Qwen3.5 397B A17B needs ~318.4 GB VRAM. AMD Instinct MI350X 288GB has 288.0 GB. With Q4_K_M quantization, expect ~15 tok/s.
Operating mode
Interactive favors responsiveness, while light API and scale-out lean harder on serving readiness. The fit stays the same, but the recommendation lens changes.
Current mode
Balanced
Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.
Select quantization to explore
30.4 GB over capacity — needs offload or smaller quantization
Fit status
Very compromised (needs ~23.1 GB host RAM)
Decode
14.6 tok/s
TTFT
13221 ms
Safe context
6K
Memory
318.4 GB / 288.0 GB
Offload
10%
It fits through host-memory offload, and offload is the main reason performance drops.
CPU or host-memory offload is active
About 10% of the working set spills out of accelerator memory, which usually hurts latency and sustained decode throughput.
Very little memory headroom
You can run the model, but there is not much room left for longer context, bigger batches, extra apps, or future model updates.
Remove offload with more accelerator memory
Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.
Buy headroom, not only minimum fit
A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.
Increase host RAM if you keep offloading
This setup may need roughly 23.1 GB of extra host RAM just for the offloaded portion, before OS and other tools.
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | C | Runs with offload (needs ~5.9 GB host RAM) | 17.2 tok/s | 6147 ms | 6K |
| Coding | D | Very compromised (needs ~23.1 GB host RAM) | 14.6 tok/s | 13221 ms | 6K |
| Agentic Coding | F | Too heavy | 11.0 tok/s | 25626 ms | 6K |
| Reasoning | D | Very compromised (needs ~23.1 GB host RAM) | 14.6 tok/s | 15625 ms | 6K |
| RAG | F | Too heavy | 11.0 tok/s | 32032 ms | 6K |
Inference speed
Estimated decode speed (tokens/sec) for Qwen3.5 397B A17B at Q4_K_M across popular GPUs and Apple Silicon, including multi-GPU rigs, using the fastest local runtime per device. Fastest is RTX 5090 32GB at ~2 tok/s. Speed is memory-bandwidth bound, so cards that fit the whole model in VRAM run far faster than ones that offload to system RAM.
| GPU / Mac | Memory | Quant | Speed (tok/s) | Fits? |
|---|---|---|---|---|
 RTX 5090 32GB | 32 GB | Q4_K_M | 2.0 | Too big |
| RTX 4090 24GB | 24 GB | Q4_K_M | 2.0 | Too big |
| RTX 4080 Super 16GB | 16 GB | Q4_K_M | 2.0 | Too big |
| RTX 3090 24GB | 24 GB | Q4_K_M | 2.0 | Too big |
| RTX 4070 12GB | 12 GB | Q4_K_M | 2.0 | Too big |
| RTX 3060 12GB | 12 GB | Q4_K_M | 2.0 | Too big |
| RTX 4060 8GB | 8 GB | Q4_K_M | 2.0 | Too big |
RX 7900 XTX 24GB | 24 GB | Q4_K_M | 2.0 | Too big |
MacBook Pro M4 Max 128GB | 128 GB | Q4_K_M | 2.0 | Too big |
Mac Studio M3 Ultra 256GB | 256 GB | Q4_K_M | 2.0 | Too big |
Mac Studio M2 Ultra 128GB | 128 GB | Q4_K_M | 2.0 | Too big |
Mac Studio M1 Ultra 128GB | 128 GB | Q4_K_M | 2.0 | Too big |
MacBook Pro M4 Max 64GB | 64 GB | Q4_K_M | 2.0 | Too big |
MacBook Pro M3 Max 64GB | 64 GB | Q4_K_M | 2.0 | Too big |
MacBook Pro M1 Max 64GB | 64 GB | Q4_K_M | 2.0 | Too big |
MacBook Pro M4 Pro 48GB | 48 GB | Q4_K_M | 2.0 | Too big |
| 2× RTX 4090 24GB | 48 GB | Q4_K_M | 2.0 | Too big |
| 2× RTX 3090 24GB | 48 GB | Q4_K_M | 2.0 | Too big |
2× RX 7900 XTX 24GB | 48 GB | Q4_K_M | 2.0 | Too big |
| 4× RTX 3060 12GB | 48 GB | Q4_K_M | 2.0 | Too big |
Estimates for single-stream decoding at Q4_K_M; real tokens/sec varies with prompt length, context, batch size, and runtime build. Prompt processing (prefill) is faster than the decode figures shown here.
How Qwen3.5 397B A17B (397B params) fits at each quantization level on AMD Instinct MI350X 288GB (288.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 154.8 GB | Low | C48 |
Q3_K_S | 3 | 194.5 GB | Low | C48 |
NVFP4Best for your GPU | 4 | 222.3 GB | Medium | C48 |
Q4_K_M | 4 | 242.2 GB | Medium | F0 |
Q5_K_M | 5 | 285.8 GB | High | F0 |
Q6_K | 6 | 325.5 GB | High | F0 |
Q8_0 | 8 | 424.8 GB | Very High | F0 |
F16 | 16 | 813.8 GB | Maximum | F0 |
Copy-paste commands to run Qwen3.5 397B A17B on your machine.
Run
lms load hf-unsloth--qwen3-5-397b-a17b-gguf && lms server startYes, AMD Instinct MI350X 288GB can run Qwen3.5 397B A17B with a D grade (Very compromised (needs ~23.1 GB host RAM)). Expected decode speed: 14.6 tok/s.
Qwen3.5 397B A17B (397B parameters) requires approximately 318.4 GB of memory with Q4_K_M quantization.
The recommended quantization for Qwen3.5 397B A17B is Q4_K_M, which balances quality and memory efficiency.
On AMD Instinct MI350X 288GB, Qwen3.5 397B A17B achieves approximately 14.6 tokens per second decode speed with a time-to-first-token of 13221ms using Q4_K_M quantization.
For coding workloads, Qwen3.5 397B A17B on AMD Instinct MI350X 288GB receives a D grade with 14.6 tok/s and 6K context.
On AMD Instinct MI350X 288GB, Qwen3.5 397B A17B can safely use up to 6K tokens of context. The model's official context limit is —, but available memory constrains the safe maximum.
Remove offload with more accelerator memory. Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.
Paste this snippet into any page to show a live fit card.
<iframe src="https://willitrunai.com/embed/hf-unsloth--qwen3-5-397b-a17b-gguf-on-instinct-mi350x-288gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>
Preview: