AMD's Ryzen AI Max+ 395 pairs a 16-core Zen 5 CPU with a 40-compute-unit RDNA 3.5 integrated GPU and a dedicated XDNA 2 NPU on one package, backed by up to 128GB of unified LPDDR5x-8000 memory (per AMD's official product specifications). That combination is why the chip shows up less in everyday laptops and more as the centerpiece of mini-PCs and small-form-factor desktops built specifically to run large language models locally without a discrete GPU.
This breakdown covers the verified specification sheet, why the memory architecture matters more than the core count, where systems built around it are actually sold, and how it stacks up against building a conventional AM4/AM5 desktop instead.
What Is the Ryzen AI Max+ 395?
"Ryzen AI Max 395" is the term most buyers search, but AMD's actual product name is the Ryzen AI Max+ 395 — the "+" designates the top SKU in the Ryzen AI Max 300 series, sitting above the Ryzen AI Max 390, 385, and 380 variants. The whole family is built on AMD's "Strix Halo" silicon: a monolithic die combining CPU cores, a large integrated GPU, and an AI accelerator on memory bandwidth normally reserved for discrete graphics cards. SpecPicks covers the next tier down in Ryzen AI Max 385 Mini PC: Specs and What It's Really For, which is a useful reference point for how AMD scales the same architecture down.
Ryzen AI Max+ 395 Specifications
Per AMD's official product listing, the core specs are:
| Spec | Ryzen AI Max+ 395 |
|---|---|
| Architecture | Zen 5 |
| Cores / Threads | 16 / 32 |
| Max boost clock | Up to 5.1 GHz |
| Integrated GPU | Radeon 8060S, RDNA 3.5, 40 compute units |
| NPU | XDNA 2, up to 50 TOPS |
| Combined platform AI throughput | Up to 126 TOPS (AMD's marketed CPU + GPU + NPU figure) |
| Memory | Up to 128GB LPDDR5x-8000, 256-bit unified bus, soldered (not user-upgradable) |
| Configurable TDP | Roughly 55W–120W depending on chassis |
The number that matters most for buyers isn't the core count — it's the memory bus. Because the GPU and CPU share the same 256-bit LPDDR5x pool instead of splitting a narrow laptop memory bus against a separate discrete VRAM pool, the Ryzen AI Max+ 395 can address far more memory as usable "VRAM" than any consumer discrete GPU near its price bracket. That's the entire reason it has become a target chip for local-LLM builds, a theme SpecPicks covers in more depth in Ryzen AI Max LLM Test: What the Public Data Shows and Ryzen AI Max+ 395 LLM Performance: What to Expect.
Why the Unified Memory Matters for Local LLMs
Running a large language model locally is mostly a memory-capacity problem before it's a compute-speed problem — a model has to fit in addressable VRAM before token throughput is even relevant. A 24GB discrete GPU caps out well below what's needed to load larger open-weight models at higher precision. Because the Ryzen AI Max+ 395's 128GB memory pool can be allocated substantially toward the integrated GPU, machines built on it can load model weights that would otherwise require multiple discrete GPUs or a workstation-class card. Actual generation speed still varies heavily by model size, quantization, and context length — SpecPicks' Ryzen AI Max for Local LLMs: What the Hardware Enables walks through that variability in more detail rather than quoting a single throughput number that wouldn't hold across workloads.
Where to Buy It: Mini-PCs and the Framework Desktop
The highest-profile system built around the Ryzen AI Max+ 395 is the Framework Desktop, a compact mini-ITX-format machine designed explicitly around Strix Halo's memory bandwidth advantage for local AI workloads (per Framework's product listing). SpecPicks' Ryzen AI Max Mini-ITX: Framework Desktop Specs Explained covers that board layout, and Ryzen AI Max Mini PC 128GB: Specs, Uses, Alternatives rounds up other mini-PC vendors shipping the same silicon in different chassis designs. If you're evaluating what's actually swappable on these boards before buying, Ryzen AI Max+ 395 Motherboard Guide: What You Can Upgrade is the more relevant read — memory is fixed at purchase, so storage and cooling are typically the only meaningful upgrade paths.
Ryzen AI Max+ 395 vs. a Traditional AM4/AM5 Desktop
For buyers who don't need 128GB of unified memory for AI workloads, a conventional desktop CPU with discrete graphics remains the more predictable — and typically cheaper — path to strong gaming performance. AMD's older Ryzen 5 3600, still a common budget desktop pick, shows real-world figures like ~39 fps in Cyberpunk 2077, ~63 fps in Baldur's Gate 3, and ~72 fps in Starfield in public GPU benchmark data, all paired with a discrete GPU rather than an integrated one.
| Path | Chip | Price | Notes |
|---|---|---|---|
| AM4 budget build | Ryzen 5 3600 | $81.00 | 6-core/12-thread, needs a discrete GPU |
| AM4 budget build (Stealth cooler) | Ryzen 5 3600 | $209.99 | Same core chip, different bundle |
| AM4 mainstream build | Ryzen 5 5600X | $179.89 | 6-core/12-thread, Zen 3, needs a discrete GPU |
| AM4 higher-core build | Ryzen 7 3700X | $314.37 | 8-core/16-thread, needs a discrete GPU |
| AM4 entry build | Ryzen 5 2600 | $245.00 | 6-core/12-thread, oldest of this group |
Pair any of those with DDR4 memory — a Crucial 16GB DDR4 3200MHz kit at $125.00 or a single Crucial 8GB DDR4 3200MHz module at $71.98 — and you have an upgradable AM4 platform where RAM and GPU can both be swapped later. That flexibility is the tradeoff against the Ryzen AI Max+ 395: its memory is soldered and fixed at purchase, and it competes on integrated-GPU capability and total AI memory bandwidth rather than raw discrete-GPU gaming throughput. If gaming FPS at a known dollar figure is the priority, a traditional AM4/AM5 build with a discrete card is still the more predictable purchase; if 128GB of GPU-addressable memory for local AI workloads is the priority, nothing in a conventional desktop build matches it at this price tier.
Thermal Design and Cooling
Because the Ryzen AI Max+ 395 is sold as part of a complete system rather than a standalone socketed chip, cooling is engineered by the system vendor rather than chosen by the buyer — Framework's desktop board, for example, ships with a specific cooler tuned to its configurable TDP range. That's a meaningfully different buying decision than a traditional AM4/AM5 build, where cooler choice materially affects sustained clocks; see Noctua NH-U12S vs MasterLiquid ML240L for a Ryzen 7 5800X for how much that choice matters on a socketed AM4 chip by comparison.
Should You Buy a Ryzen AI Max+ 395 System?
It makes the most sense for buyers who specifically want to run larger local LLMs without assembling a multi-GPU workstation, and who are comfortable with a fixed memory configuration chosen at purchase. For general desktop use, content creation, or gaming-first builds, a traditional AM4/AM5 platform with a discrete GPU — using components like those above — remains the more flexible and often cheaper route, with the ability to upgrade the GPU and RAM independently over time.
Frequently Asked Questions
Is it the Ryzen AI Max 395 or the Ryzen AI Max+ 395? AMD's official product name is Ryzen AI Max+ 395 — the "+" marks it as the top SKU in the Ryzen AI Max 300 ("Strix Halo") lineup.
How much memory does the Ryzen AI Max+ 395 support? Up to 128GB of LPDDR5x-8000 across a 256-bit unified bus, per AMD's specifications.
Can you upgrade the RAM later? No — the memory is soldered to the package for bandwidth reasons, unlike a traditional AM4/AM5 build with swappable DIMMs.
What NPU does it use? AMD's XDNA 2 NPU, rated up to 50 TOPS on its own, with a marketed combined platform figure of up to 126 TOPS across CPU, GPU, and NPU.
Which computers use it? Small-form-factor mini-PCs, most notably the Framework Desktop, rather than mainstream laptops.
Is it good for gaming? Its integrated Radeon 8060S GPU is among the strongest AMD has shipped, but this synthesis has not verified specific gaming frame-rate figures for the chip itself — buyers prioritizing gaming FPS at a known price point may find a discrete-GPU desktop build more predictable.
Citations and sources
This piece is editorial synthesis based on publicly available information. No independent first-party benchmarking is reported.
