitted |
| r_vhbqyt1f_xr | 2026-10-17 | concurrent-decode | Q4_K_M | llama.cpp · commit 1268eda | 1024 | 256 | 68.5 | – | – | – | Measured by llm-speed@llmspeed |
| r_-txe_hiq44n | 2026-07-01 | concurrent-decode | – | llama.cpp · not reported | 1024 | 256 | 66.6 | – | – | – | Community-submitted |
Questions
Can an RTX 5090 run Qwen3-32B?
Yes: an RTX 5090 runs Qwen3 32B at 69.8 tok/s at Q4_K_M, measured on llama.cpp with a 110-token prompt in October 2026.
How many tokens per second does Qwen3-32B get on an RTX 5090?
Decode speed measured by llm-speed, by quantization and runtime: Q4_K_M on llama.cpp: 69.8 tok/s. Based on 1 run measured by llm-speed; single-request chat-short unless noted. 2 community-submitted runs are listed on the page, labelled, and not counted here.
What is the fastest quant of Qwen3-32B that fits on an RTX 5090?
Q4_K_M on llama.cpp was the fastest measured setting at 69.8 tok/s; it fit because it ran. Other quants are listed only where a run measured them.
How much VRAM does Qwen3-32B use on an RTX 5090?
Memory use was not recorded for these runs. The VRAM calculator gives an estimate, not a measurement.