AMD's Ryzen AI Max+ 395 is the flagship SKU in the Ryzen AI Max 300 series — the commercial name for the silicon AMD developed under the codename "Strix Halo." It's not a datacenter accelerator; it's a single-chip mobile/mini-PC APU that combines a full 16-core Zen 5 CPU, a 40-compute-unit RDNA 3.5 integrated GPU, and a dedicated NPU on one package, paired with a large pool of unified LPDDR5X memory. That combination — especially the up-to-128GB memory ceiling — is what has made it a talking point among people building compact local-AI workstations rather than renting cloud GPU time.
This synthesis pulls together AMD's own published specifications and public reporting to lay out what the chip actually is, where it fits, and who should (and shouldn't) consider a system built around it. For deeper spec-by-spec breakdowns and platform-specific guides, see the companion pieces on Ryzen AI Max+ 395 specs and benchmarks, the motherboard and upgrade guide, and what to expect from its LLM performance.
Key specifications
| Spec | Ryzen AI Max+ 395 |
|---|---|
| CPU cores / threads | 16 / 32 (Zen 5) |
| Integrated GPU | Radeon 8060S, RDNA 3.5, 40 compute units |
| NPU | AMD XDNA 2 |
| Memory type | LPDDR5X, unified, up to 128GB |
| Memory bus | Quad-channel, 256-bit |
| Configurable TDP | 45W–120W (platform-dependent) |
| Package | Single-die APU (CPU + GPU + NPU) |
Per AMD's published specifications, the Ryzen AI Max+ 395 is positioned as the top-end part in the lineup, above the Ryzen AI Max 385 and the base Ryzen AI Max 300-series chips. The defining architectural choice is that CPU, GPU, and NPU all draw from the same unified LPDDR5X memory pool instead of the GPU having its own dedicated VRAM — a design AMD borrowed conceptually from how Apple Silicon handles unified memory, though the underlying silicon and software stack are entirely different.
Architecture: CPU, iGPU, and NPU on one die
The 16 Zen 5 cores are the same CPU architecture AMD ships in its desktop Ryzen 9000 series, just tuned for a mobile/SFF power envelope. For general compute, multithreaded workloads, and everyday desktop use, that gives the Ryzen AI Max+ 395 CPU performance in the same class as recent high-core-count Zen 5 parts, scaled to the platform's TDP.
The Radeon 8060S integrated GPU is the largest iGPU AMD has shipped in a consumer part — 40 compute units built on the RDNA 3.5 architecture. That's roughly comparable in shader count to some recent midrange discrete mobile GPUs, which is why systems built around this chip are marketed as being able to handle 1080p and some 1440p gaming without a separate graphics card, in addition to AI workloads.
The XDNA 2 NPU handles dedicated AI inference tasks separately from the CPU and GPU, following the same NPU architecture AMD introduced in its Ryzen AI 300 laptop chips. AMD markets it as meeting Microsoft's Copilot+ PC NPU performance requirement, though the specific throughput figure depends on precision and workload, so it's best treated as a platform capability rather than a single fixed number.
Memory: why the unified pool matters more than the GPU
The headline feature isn't the CPU or the iGPU — it's the memory architecture. Up to 128GB of LPDDR5X runs across a quad-channel, 256-bit bus, and because it's unified, the GPU and NPU can address large portions of that pool directly rather than being capped by a fixed VRAM allocation the way a discrete graphics card is.
That matters most for local large-language-model inference. Most consumer discrete GPUs top out well under 32GB of VRAM, which puts a hard ceiling on how large a model (or how little it can be quantized) before it simply won't load. A system with 128GB of unified memory can hold much larger models in memory at all — the tradeoff being that LPDDR5X's bandwidth is lower than the GDDR6X or HBM memory used in high-end discrete cards, so raw token-generation throughput on a given model size varies by quantization, context length, and inference engine rather than following one universal number. The Ryzen AI Max LLM test roundup and Ryzen AI Max+ 395 LLM performance piece dig further into how that tradeoff plays out across model sizes.
One firm limitation worth flagging: because the memory is soldered LPDDR5X rather than socketed DIMMs, capacity is fixed at time of purchase. A conventional DDR4/DDR5 desktop RAM kit is not a compatible upgrade path here — buyers need to choose their memory tier upfront, unlike a standard desktop build where RAM can be added later.
Where you'll find it: mini-PCs, not laptops-only
Unlike most high-end mobile silicon, the Ryzen AI Max+ 395 has shown up prominently in small-form-factor desktop and mini-PC designs rather than exclusively in laptops — a notable departure from how AMD has historically positioned its top mobile chips. The Ryzen AI Max mini-ITX guide covers Framework's mainboard-based desktop approach, while the 128GB Ryzen AI Max mini PC review looks at fully-assembled small-form-factor systems from vendors like GMKtec. Both formats put the chip in a case with real desktop-style I/O — multiple USB4/USB-C ports, display outputs, and 2.5GbE — rather than the more constrained port selection typical of thin laptops.
Storage and expansion
Most Ryzen AI Max+ 395 mini-PCs ship with a fixed amount of internal NVMe storage and one or two expansion slots, but external storage over USB4/USB-C is a straightforward way to add capacity for model files, datasets, or media libraries. A 2TB SATA SSD or 480GB SSD in a USB enclosure is a common, low-cost way to extend storage without opening the case, and a reliable USB-C cable matters more than it sounds on these systems since USB4 ports carry both data and, on some models, power delivery. See the motherboard guide for what internal upgrade paths (if any) a given board actually supports before assuming standard desktop expandability.
Power and thermals
AMD specifies the Ryzen AI Max+ 395 as configurable across a 45W–120W TDP range depending on the platform vendor's tuning, which is why gaming and sustained-workload performance can vary meaningfully between different mini-PCs and laptops using the same chip — a vendor that tunes toward the lower end of that range will run cooler and quieter but leave performance on the table compared to one that pushes closer to 120W with a beefier cooler. This is a normal characteristic of AMD's configurable-TDP mobile silicon and not unique to this chip, but it means published benchmark numbers for one specific system don't automatically transfer to a different Ryzen AI Max+ 395 device.
Who should buy it
The Ryzen AI Max+ 395 makes the most sense for buyers who want a single, quiet, compact box that can do meaningfully more than typical integrated graphics — running larger local LLMs, doing light-to-moderate gaming, and handling general desktop work — without assembling a discrete-GPU tower. It's a poor fit for anyone whose primary use case is maximum LLM inference throughput or high-refresh competitive gaming at high settings, where a discrete GPU with faster (if smaller) memory will outperform it on raw speed. For buyers deciding between the 395 and the step-down Ryzen AI Max 385, the gap comes down mainly to CPU core count, GPU compute units, and maximum memory tier rather than a fundamentally different architecture.
Buyers assembling a more traditional AMD desktop instead — where cooling choice is a separate, user-controlled decision — may find the comparison in Noctua NH-U12S vs MasterLiquid ML240L on a Ryzen 7 5800X useful context for how much thermal headroom matters on conventional socketed AMD platforms, even though that comparison covers a different, non-unified-memory chip.
Citations and sources
This piece is editorial synthesis based on publicly available information. No independent first-party benchmarking is reported.
