Models · chat and writing
Qwen3 0.6B: VRAM requirements and GPUs that run it
Tiny. Runs on anything, including phones. Good for autocomplete and classification. At Q4 with 8K context it needs about 0.6 GB, so you want a GPU with at least 4 GB of VRAM or a Mac with 8 GB of unified memory.
Memory needed by quantization
| Quant | Weights | Total at 8K | Total at 32K |
|---|---|---|---|
| Q4_K_M | 0.4 GB | 0.6 GB | 1.3 GB |
| Q5_K_M | 0.5 GB | 0.7 GB | 1.3 GB |
| Q8_0 | 0.7 GB | 0.9 GB | 1.6 GB |
| FP16 | 1.3 GB | 1.5 GB | 2.2 GB |
Totals include about 10% runtime overhead and the KV cache. Its conversation memory (KV cache) is unusually small, so going from 8K to 32K context adds only about 0.7 GB. Most people should download Q4_K_M; Q5 or Q8 only make sense if you have spare memory. Read which quantization to pick.
Which GPUs can run Qwen3 0.6B?
40 of the 40 cards we track run it fully in VRAM at 8K context. The cheapest entry point is a 4 GB card such as the GTX 1650, and the fastest is the RTX 5090 at about 1830 tokens per second.
| GPU | VRAM | Status | Quant | Speed |
|---|---|---|---|---|
| RTX 5090 | 32 GB | Runs well | Q8_0 | ~1830 tok/s |
| RTX 4090 | 24 GB | Runs well | Q8_0 | ~1030 tok/s |
| RTX 5080 | 16 GB | Runs well | Q8_0 | ~980 tok/s |
| RX 7900 XTX | 24 GB | Runs well | Q8_0 | ~980 tok/s |
| RTX 3090 / 3090 Ti | 24 GB | Runs well | Q8_0 | ~960 tok/s |
| RTX 5070 Ti | 16 GB | Runs well | Q8_0 | ~920 tok/s |
| RX 7900 XT | 20 GB | Runs well | Q8_0 | ~820 tok/s |
| RTX 3080 | 10 GB | Runs well | Q8_0 | ~780 tok/s |
| RTX 4080 / 4080 Super | 16 GB | Runs well | Q8_0 | ~750 tok/s |
| RTX 5070 | 12 GB | Runs well | Q8_0 | ~690 tok/s |
| RTX 4070 Ti Super | 16 GB | Runs well | Q8_0 | ~690 tok/s |
| RX 9070 XT | 16 GB | Runs well | Q8_0 | ~650 tok/s |
| RX 7800 XT | 16 GB | Runs well | Q8_0 | ~640 tok/s |
| RTX 4090 Laptop | 16 GB | Runs well | Q8_0 | ~590 tok/s |
| Arc A770 | 16 GB | Runs well | Q8_0 | ~570 tok/s |
| Arc A750 | 8 GB | Runs well | Q8_0 | ~520 tok/s |
| RX 6800 XT | 16 GB | Runs well | Q8_0 | ~520 tok/s |
| RTX 4070 / 4070 Super | 12 GB | Runs well | Q8_0 | ~520 tok/s |
| GTX 1080 Ti | 11 GB | Runs well | Q8_0 | ~490 tok/s |
| Arc B580 | 12 GB | Runs well | Q8_0 | ~470 tok/s |
| RTX 5060 Ti | 8 GB | Runs well | Q8_0 | ~460 tok/s |
| RTX 5060 | 8 GB | Runs well | Q8_0 | ~460 tok/s |
| RTX 3070 / 3070 Ti | 8 GB | Runs well | Q8_0 | ~460 tok/s |
| RTX 3060 Ti | 8 GB | Runs well | Q8_0 | ~460 tok/s |
| RTX 5060 Ti | 16 GB | Runs well | Q8_0 | ~460 tok/s |
| RTX 4080 Laptop | 12 GB | Runs well | Q8_0 | ~440 tok/s |
| RX 7700 XT | 12 GB | Runs well | Q8_0 | ~440 tok/s |
| RX 6700 XT | 12 GB | Runs well | Q8_0 | ~390 tok/s |
| RTX 3060 | 12 GB | Runs well | Q8_0 | ~370 tok/s |
| GTX 1660 / 1660 Super | 6 GB | Runs well | Q8_0 | ~340 tok/s |
| RTX 3060 Laptop | 6 GB | Runs well | Q8_0 | ~340 tok/s |
| RTX 4060 Ti | 8 GB | Runs well | Q8_0 | ~290 tok/s |
| RX 7600 | 8 GB | Runs well | Q8_0 | ~290 tok/s |
| RTX 4060 Ti | 16 GB | Runs well | Q8_0 | ~290 tok/s |
| RTX 4060 | 8 GB | Runs well | Q8_0 | ~280 tok/s |
| RTX 4070 Laptop | 8 GB | Runs well | Q8_0 | ~260 tok/s |
| RTX 4060 Laptop | 8 GB | Runs well | Q8_0 | ~260 tok/s |
| RTX 3050 | 8 GB | Runs well | Q8_0 | ~230 tok/s |
| RTX 4050 Laptop | 6 GB | Runs well | Q8_0 | ~200 tok/s |
| GTX 1650 | 4 GB | Runs well | Q8_0 | ~130 tok/s |
Running it on a Mac
Apple Silicon Macs can give a model about 72% of their unified memory. Qwen3 0.6B fits from 8 GB (Apple M1, about 69 tokens per second). With a faster chip like the Apple M3 Ultra (96 GB) expect around 840 tokens per second. Compare the options in our Mac vs NVIDIA guide.
Compare Qwen3 0.6B
Questions
How much VRAM does Qwen3 0.6B need?
About 0.6 GB at Q4_K_M with 8K context, 0.9 GB at Q8_0, and 1.5 GB at full FP16. In practice that means a card with at least 4 GB of VRAM for Q4.
Can Qwen3 0.6B run on an 8 GB GPU?
Yes, at Q8_0, using about 0.9 GB.
Can I run Qwen3 0.6B on a Mac?
Yes. The smallest configuration that runs it fully is Apple M1 with 8 GB of unified memory, at about 69 tokens per second. More memory bandwidth (Pro, Max, Ultra chips) means faster replies.
Estimates based on model size, architecture and memory bandwidth. Real numbers vary with software, drivers and context length.