Qwen 3.5 122B-A10B (MoE)
Qwen 3.5 122B-A10B (MoE) needs roughly 61.0GB VRAM at Q4 quantization (244.0GB at FP16). 24 GPUs we track can run it fully in VRAM at 8k context.
Alibaba122B params10B active (MoE)256k contextApache 2.0Commercial use ok
VRAM at each quantization
Assumes 8k context. KV cache grows linearly with context length.
| Quant | Weights | KV cache | Total |
|---|---|---|---|
| FP16 | 244.0 GB | 2.15 GB | 275.7 GB |
| Q8 | 122.0 GB | 2.15 GB | 139.1 GB |
| Q6_K | 91.5 GB | 2.15 GB | 104.9 GB |
| Q5_K_M | 76.3 GB | 2.15 GB | 87.8 GB |
| Q4_K_M | 61.0 GB | 2.15 GB | 70.7 GB |
| Q3_K_M | 48.8 GB | 2.15 GB | 57.1 GB |
| Q2_K | 36.6 GB | 2.15 GB | 43.4 GB |
Benchmarks
GPUs that run Qwen 3.5 122B-A10B (MoE) natively (24)
- NVIDIA H100 80GBQ4_K_M · 737 t/s
- NVIDIA A100 80GBQ4_K_M · 448.6 t/s
- NVIDIA L40SQ2_K · 316.8 t/s
- NVIDIA RTX A6000Q2_K · 281.6 t/s
- NVIDIA RTX 6000 AdaQ2_K · 352 t/s
- AMD Instinct MI300XQ8 · 583 t/s
- Apple M4 Ultra (384GB)FP16 · 60.1 t/s
- Apple M4 Ultra (192GB)Q8 · 120.1 t/s
- Apple M4 Max (128GB)Q6_K · 80.1 t/s
- Apple M4 Max (96GB)Q5_K_M · 96.1 t/s
- Apple M4 Max (64GB)Q3_K_M · 150.2 t/s
- Apple M4 Max (48GB)Q2_K · 200.2 t/s
- Apple M4 Pro (48GB)Q2_K · 100.1 t/s
- Apple M3 Max (128GB)Q6_K · 58.7 t/s
- Apple M3 Max (96GB)Q5_K_M · 70.4 t/s
- Apple M3 Max (64GB)Q3_K_M · 110 t/s
- Apple M3 Max (48GB)Q2_K · 146.7 t/s
- Apple M2 Ultra (384GB)FP16 · 44 t/s
- Apple M2 Ultra (192GB)Q8 · 88 t/s
- Apple M2 Max (96GB)Q5_K_M · 70.4 t/s
- Apple M2 Max (64GB)Q3_K_M · 110 t/s
- Apple M1 Ultra (128GB)Q6_K · 117.3 t/s
- Apple M1 Ultra (64GB)Q3_K_M · 220 t/s
- Apple M1 Max (64GB)Q3_K_M · 110 t/s
Plus 7 GPUs that run it with CPU offload (slower)
- NVIDIA RTX 5090Q2_K · 149.3 t/s
- NVIDIA RTX 4090Q2_K · 84 t/s
- NVIDIA RTX 3090Q2_K · 78 t/s
- NVIDIA RTX 3090 TiQ2_K · 84 t/s
- NVIDIA A100 40GBQ3_K_M · 97.2 t/s
- AMD Radeon RX 7900 XTXQ2_K · 80 t/s
- AMD Radeon RX 7900 XTQ2_K · 66.7 t/s