Can RTX 4090 24GB run Qwen3-VL 8B?
Measured results for Qwen3-VL 8B (8.77B dense) on RTX 4090 24GB, plus what fits and the commands to run it.
Short answer
Yes. Qwen3-VL 8B Q8_0 fits fully on RTX 4090 24GB at 16k context, using 11.1 of 23.0 GB usable, and generates about 102.3 tokens/s in our own run. As the context fills it slows down: 80.5 tok/s with 16k tokens already in context (79% of the empty-context speed), measured. For Q8_0, our bandwidth-only estimate was ~64 tok/s against a measured median of ~102 (38% too pessimistic).
Measured generation speed by quantization
| Quant | Generation tok/s | Range | Prompt tok/s | Runs | Origin |
|---|---|---|---|---|---|
| Q4 class (median) | 158.5 | 158.5–158.5 | 10766 | 1 | includes our runs |
| Q8 class (median) | 102.3 | 102.3–102.3 | 10242 | 1 | includes our runs |
| Q4_K_M (our run, depth 0) | 158.5 | 3 repeats | 10766 | 3 | own measurement |
| Q8_0 (our run, depth 0) | 102.3 | 3 repeats | 10242 | 3 | own measurement |
Generation = tg128 (128 generated tokens, short context). Public rows are medians of community llama.cpp results (sources); our own runs use llama-bench b11658 on a rented GPU.
Speed vs context depth measured
Our own llama-bench runs on GeForce RTX 4090 (flash attention on, 3 repeats): generation and prompt speed with the given number of tokens already in the context.
| Quant | Context depth | Generation tok/s | vs depth 0 | Prompt tok/s |
|---|---|---|---|---|
| Q4_K_M | 0 | 158.5 | 100% | 10766 |
| Q4_K_M | 4k | 143.1 | 90% | 9686 |
| Q4_K_M | 16k | 111.6 | 70% | 6689 |
| Q4_K_M | 32k | 87.5 | 55% | 4899 |
| Q4_K_M | 64k | 60.4 | 38% | 3108 |
| Q4_K_M | 128k | 37.2 | 23% | 1770 |
| Q8_0 | 0 | 102.3 | 100% | 10242 |
| Q8_0 | 4k | 95.7 | 94% | 10121 |
| Q8_0 | 16k | 80.5 | 79% | 6872 |
Alternatives
Less memory that still fits Qwen3-VL 8B
Smaller memory usually means a cheaper card or Mac (we do not list prices).
- Mac mini / iMac M4 16GB: Q4_K_M, 11 GB usable, ~16 tok/s scaled from measured
- RTX 5070 12GB: Q6_K, 11 GB usable, ~94 tok/s scaled from measured
- RTX 4070 12GB: Q6_K, 11 GB usable, ~71 tok/s scaled from measured
Faster for Qwen3-VL 8B
- RTX 5090 32GB: Q8_0, 31 GB usable, ~167 tok/s measured
- RTX PRO 6000 96GB: Q8_0, 95 GB usable, ~163 tok/s measured
Other models on RTX 4090 24GB, similar size
- Gemma 4 E4B (hardware page): Q8_0, fits fully, ~96 tok/s (scaled from measured)
- Qwen3.5 9B: Q8_0, fits fully, ~95 tok/s (measured)
- Ornith 1.5 9B (hardware page): Q8_0, fits fully, ~95 tok/s (scaled from measured)
- DeepSeek R1 Distill Qwen 7B (hardware page): Q8_0, fits fully, ~97 tok/s (scaled from measured)
Memory and context limits
Q8_0 weights are 8.7 GB; the 16k f16 KV cache adds 2.42 GB. Largest context that fits in 23.0 GB usable:
| KV cache type | Max context |
|---|---|
| f16 | 64k |
| q8_0 | 128k |
| q4_0 | 256k |
Native limit 256k, 1M with YaRN. Full KV cache table for Qwen3-VL 8B.
Copy-paste commands
The advisor's pick for this hardware: Q8_0, 16k context.
llama-server -hf unsloth/Qwen3-VL-8B-Instruct-GGUF:Q8_0 -c 16384 -fa on --jinja -ngl 99
OLLAMA_CONTEXT_LENGTH=16384 ollama serve ollama run hf.co/unsloth/Qwen3-VL-8B-Instruct-GGUF:Q8_0
FAQ
How fast is Qwen3-VL 8B on RTX 4090 24GB?
About 102.3 tokens/s for generation (Q8_0 on llama.cpp CUDA, 128-token generation), 10242 tokens/s for prompt processing. Speed drops as the context fills; see the depth table.
How much memory does Qwen3-VL 8B need on RTX 4090 24GB?
The advisor's pick, Q8_0, is 8.7 GB of weights plus 2.42 GB of KV cache at 16k context (11.1 GB), against 23.0 GB usable.
Is the Qwen3-VL 8B speed on RTX 4090 24GB measured or estimated?
Measured: our own llama-bench run on a rented GeForce RTX 4090, including runs with 4k to 128k tokens already in context.
What can I use instead of RTX 4090 24GB for Qwen3-VL 8B?
With less memory, Mac mini / iMac M4 16GB (~16 tok/s), RTX 5070 12GB (~94 tok/s), RTX 4070 12GB (~71 tok/s) also fit it fully. Faster: RTX 5090 32GB (~167 tok/s), RTX PRO 6000 96GB (~163 tok/s).
What is the maximum context for Qwen3-VL 8B on RTX 4090 24GB?
With an f16 KV cache about 64k tokens, 128k with q8_0 and 256k with q4_0, bounded by the model's own limit.
Related
All models on RTX 4090 24GB · Qwen3-VL 8B on all hardware · VRAM tier · all measured pages · leaderboard
Compare: RTX 3090 24GB vs RTX 4090 24GB · RTX 4090 24GB vs RTX 5090 32GB · RTX 4090 24GB vs AMD RX 7900 XTX 24GB · RTX 5080 16GB vs RTX 4090 24GB