vLLM on the NVIDIA RTX 5090
On the NVIDIA RTX 5090, vLLM runs Qwen3.8 27B at 4-bit at a median 77.2 tokens per second for generation, from 1 plain run (256K context).
Speed by model on the NVIDIA RTX 5090
Plain runs only: one device, one request, no speculative decoding, the whole model in memory. Each row is one model size at one bit class, with the other engines' plain runs of the same model on this card beside it. Context length and build still differ between runs, and each side shows its own.
| Model | vLLM | Other engines |
|---|---|---|
| Qwen3.8 27B 4-bit | 77.2 t/sone report, 256K context | no other engine reported |
| Qwen3.8 27B 5-bit | 59.0 t/sone report, 64K context | no other engine reported |
11 vLLM reports on the NVIDIA RTX 5090
Newest first. A run on several cards, with speculative decoding, with part of the model in system RAM or with batched requests says so, since each describes a different setup.
- Qwen3.8 27B · NVFP4304.0 t/s(speculative decoding)
- DeepSeek V4.1 Flash 552B · 16B active · MXFP497.2 t/s(8 cards, part in system RAM)
- Qwen3.8 27B · FP881.3 t/s(2 cards, speculative decoding)
- Qwen3.8 27B · NVFP477.2 t/s
- DeepSeek V4 Flash 284B · 13B active15.0 t/s(speculative decoding, part in system RAM)
- DeepSeek V4 Flash 284B · 13B active13.8 t/s(speculative decoding, part in system RAM)
- Qwen3.8 27B · 5.5-bit59.0 t/s
- Qwen3.8 27B · NVFP4120.0 t/s(speculative decoding, 3 requests at once)
- Gemma 4 26B · 4B active · AWQ-4bit578.0 t/s(speculative decoding)
- Qwen3.6 27B · NVFP473.6 t/s(speculative decoding)
- Qwen3.6 27B · NVFP480.0 t/s(speculative decoding)
Frequently asked
How fast is vLLM on the NVIDIA RTX 5090?
On the NVIDIA RTX 5090, vLLM runs Qwen3.8 27B at 4-bit at a median 77.2 tokens per second for generation, from 1 plain run (256K context).