Only with CPU offload
Qwen3.5 122B-A10B at Q4_K_M needs 71.1 GB but only 46.4 GB is addressable, so about 35% of the layers would stream from system RAM at roughly 60 GB/s. Expect around 6.3 tokens per second — usable for batch work, painful for chat.
The 96–128 GB unified-memory model: 122B of knowledge at 10B-active speed.
Hardware requirements What hardware do I need to run Qwen3.5 122B-A10B? →At Q4_K_M with an 8K context: a 141 GB GPU (H200 SXM) or a 128 GB Mac. Minimum, recommended, high-quality and long-context tiers, with all 114 devices checked at every quantisation.
Fits instead: Qwen3.5 9B (6.3 GB) · Qwen3.5 4B (3.5 GB)
The VRAM budget
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 123.7 GB | 124.5 GB | — | ~2.1 | −0.1% ppl | 78.1 GB over |
| Q6_K | 95.5 GB | 96.2 GB | — | ~3.3 | −0.4% ppl | 49.8 GB over |
| Q5_K_M | 82.5 GB | 83.3 GB | — | ~4.4 | −0.8% ppl | 36.9 GB over |
| Q4_K_M | 70.3 GB | 71.1 GB | — | ~6.3 | −1.9% ppl | 24.7 GB over |
| Q3_K_M | 56.9 GB | 57.7 GB | — | ~12 | −5.4% ppl | 11.3 GB over |
| Q2_K | 48.7 GB | 49.5 GB | — | ~28 | −15% ppl | 3.1 GB over |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. Only 12 of its 48 blocks keep a per-token KV cache; the rest are linear-attention, Mamba or convolution blocks with a fixed-size state.
How to run it
$ llama-server \
-hf Qwen/Qwen3.5-122B-A10B:Q4_K_M \
-c 8192 -ngl 31
The engine underneath most of the others. Every knob is exposed. More on llama.cpp.
Related models on a Radeon Pro W7900
| Model | Total | Tok/s est. | |
|---|---|---|---|
| Qwen3.5 397B-A17B 403B MoE · same family |
Q4_K_M · 227.4 GB | — | Won't fit |
| Qwen3.5 9B 9.65B · same family |
Q4_K_M · 6.3 GB | 96 | Runs great |
| Qwen3.5 4B 4.66B · same family |
Q4_K_M · 3.5 GB | 200 | Runs great |
| Qwen3.5 2B 2.27B · same family |
Q4_K_M · 2.0 GB | 410 | Runs great |
| Nemotron 3 Super 120B-A12B 124B MoE · similar size |
Q4_K_M · 70.4 GB | ~5.3 | CPU offload |
| Ling 3.0 Flash 124B-A5B 124B MoE · similar size |
Q4_K_M · 70.4 GB | ~13 | CPU offload |
| Mistral Small 4 119B-A6B 119B MoE · similar size |
Q4_K_M · 67.7 GB | ~10 | CPU offload |
Qwen3.5 122B-A10B on other hardware
| Device | Total | Tok/s est. | |
|---|---|---|---|
| GeForce RTX 4060 Laptop 7.0 GB usable · 256 GB/s |
71.1 GB | — | Won't fit |
| GeForce RTX 3060 12 GB 10.6 GB usable · 360 GB/s |
71.1 GB | — | Won't fit |
| M4 · 16 GB 10.7 GB usable · 120 GB/s |
71.1 GB | — | Won't fit |
| GeForce RTX 4090 22.4 GB usable · 1008 GB/s |
71.1 GB | ~3.5 | CPU offload |
| Radeon RX 7900 XTX 22.4 GB usable · 960 GB/s |
71.1 GB | ~3.5 | CPU offload |
| M3 Max · 64 GB 48.0 GB usable · 400 GB/s |
71.1 GB | — | Won't fit |
| DGX Spark · 128 GB 96.0 GB usable · 273 GB/s |
71.1 GB | 13 | Runs great |
| H200 SXM 139.4 GB usable · 4800 GB/s |
71.1 GB | 199 | Runs great |