Can an RTX 3090 run Qwen3.8-27B?
Yes: an RTX 3090 runs Qwen3.8 27B at 38.8 tok/s at Q4_K_M, measured on llama.cpp with 8K context in October 2026.
5 signed runs on llm-speed. Single-request decode speed in tokens per second; the short-chat workload unless noted. 4 of 5 runs were measured by llm-speed (@llmspeed); the other one is 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 a37cc4610 | 38.8 | chat-short | 4 | Measured by llm-speed@llmspeed |
| Q4_K_M | llama.cpp | commit 8c2134a | 37.6 | chat-short | 1 | Community-submitted |
Fastest quant that fits: Q4_K_M on llama.cpp, 38.8 tok/s (measured by llm-speed; it fits because it ran on this hardware). Quants without a run are not listed.
VRAM: not recorded for these runs. The VRAM calculator gives an estimate (predicted, not measured).
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_w40_vv7f77w | 2026-10-22 | chat-short | Q4_K_M | llama.cpp · commit a37cc4610 | 154 | 256 | 38.8 | 479.1 | 321 | – | Measured by llm-speed@llmspeed |
| r_7250vqsziqj | 2026-10-22 | chat-short | Q4_K_M | llama.cpp · commit a37cc4610 | 154 | 256 | 38.7 | 380.6 | 405 | – | Measured by llm-speed@llmspeed |
| r_mmtzel_i367 | 2026-10-22 | chat-short | Q4_K_M | llama.cpp · commit a37cc4610 | 154 | 256 | 38.6 | 482.3 | 319 | – | Measured by llm-speed@llmspeed |
| r_gscy1f7ey_k | 2026-10-22 | chat-short | Q4_K_M | llama.cpp · commit a37cc4610 | 154 | 256 | 38.6 | 482.7 | 319 | – | Measured by llm-speed@llmspeed |
| r_ptct8o01o5l | 2026-10-14 | chat-short | Q4_K_M | llama.cpp · commit 8c2134a | 154 | 256 | 37.6 | 368.5 | 418 | – | Community-submitted |
| r_w40_vv7f77w | 2026-10-22 | chat-long | Q4_K_M | llama.cpp · commit a37cc4610 | 3224 | 1024 | 38.4 | 1237.5 | 2605 | – | Measured by llm-speed@llmspeed |
| r_7250vqsziqj | 2026-10-22 | chat-long | Q4_K_M | llama.cpp · commit a37cc4610 | 3224 | 1024 | 38.2 | 1128.0 | 2858 | – | Measured by llm-speed@llmspeed |
| r_mmtzel_i367 | 2026-10-22 | chat-long | Q4_K_M | llama.cpp · commit a37cc4610 | 3224 | 872 | 38.2 | 1222.7 | 2637 | – | Measured by llm-speed@llmspeed |
| r_gscy1f7ey_k | 2026-10-22 | chat-long | Q4_K_M | llama.cpp · commit a37cc4610 | 3224 | 1024 | 38.1 | 1208.6 | 2667 | – | Measured by llm-speed@llmspeed |
| r_ptct8o01o5l | 2026-10-14 | chat-long | Q4_K_M | llama.cpp · commit 8c2134a | 3224 | 1024 | 37.2 | 1103.5 | 2921 | – | Community-submitted |
| r_7250vqsziqj | 2026-10-22 | agent-trace | Q4_K_M | llama.cpp · commit a37cc4610 | 3496 | 840 | 38.0 | 2332.0 | 907 | – | Measured by llm-speed@llmspeed |
| r_w40_vv7f77w | 2026-10-22 | agent-trace | Q4_K_M | llama.cpp · commit a37cc4610 | 3496 | 840 | 37.9 | 2399.7 | 865 | – | Measured by llm-speed@llmspeed |
| r_gscy1f7ey_k | 2026-10-22 | agent-trace | Q4_K_M | llama.cpp · commit a37cc4610 | 3496 | 840 | 37.8 | 2408.1 | 880 | – | Measured by llm-speed@llmspeed |
| r_mmtzel_i367 | 2026-10-22 | agent-trace | Q4_K_M | llama.cpp · commit a37cc4610 | 3496 | 840 | 37.8 | 2401.3 | 862 | – | Measured by llm-speed@llmspeed |
| r_ptct8o01o5l | 2026-10-14 | agent-trace | Q4_K_M | llama.cpp · commit 8c2134a | 3496 | 840 | 36.8 | 2190.5 | 968 | – | Community-submitted |
| r_w40_vv7f77w | 2026-10-22 | concurrent-decode | Q4_K_M | llama.cpp · commit a37cc4610 | 1024 | 256 | 38.5 | – | – | – | Measured by llm-speed@llmspeed |
| r_7250vqsziqj | 2026-10-22 | concurrent-decode | Q4_K_M | llama.cpp · commit a37cc4610 | 1024 | 256 | 38.5 | – | – | – | Measured by llm-speed@llmspeed |
| r_mmtzel_i367 | 2026-10-22 | concurrent-decode | Q4_K_M | llama.cpp · commit a37cc4610 | 1024 | 256 | 38.5 | – | – | – | Measured by llm-speed@llmspeed |
| r_gscy1f7ey_k | 2026-10-22 | concurrent-decode | Q4_K_M | llama.cpp · commit a37cc4610 | 1024 | 256 | 38.4 | – | – | – | Measured by llm-speed@llmspeed |
| r_ptct8o01o5l | 2026-10-14 | concurrent-decode | Q4_K_M | llama.cpp · commit 8c2134a | 1024 | 256 | 37.5 | – | – | – | Community-submitted |
Questions
Can an RTX 3090 run Qwen3.8-27B?
Yes: an RTX 3090 runs Qwen3.8 27B at 38.8 tok/s at Q4_K_M, measured on llama.cpp with 8K context in October 2026.
How many tokens per second does Qwen3.8-27B get on an RTX 3090?
Decode speed measured by llm-speed, by quantization and runtime: Q4_K_M on llama.cpp: 38.8 tok/s. Based on 4 runs measured by llm-speed; single-request chat-short unless noted. 1 community-submitted run is listed on the page, labelled, and not counted here.
What is the fastest quant of Qwen3.8-27B that fits on an RTX 3090?
Q4_K_M on llama.cpp was the fastest measured setting at 38.8 tok/s; it fit because it ran. Other quants are listed only where a run measured them.
How much VRAM does Qwen3.8-27B use on an RTX 3090?
Memory use was not recorded for these runs. The VRAM calculator gives an estimate, not a measurement.