AMD Ryzen AI MAX+ PRO 495 Review: The Most Powerful Mobile APU Yet

Specification Overview

Code name: Gorgon Halo (Ryzen AI MAX+ PRO 495)

CPU architecture: Zen 5, 16 cores / 32 threads

Base clock: ~2.6 GHz, boost ~5.2 GHz

Memory support: LPDDR5X‑8533, up to 192 GB (some units 256 GB)

Integrated GPU: Radeon 8065S (upgraded 8060S, same CU count, higher clocks)

NPU performance: 55‑60 TOPS

Total heterogeneous compute: >80 TOPS (CPU+GPU+NPU)

TDP: 55 W (configurable)

Memory Bandwidth Leap

LPDDR5X‑8533 raises the memory frequency by about 6.6 % over the previous LPDDR5X‑8000 and expands the maximum capacity from 128 GB to 192 GB (some test units reach 256 GB). 128 GB can run a 70 B‑parameter model locally; 192 GB enables a full 405 B model or multiple 70 B models side‑by‑side.

GPU Increment

The Radeon 8065S is a minor refresh of the 8060S: compute unit count is unchanged, but clock speeds are slightly higher. In 3DMark Time Spy the GPU is expected to be 8 %–12 % faster than the previous generation.

NPU Advancement

The NPU delivers 55‑60 TOPS, comparable to Intel’s Lunar Lake (48 TOPS) and Qualcomm’s Hexagon (45 TOPS). Combined with CPU and GPU, the platform exceeds 80 TOPS, sufficient for on‑device LLM inference, AI‑generated imagery, and video enhancement.

Measured Performance (PassMark)

Multi‑core performance: about 4 % higher than the Ryzen AI MAX+ PRO 395. Single‑core performance: about 3 % higher than the 395.

These gains occur at the same 55 W TDP. The modest ~100 MHz frequency bump is amplified by the memory‑bandwidth increase, which benefits large‑model inference and other high‑bandwidth workloads.

Target Use Cases

Ideal Scenarios

Local large‑model inference: 192 GB unified memory allows running 70 B models without a discrete GPU and experimenting with 405 B models.

Creative workstation: Radeon 8065S approaches RTX 3060 performance in some tests, enabling video editing, 3D rendering, and AI‑generated images on a thin laptop.

Enterprise AI PC: PRO series includes AMD Memory Guard (full memory encryption), Secure Boot, and remote‑management support, pairing well with the 60 TOPS NPU for corporate AI deployments.

Unsuitable Scenarios

Pure gaming: Integrated graphics have an upper limit; gamers may prefer a future Strix Point APU combined with a discrete GPU.

Budget‑sensitive buyers: Expected price range is roughly ¥20 000–¥40 000, which may be prohibitive for cost‑conscious users.

Competitor Comparison

Memory capacity: AMD 495 – 192 GB (vs. Intel ARL‑H – 96 GB, Qualcomm X Elite – 64 GB).

Integrated graphics level: AMD 495 – RTX 3060‑class, Intel ARL‑H – RTX 3050‑class, Qualcomm X Elite – RTX 3050‑class.

NPU throughput: AMD 495 – 55‑60 TOPS, Intel ARL‑H – 48 TOPS, Qualcomm X Elite – 45 TOPS.

CPU core count: AMD 495 – 16C/32T, Intel ARL‑H – 12C/24T, Qualcomm X Elite – 12C/12T.

Memory bandwidth: AMD 495 – LPDDR5X‑8533, Intel ARL‑H – LPDDR5X‑8400, Qualcomm X Elite – LPDDR5X‑8533.

x86 compatibility: AMD 495 and Intel ARL‑H – perfect; Qualcomm X Elite – requires translation.

Selection Guidance

Run models up to 70 B: choose Ryzen AI MAX+ PRO 395 (already released, lower price).

Run models over 100 B: wait for the 495 (memory capacity determines the ceiling).

Video editing + AI image generation: Ryzen AI MAX+ PRO 395 offers the best price‑performance.

Enterprise bulk AI PC purchase: Ryzen AI MAX+ PRO 495 PRO series provides the full security feature set.

Conclusion

The Ryzen AI MAX+ PRO 495 (Gorgon Halo) marks a significant architecture upgrade for x86 mobile workstations, combining a 192 GB unified memory pool, up to 60 TOPS NPU, and GPU performance near RTX 3060, positioning AMD for the AI‑driven mobile PC market.