Qwen 3 8B VRAM Requirements
Qwen 3 8B needs about 6.6 GB of VRAM at Q4_K_M with an 8K context: 4.6 GB for weights, 1.1 GB for the KV cache and 0.9 GB of runtime overhead. At Q8_0 it needs 10.1 GB and at full FP16 precision 17.2 GB. The smallest GPU that runs it fully is the RTX 3050 8GB.
6.6 GB VRAM
Qwen 3 8B at Q4_K_M with 8K context and the KV cache in FP16.
- Weights 4.6 GB
- KV cache 1.1 GB
- Runtime overhead 0.9 GB
Which GPUs can run it
- RTX 3060 12GBFits~41 tok/s
- RTX 4060 Ti 16GBFits~33 tok/s
- RTX 3090 24GBFits~105 tok/s
- Radeon RX 7900 XTX 24GBFits~92 tok/s
- RTX 4090 24GBFits~113 tok/s
- RTX 5090 32GBFits~195 tok/s
- RTX 6000 Ada 48GBFits~108 tok/s
- A100 80GBFits~219 tok/s
- H100 80GB SXMFits~343 tok/s
- Ryzen AI Max+ 395 128GBFits~27 tok/s
- Apple M4 Max 128GBFits~62 tok/s
- Apple M5 Max 128GBFits~70 tok/s
- RTX PRO 6000 Blackwell 96GBFits~195 tok/s
- NVIDIA DGX Spark 128GBFits~29 tok/s
- Apple M3 Ultra 512GBFits~92 tok/s
Qwen 3 8B VRAM by quantization
Total memory with an FP16 KV cache. "Smallest tier" is the smallest standard VRAM size with a 6% safety margin at 8K.
| Quantization | Bits/weight | Weights | Total at 8K | Total at 32K | Smallest tier |
|---|---|---|---|---|---|
| Q4_K_M | 4.85 | 4.6 GB | 6.6 GB | 10.0 GB | 8 GB |
| Q5_K_M | 5.69 | 5.4 GB | 7.4 GB | 10.8 GB | 8 GB |
| Q6_K | 6.59 | 6.3 GB | 8.3 GB | 11.6 GB | 10 GB |
| Q8_0 | 8.50 | 8.1 GB | 10.1 GB | 13.5 GB | 12 GB |
| FP16 / BF16 | 16.00 | 15.3 GB | 17.2 GB | 20.6 GB | 20 GB |
| Q3_K_M | 3.91 | 3.7 GB | 5.7 GB | 9.1 GB | 8 GB |
| Q2_K | 3.00 | 2.9 GB | 4.8 GB | 8.2 GB | 6 GB |
VRAM by context length
At Q4_K_M. The KV cache grows linearly with context: 0.14 GB per 1,000 tokens in FP16.
| Context | KV cache (FP16) | KV cache (Q8_0) | Total, FP16 cache | Total, Q8_0 cache |
|---|---|---|---|---|
| 4K tokens | 0.6 GB | 0.3 GB | 6.0 GB | 5.8 GB |
| 8K tokens | 1.1 GB | 0.6 GB | 6.6 GB | 6.1 GB |
| 32K tokens | 4.5 GB | 2.4 GB | 10.0 GB | 7.9 GB |
| 40K tokens | 5.6 GB | 3.0 GB | 11.1 GB | 8.5 GB |
GPUs that can run Qwen 3 8B
At Q4_K_M with 8K context. "Best quant" is the highest-quality format that fits fully; "max context" is at Q4_K_M.
| GPU | Usable memory | Status | Best quant | Max context | Speed |
|---|---|---|---|---|---|
| RTX 3050 8GB | 8.0 GB | Fits | Q5_K_M | 14K | ~26 tok/s |
| RTX 4060 8GB | 8.0 GB | Fits | Q5_K_M | 14K | ~31 tok/s |
| RTX 4060 Ti 8GB | 8.0 GB | Fits | Q5_K_M | 14K | ~33 tok/s |
| Radeon RX 7600 8GB | 8.0 GB | Fits | Q5_K_M | 14K | ~28 tok/s |
| RTX 3060 Ti 8GB | 8.0 GB | Fits | Q5_K_M | 14K | ~51 tok/s |
| RTX 3070 8GB | 8.0 GB | Fits | Q5_K_M | 14K | ~51 tok/s |
| RTX 5060 8GB | 8.0 GB | Fits | Q5_K_M | 14K | ~51 tok/s |
| RTX 5060 Ti 8GB | 8.0 GB | Fits | Q5_K_M | 14K | ~51 tok/s |
| RTX 3070 Ti 8GB | 8.0 GB | Fits | Q5_K_M | 14K | ~69 tok/s |
| Intel Arc B570 10GB | 10.0 GB | Fits | Q6_K | 27K | ~34 tok/s |
| RTX 3080 10GB | 10.0 GB | Fits | Q6_K | 27K | ~86 tok/s |
| Apple M4 16GB | 10.7 GB | Fits | Q6_K | 32K | ~14 tok/s |
| RTX 2080 Ti 11GB | 11.0 GB | Fits | Q8_0 | 34K | ~70 tok/s |
| RTX 2060 12GB | 12.0 GB | Fits | Q8_0 | 40K | ~39 tok/s |
About Qwen 3 8B
Supports switching between thinking and non-thinking modes. With 36 layers its cache is 144 KB per token in FP16, about 12% more than Llama 3.1 8B. The native window is 32K (40,960 in config.json), extendable to 128K with YaRN.
| Spec | Value |
|---|---|
| Parameters | 8.19B |
| Layers | 36 |
| Hidden size | 4096 |
| Attention heads / KV heads | 32 / 8 |
| Head dimension | 128 |
| Vocabulary | 151,936 |
| Max context (used here) | 40K tokens |
| KV cache per token (FP16) | 144 KB |
| Released | 2025-04 |
| License | Apache 2.0 |
Qwen 3 8B VRAM: common questions
Can Qwen 3 8B run on an 8GB GPU?
Yes at Q4_K_M with an 8K context (about 6.6 GB). Thinking mode can use long traces, so a Q8_0 KV cache helps on 8 GB cards.
How much VRAM does Qwen 3 8B need?
About 6.6 GB at Q4_K_M with an 8K context, 10.1 GB at Q8_0 and 17.2 GB at FP16. Longer contexts add 0.14 GB per 1,000 tokens with an FP16 cache.
What is the cheapest GPU that can run Qwen 3 8B?
The smallest listed device that fits it fully at Q4_K_M with 8K context is the RTX 3050 8GB (8.0 GB usable).