Can Qwen3 VL 8B Instruct run on the RTX 3090 Ti?
Yes. At 8K context the best fit is F16, using about 17.1 GB of the 22.8 GB usable on the RTX 3090 Ti. Estimated speed ceiling: ~62 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | 128K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|---|
| 2-bit (Q2_K) | 4.3 GB Fits | 4.9 GB Fits | 8.3 GB Fits | 21.8 GB Tight | 307 |
| 3-bit (Q3_K_M) | 5.2 GB Fits | 5.8 GB Fits | 9.2 GB Fits | 22.7 GB Tight | 235 |
| 4-bit (Q4_K_M) • | 5.9 GB Fits | 6.5 GB Fits | 9.9 GB Fits | 23.4 GB No | 200 |
| 5-bit (Q5_K_M) | 7.1 GB Fits | 7.6 GB Fits | 11.0 GB Fits | 24.5 GB No | 161 |
| 6-bit (Q6_K) | 7.9 GB Fits | 8.5 GB Fits | 11.9 GB Fits | 25.4 GB No | 140 |
| 8-bit (Q8_0) • | 9.4 GB Fits | 9.9 GB Fits | 13.3 GB Fits | 26.8 GB No | 116 |
| 16-bit (F16/BF16) • | 16.5 GB Fits | 17.1 GB Fits | 20.5 GB Fits | 34.0 GB No | 62 |
• real GGUF file size from Qwen/Qwen3-VL-8B-Instruct-GGUF. Speed is a bandwidth ceiling (1008 GB/s), not a benchmark. Methodology.
More
Qwen3 VL 8B Instruct on every GPU and Mac · Every model that runs on the RTX 3090 Ti · Other models that fit 24 GB VRAM · Try other settings in the calculator
Data: Hugging Face Hub (Qwen/Qwen3-VL-8B-Instruct), NVIDIA specs. Data as of 2026-11-17.