Qwen3-VL-8B-Instruct on RTX 5090: Runs comfortably (Sep 2026)
Conditional memory judgement for one configuration. Change context, quant, or engine and the conclusion can change. Configuration used on this page: BF16 safetensors, 8,192 cached tokens, 1 sequence, engine vllm. Status: Likely fits. Software: Not verified..
Verdict on GeForce RTX 5090 32GB (desktop)
Yes, run Q8_0. The calculator puts it at 9.4 GiB–10.0 GiB at 8,192 tokens, inside 30.0 GiB of usable VRAM (after the driver reserve).
With Q8_0, the longest context that still counts as a fit is about 124,037 tokens of a 262,144-token native window, one sequence, fp16 KV cache.
| GGUF quant | File weights | KV cache | Estimated total | Verdict |
|---|---|---|---|---|
| Q8_0 (recommended) | 8.1 GiB | 1.1 GiB | 9.4 GiB–10.0 GiB | Comfortable |
| Q6_K | 6.3 GiB | 1.1 GiB | 7.5 GiB–8.1 GiB | Comfortable |
| Q5_K_M | 5.4 GiB | 1.1 GiB | 6.7 GiB–7.3 GiB | Comfortable |
| Q4_K_M | 4.7 GiB | 1.1 GiB | 5.9 GiB–6.5 GiB | Comfortable |
Engine llama.cpp, one sequence, fp16 KV. File weights are the published GGUF sizes in the catalog. Totals come from the same calculator that powers the tool below; they are estimates, not a measurement of this setup.
Fixed page configuration
Qwen3-VL-8B-Instruct BF16 safetensors on the catalog RTX 5090 32 GB SKU. The page uses that published artifact and the calculator; it is not a new measurement.
- ModelQwen3-VL-8B-Instruct
- HardwareGeForce RTX 5090 32GB (desktop)
- QuantizationBF16 safetensors
- Context length (cached tokens)8,192
- Engine
vllm - Concurrent sequences1
- KV cache typefp16
Technical identifiers and share data
- Model
qwen3-vl-8b-instruct - Hardware
nvidia-geforce-rtx-5090-32gb-desktop - Quantization
bf16 - Share fragment for this identity
v2.eyJzY2hlbWFWZXJzaW9uIjoyLCJjYWxjdWxhdG9yVmVyc2lvbiI6IjAuMy4wIiwiZGF0YXNldFZlcnNpb24iOiIyMDI2LTEwLTAyLjEiLCJjb25maWciOnsibW9kZWxJZCI6InF3ZW4zLXZsLThiLWluc3RydWN0IiwiaGFyZHdhcmVJZCI6Im52aWRpYS1nZWZvcmNlLXJ0eC01MDkwLTMyZ2ItZGVza3RvcCIsImRldmljZUNvdW50IjoxLCJxdWFudGl6YXRpb25JZCI6ImJmMTYiLCJjb250ZXh0VG9rZW5zIjo4MTkyLCJjb25jdXJyZW50U2VxdWVuY2VzIjoxLCJlbmdpbmUiOiJ2bGxtIiwia3ZUeXBlIjoiZnAxNiJ9fQ
Memory breakdown
- Weight artifact
- 16.33 GiB (17,534,247,392 bytes)
- Raw KV cache
- 1.13 GiB (1,207,959,552 bytes)
- Engine extras / workspace
- 0.63 GiB–2.53 GiB
- Total estimated demand
- 18.08 GiB–19.99 GiB
- Net calculator budget
- 30.00 GiB
- OS / display reserve
- 2.00 GiB
- Remaining vs net budget
- 10.01 GiB–11.92 GiB
19.99 GiB high estimate vs 30.00 GiB net
Enough memory does not mean this engine, quant, and OS will run. Support does not mean it will be fast.
Other published files on this device
Same model, device and context (8,192 tokens, 1 sequence, FP16 KV). Each row is one published weight-file set with its usual engine, sorted by memory.
| Quantization | Engine | Estimated memory | Fit |
|---|---|---|---|
| BF16 safetensors This page | vllm | 18.08 GiB–19.99 GiB | Runs comfortably |
| GGUF Q8_0 (Qwen) | llama.cpp | 9.36 GiB–9.98 GiB | Runs comfortably |
| GGUF Q6_K (Unsloth) | llama.cpp | 7.51 GiB–8.09 GiB | Runs comfortably |
| GGUF Q5_K_M (Unsloth) | llama.cpp | 6.70 GiB–7.26 GiB | Runs comfortably |
| GGUF Q4_K_M (Qwen) | llama.cpp | 5.93 GiB–6.48 GiB | Runs comfortably |
Assumptions that affect this result
- Requested context is within the recorded native window of 262144 tokens.
- Full-resident weight estimates use total parameters, including inactive MoE experts.
- Weight bytes come from the selected weight files only, not from every file in the repository.
- Download file size is not peak GPU memory. Runtime layout, allocator padding, and KV are extra.
- Counted files: model-00001-of-00004.safetensors, model-00002-of-00004.safetensors, model-00003-of-00004.safetensors, model-00004-of-00004.safetensors.
- This checkpoint has a vision component. Text-only inference does not include vision encoder or projector bytes.
- Context budget is the number of cached tokens per sequence (prompt plus reserved generation).
- Concurrent sequences share one model replica; weights are not multiplied by sequence count.
Related pages
Sources
- Qwen3-VL-8B-Instruct model card
- config.json — revision 0c351dd01ed87e9c1b53cbc748cba10e6187ff3b
- safetensors index — metadata.total_size=17534247392
- Artifact repo Qwen/Qwen3-VL-8B-Instruct-GGUF
- Artifact repo unsloth/Qwen3-VL-8B-Instruct-GGUF
- NVIDIA GeForce RTX 5090 product page — Standard Memory Config 32 GB GDDR7
Calculator
This combo page keeps the route model and device. Customize to change them on the main calculator; the share then keeps that identity.
Loading catalog snapshot…
Search or pick a catalog checkpoint to see memory demand.