# Ryzen AI Max+ 395 (Strix Halo) systems

> OnPrem World · https://onpremworld.com/en/hardware/amd-ryzen-ai-max-plus-395 · Last updated: 2026-10-14 · CC BY 4.0

Strix Halo: up to 128 GB of unified memory at 256 GB/s, with up to 96 GB assignable to the GPU on Windows; its $1,999 reference system (Framework Desktop 128 GB) has the lowest launch price for 128 GB in our catalog.

## Specifications

|  |  |
|---|---|
| Memory | 32–128 GB LPDDR5X |
| Memory bandwidth | 256 GB/s |
| Power | 120 W |
| Launch | 2025-03 |
| Launch price | US$1,999 |
| Architecture | Zen 5 (16 cores) + RDNA 3.5 (Radeon 8060S, 40 CUs) + XDNA 2 NPU |
| Software | ROCm / Vulkan |

## What it can run (128 GB, 8K)

Estimates, not benchmarks. See the methodology.

| Model | Parameters | Best quant | Total | tok/s (est.) |
|---|---|---|---|---|
| Qwen3.5-122B-A10B | 122B | Q5_K_M | 85.5 GiB | 22 |
| Mistral Small 4 119B (2603) | 119B | Q5_K_M | 83.4 GiB | 33 |
| gpt-oss-120b | 117B | Q6_K | 95.0 GiB | 37 |
| Llama 4 Scout 17B-16E Instruct | 109B | Q6_K | 89.8 GiB | 11 |
| GLM-4.5-Air | 106B | Q6_K | 87.3 GiB | 16 |
| Qwen3-Coder-Next (80B-A3B) | 80B | Q8_0 | 83.8 GiB | 48 |
| Llama 3.3 70B Instruct | 70B | Q8_0 | 75.7 GiB | 2.1 |
| Qwen3.5-35B-A3B | 35B | FP16 | 69.1 GiB | 26 |
| Qwen3.6-35B-A3B | 35B | FP16 | 69.1 GiB | 26 |
| Qwen3-32B | 32.8B | FP16 | 66.6 GiB | 2.3 |
| DeepSeek-R1-Distill-Qwen-32B | 32.76B | FP16 | 66.6 GiB | 2.3 |
| Gemma 4 31B IT | 30.7B | FP16 | 61.6 GiB | 2.5 |
| Qwen3-30B-A3B (2507) | 30.5B | FP16 | 60.9 GiB | 23 |
| Qwen3-Coder-30B-A3B-Instruct | 30.5B | FP16 | 60.9 GiB | 23 |
| GLM-4.7-Flash (30B-A3B) | 30B | FP16 | 59.6 GiB | 26 |
| Qwen3.5-27B | 27B | FP16 | 53.8 GiB | 2.8 |
| Qwen3.6-27B | 27B | FP16 | 53.8 GiB | 2.8 |
| Qwen3.8-27B | 27B | FP16 | 53.8 GiB | 2.8 |
| Gemma 3 27B IT | 27B | FP16 | 54.3 GiB | 2.8 |
| Gemma 4 26B A4B IT | 25.2B | FP16 | 50.1 GiB | 20 |
| Mistral Small 3.2 24B Instruct (2506) | 24B | FP16 | 48.7 GiB | 3.2 |
| Magistral Small 1.2 (2509) | 24B | FP16 | 48.7 GiB | 3.2 |
| Devstral Small 2 24B Instruct (2512) | 24B | FP16 | 48.7 GiB | 3.2 |
| gpt-oss-20b | 21B | FP16 | 41.8 GiB | 21 |
| Phi-4 (14B) | 14B | FP16 | 29.4 GiB | 5.5 |
| Gemma 3 12B IT | 12B | FP16 | 24.8 GiB | 6.4 |
| Gemma 4 12B Unified IT | 11.95B | FP16 | 24.2 GiB | 6.4 |
| Qwen3.5-9B | 9B | FP16 | 18.4 GiB | 8.5 |
| Qwen3-8B | 8.2B | FP16 | 17.7 GiB | 9.4 |
| DeepSeek-R1-0528-Qwen3-8B | 8.19B | FP16 | 17.6 GiB | 9.4 |
| Llama 3.1 8B Instruct | 8B | FP16 | 17.1 GiB | 9.6 |
| Gemma 4 E4B IT | 8B | FP16 | 16.4 GiB | 9.6 |
| Llama 3.2 3B Instruct | 3.21B | FP16 | 7.7 GiB | 24 |
| Kimi K2.6 | 1000B | Doesn't fit | – | – |
| DeepSeek-R1-0528 (671B) | 671B | Doesn't fit | – | – |
| DeepSeek-V4-Flash | 284B | Doesn't fit | – | – |
| Qwen3.8-Flash-Next | 180B | Doesn't fit | – | – |

## Prices

No prices recorded for this device yet.

## Frequently asked questions

### Which LLMs can the Ryzen AI Max+ 395 (Strix Halo) systems run?

With 128 GB and an 8K context, the largest models that fit include: Qwen3.5-122B-A10B, Mistral Small 4 119B (2603), gpt-oss-120b, Llama 4 Scout 17B-16E Instruct, GLM-4.5-Air. See the full table below.

### How fast is the Ryzen AI Max+ 395 (Strix Halo) systems for local LLMs?

Generation speed is mostly set by memory bandwidth (256 GB/s). For Llama 3.1 8B Instruct at FP16, expect roughly 9.6 tokens/s; bigger models are slower.

### Can the Ryzen AI Max+ 395 (Strix Halo) systems run 70B models?

Yes. Llama 3.3 70B Instruct fits on the Ryzen AI Max+ 395 (Strix Halo) systems (128 GB) at Q8_0 with an 8K context, using about 75.7 GiB, at an estimated 2.1 tokens/s.

## Sources

- https://www.amd.com/en/blogs/2025/amd-ryzen-ai-max-395-processor-breakthrough-ai-.html
- https://www.amd.com/en/products/processors/laptop/ryzen/ai-300-series/amd-ryzen-ai-max-plus-395.html
- https://frame.work/blog/introducing-the-framework-desktop