AMD Ryzen AI Max+ PRO 495 for local LLMs
Written by Jakub Rusinowski · Last updated
With 192 GB of LPDDR5X at 273 GB/s, the AMD Ryzen AI Max+ PRO 495 runs 164 catalogued models at Q4_K_M with 8K context. The largest that fits is Qwen 3 235B-A22B (MoE) (~144.3 GB), and the top pick is MiniMax M2.7 230B-A10B at 11–24 tok/s.
"Gorgon Halo": the Strix Halo design with 192 GB of LPDDR5X-8533 at 273 GB/s, 16 Zen 5 cores and a 40-CU Radeon 8065S. On Windows up to 160 GB can be given to the GPU as VRAM, and fit verdicts here use 160 GB; on Linux nearly all of the 192 GB can be used. Sold in mini-PCs from GMKtec, Minisforum, Framework, Bosgame, TOPC, Acemagic and AMD's own Ryzen AI Halo, and in the HP ZBook Ultra G3a laptop.
Models that run on the AMD Ryzen AI Max+ PRO 495
Q4_K_M, 8K context, 160 GB usable. Ranked by quality and speed.
| Model | Memory · marker = 160 GB | VRAM | Speed |
|---|---|---|---|
| MiniMax M2.7 230B-A10B MiniMax M2.7 | 139.7 GB | 11–24 tok/s | |
| MiniMax M2.5 230B MiniMax M2.5 | 139.6 GB | 11–24 tok/s | |
| Step 3.7 Flash 198B-A11B StepFun Step | 119.7 GB | 15–31 tok/s | |
| Qwen3.8-Flash-Next Qwen3.8 | 109.7 GB | 26–54 tok/s | |
| Qwen 3.5 122B-A10B Qwen 3.5 | 74.5 GB | 12–25 tok/s | |
| Nemotron 3 Super 120B-A12B Nemotron 3 Super | 73.3 GB | 11–22 tok/s | |
| Nemotron 3.5 Super VL 120B-A12B Nemotron 3.5 | 73.3 GB | 11–22 tok/s | |
| Mistral Small 4 119B-A6.5B Mistral Small 4 | 72.6 GB | 16–32 tok/s | |
| GPT-OSS 120B GPT-OSS | 71.6 GB | 29–59 tok/s | |
| Llama 4 Scout 17B Llama 4 | 68.2 GB | 9.2–19 tok/s | |
| Llama 4.5 Scout Llama 4.5 | 66.6 GB | 8.5–18 tok/s | |
| Qwen3-Coder-Next (80B-A3B MoE) Qwen3-Coder | 49.4 GB | 43–88 tok/s |
Buy it or rent the same memory
Buy the card, or rent a GPU with the same memory by the hour to try models first.
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
Speed vs other GPUs
Llama 3.1 8B, Q4_K_M. Estimated ranges. How this is calculated
Specifications
Specs last updated 2026-11-11.
- Memory
- 192 GB LPDDR5X
- Memory bandwidth
- 273 GB/s
- Memory bus
- 256-bit
- Architecture
- Zen 5 + RDNA 3.5 "Gorgon Halo"
- Series
- AI Workstation
- Release year
- 2026
- Compute backends
- ROCM, VULKAN
- Usable for models
- 160 GB
Similar GPUs
Frequently asked questions
Can the AMD Ryzen AI Max+ PRO 495 run local LLMs?
Yes. With 192 GB (160 GB usable by a model) it runs 164 of the catalogued models at Q4_K_M with 8K context; the largest is Qwen 3 235B-A22B (MoE), needing about 144.3 GB.
How fast is the AMD Ryzen AI Max+ PRO 495 for AI inference?
It is estimated to run Llama 3.1 8B at 19–40 tok/s at Q4_K_M. Llama 3.3 70B is estimated at 2.6–5.4 tok/s. These are modelled estimates from memory bandwidth, not measurements; the methodology page shows the formula.
What LLMs can I run on 192 GB?
Among the best that fit: MiniMax M2.7 230B-A10B, MiniMax M2.5 230B, Step 3.7 Flash 198B-A11B, Qwen3.8-Flash-Next, Qwen 3.5 122B-A10B.