Advisor / Models / Qwen3-VL 8B

Qwen3-VL 8B: VRAM requirements & speed

Qwen3-VL 8B (8.77B dense, apache-2.0). Native context 256k, 1M with YaRN. Smallest GGUF: 5.03 GB.

5.03 GBsmallest quant file
25hardware pages that fit it fully
1run with offload
0measured data points

Memory needed by quantization and context

GGUF file size plus f16 KV cache (and fixed overhead of ~1 GB per GPU not shown). Source: unsloth/Qwen3-VL-8B-Instruct-GGUF.

QuantFile+8k ctx+32k ctx+128k ctx+256k ctx+1M ctx
Q4_K_M5.03 GB6.2 GB9.9 GB24.4 GB43.7 GB152.5 GB
Q5_K_M5.85 GB7.1 GB10.7 GB25.2 GB44.5 GB153.3 GB
Q6_K6.73 GB7.9 GB11.6 GB26.1 GB45.4 GB154.2 GB
Q8_08.71 GB9.9 GB13.5 GB28.0 GB47.4 GB156.2 GB

KV cache size by context length

Computed from the model's attention layout.

Contextf16q8_0q4_0
8k1.21 GB0.64 GB0.34 GB
32k4.83 GB2.57 GB1.36 GB
128k19.33 GB10.27 GB5.44 GB
256k38.65 GB20.54 GB10.87 GB
1M YaRN147.46 GB78.34 GB41.47 GB

Which hardware runs Qwen3-VL 8B? · 16k context

HardwareQuantFitMemoryTok/sSpeed source
RTX 4090 24GBQ8_0fits fully11.1 GB102scaled from measured
RTX 3090 24GBQ8_0fits fully11.1 GB91scaled from measured
Mac Studio M5 Ultra 256GBQ8_0fits fully11.1 GB87estimated
Mac Studio M5 Ultra 512GB (ships late Oct 2026)Q8_0fits fully11.1 GB87estimated
RTX 5070 12GBQ6_Kfits fully9.1 GB74scaled from measured
2x RTX 3090 (48GB total)Q8_0fits fully11.1 GB74estimated
RTX PRO 6000 96GB (bonus page)Q8_0fits fully11.1 GB66scaled from measured
RTX 4070 12GBQ6_Kfits fully9.1 GB62scaled from measured
RTX 5090 32GBQ8_0fits fully11.1 GB43scaled from measured
RTX 3060 12GB (32GB RAM)Q6_Kfits fully9.1 GB42scaled from measured
Mac Studio M5 Max 128GBQ8_0fits fully11.1 GB41scaled from measured
MacBook Pro 14/16 M5 Max 128GBQ8_0fits fully11.1 GB41scaled from measured
Mac Studio (2025) M3 Ultra 96GBQ8_0fits fully11.1 GB37scaled from measured
RTX 5080 16GBQ8_0fits fully11.1 GB34scaled from measured
16GB VRAM class (RTX 4060 Ti 16GB, 32GB RAM)Q8_0fits fully11.1 GB33scaled from measured
RTX 5070 Ti 16GBQ8_0fits fully11.1 GB32scaled from measured
8GB VRAM class (RTX 4060 8GB, 32GB RAM)Q4_K_Mpartial offload (slow)7.4 GB31estimated
MacBook Pro 14/16 (2024) M4 Max 128GBQ8_0fits fully11.1 GB30scaled from measured
AMD RX 7900 XTX 24GBQ8_0fits fully11.1 GB28scaled from measured
MacBook Pro 14/16 M5 Pro 48GBQ8_0fits fully11.1 GB24scaled from measured
NVIDIA DGX Spark 128GBQ8_0fits fully11.1 GB19scaled from measured
Mac mini M6 32GBQ6_Kfits fully9.1 GB18scaled from measured
Mac mini (2024) M4 Pro 64GBQ8_0fits fully11.1 GB18scaled from measured
MacBook Air M5 32GBQ5_K_Mfits fully8.3 GB17scaled from measured
Mac mini / iMac M4 16GBQ4_K_Mfits fully7.4 GB16scaled from measured
AMD Ryzen AI Max+ 395 (Strix Halo) 128GBQ6_Kfits fully9.1 GB16scaled from measured

Discrete GPUs are modelled with 32 GB of system RAM available for offload. Fits (at least with offload) on every listed configuration.

Links

Model card · GGUF files · Open in the advisor · last verified 2026-10-14

FAQ

How much VRAM does Qwen3-VL 8B need?

The smallest quant (Q4_K_M) is 5.03 GB, so you need roughly 7 GB for an 8k context including KV cache and overhead. At 1M context the f16 KV cache alone is 147.5 GB.

Can I run Qwen3-VL 8B on a RTX 4090 24GB?

Yes: on RTX 4090 24GB it fits fully as Q8_0 at about 102 tok/s (scaled from measured). 25 of 26 hardware configurations fit it fully at 16k context.

What is the maximum context for Qwen3-VL 8B?

256k tokens natively, up to 1M with YaRN as documented on the model card. Memory, not the model, is usually the limit: see the KV cache table above.

All models: model index · by memory: 8 GB · 12 GB · 16 GB · 24 GB · 32 GB · 48 GB · 64 GB · 96 GB · 128 GB · 192 GB · 256 GB · 512 GB. Numbers use the same fit math as the advisor; sources and methods.