Quick Verdict
The most integrated Strix Halo part AMD ships, blending 16 Zen 5 cores, a desktop-class 40-CU iGPU, and 50 NPU TOPS with enterprise-grade PRO security and manageability; the trade-offs are a soldered FP11 package, locked multiplier, and the need for high-end cooling to sustain the 120W cTDP ceiling.
Overview
Launch
2025
Status
ActiveGeneration
Ryzen AI Max PRO 300 Series (1st Gen Max)
Market
Mobile Workstation / Premium AI PC
The AMD Ryzen AI Max+ PRO 395 is a 16-core, 32-thread mobile APU built on the Zen 5 'Strix Halo' design, pairing a 40-CU Radeon 8060S integrated GPU with a 50-TOPS XDNA 2 NPU and a 256-bit LPDDR5x memory interface for workstation-class throughput in thin, light, and small-form-factor systems.
Announced at CES 2025 and built on TSMC's 4nm process, the Ryzen AI Max+ PRO 395 is the flagship of AMD's Ryzen AI Max PRO 300 Series. It carries 16 full Zen 5 cores and 32 threads clocked up to 5.1 GHz, 16 MB of L2 and 64 MB of L3 cache, and a 256-bit LPDDR5x-8000 memory bus that can address up to 128 GB of system memory, with up to 96 GB convertible to GPU VRAM via AMD Variable Graphics Memory.
The integrated Radeon 8060S packs 40 RDNA 3.5 compute units running at 2.9 GHz, while the XDNA 2 NPU delivers up to 50 TOPS, contributing to an aggregate 126 TOPS platform rating.
The PRO SKU adds AMD Memory Guard, DASH manageability, Secured-core PC support, and extended planned availability over the consumer Ryzen AI Max+ 395.
Specifications
Performance
Sixteen Zen 5 cores push PassMark CPU Mark scores above 51,000 and Cinebench R23 multi-core results near 35,000, placing the PRO 395 alongside 16-core desktop Ryzen 9000 parts in multi-threaded throughput.
Full AMD-V, AMD-Vi IOMMU, and nested paging support combined with up to 128 GB of memory make the PRO 395 well suited to running several VMs or containers from a compact workstation.
The Radeon 8060S iGPU comfortably handles 1080p high settings and many 1440p titles, with performance broadly comparable to a mobile RTX 4050-4060 depending on title and power envelope; CPU-bound esports titles scale well thanks to the 5.1 GHz boost.
Zen 5 on TSMC 4nm is competitive per watt at the 55W default TDP, but sustaining the 120W cTDP ceiling in a compact chassis demands substantial cooling, and the locked multiplier limits manual tuning.
- •40-CU Radeon 8060S approaches entry-level discrete mobile GPU performance
- •256 GB/s memory bandwidth from the wide LPDDR5x bus feeds the iGPU effectively
- •5.1 GHz boost on Zen 5 cores keeps CPU-bound titles running smoothly
- •Best suited to 1080p high or 1440p medium settings rather than 4K ultra
- •50 TOPS XDNA 2 NPU for Copilot+ workloads and sustained low-power inference
- •126 TOPS aggregate platform rating (CPU + iGPU + NPU)
- •Up to 96 GB of unified memory allocatable as VRAM via AMD Variable Graphics Memory
- •Capable of running 70B-parameter class models locally with quantization, a feat impractical on most discrete mobile GPUs
Architecture
TSMC 4nm FinFET
Process Node
Strix Halo
Codename
16C / 32T
Core Config
64 MB
L3 Cache
55 W
TDP
Architecture Overview
Strix Halo is AMD's largest monolithic client SoC to date, designed to collapse the traditional CPU-dGPU-NPU trio into a single piece of silicon. Rather than reusing the chiplet layout of Ryzen 9000 desktop parts, AMD placed 16 Zen 5 cores, a 40-CU RDNA 3.5 graphics complex, an XDNA 2 NPU, a 256-bit LPDDR5x memory controller, and a PCIe 4.0 I/O fabric on one TSMC 4nm die. The PRO 395 is the full-enabled SKU of that die.
CPU Design
All 16 cores are full-fat Zen 5 units with 1 MB of dedicated L2 cache each (16 MB total) sharing a 64 MB L3 cache, supporting SMT for 32 threads. Unlike Zen 5c-based designs, every core can hit the same boost ceiling, which helps in lightly-threaded and latency-sensitive workloads. AVX-512 and the wider Zen 5 front-end benefit vector code, compilation, and emulation tasks.
Memory Subsystem
The standout architectural choice is the 256-bit LPDDR5x-8000 memory controller. Where mainstream mobile Ryzen parts use a 128-bit bus, Strix Halo doubles the width to feed both the CPU and the large iGPU, delivering up to 256 GB/s of unified bandwidth. Memory is soldered and not user-upgradable, but capacity scales to 128 GB, and AMD Variable Graphics Memory can reassign up to 96 GB of that pool as GPU-accessible VRAM for large model inference.
PCIe & I/O
The CPU exposes 16 native PCIe 4.0 lanes, typically allocated to a fast NVMe boot drive and a USB4 or external GPU link. Strix Halo also integrates native USB4 (40 Gbps) and DisplayPort 2.1 UHBR20 output, supporting up to four displays including 8K60 panels.
Overclocking
The multiplier is locked, so manual core overclocking is not available. Enthusiast tuning is limited to Precision Boost Overdrive and Curve Optimizer within OEM-permitted limits, along with memory timing via EXPO where supported. Thermal and power headroom, governed by the 45-120W cTDP range, is the primary lever for sustained performance.
- Monolithic 16-core Zen 5 die replaces chiplet Dragon Range layout
- Integrated 40-CU Radeon iGPU removes need for a discrete mobile GPU in many builds
- 256-bit LPDDR5x bus doubles memory bandwidth versus 128-bit predecessors
- 50-TOPS XDNA 2 NPU adds dedicated Copilot+ AI acceleration
- AMD PRO Technologies extend enterprise security and manageability over consumer Max+ 395
Key Highlights
- 16 full Zen 5 cores on a single monolithic die with low inter-core latency
- 40-CU Radeon 8060S iGPU approaches entry-level discrete mobile GPU performance
- 256-bit LPDDR5x-8000 bus delivers up to 256 GB/s of unified bandwidth
- Up to 96 GB of system memory allocatable as VRAM for large local LLMs
- 50-TOPS XDNA 2 NPU and 126 TOPS platform rating for Copilot+ workloads
- AMD PRO Technologies add enterprise security, DASH manageability, and extended availability
- Native USB4, DisplayPort 2.1 UHBR20, and AV1 encode/decode support
- Soldered FP11 BGA package with no socketed upgrade path
- Locked multiplier limits manual overclocking
- LPDDR5x is soldered and not user-upgradable after purchase
- Only 16 native CPU PCIe 4.0 lanes, fewer than desktop workstation platforms
- Sustained 120W cTDP requires robust cooling in compact chassis
- Premium system pricing reflects the integrated high-bandwidth design
History
Strix Halo had been rumoured for years as AMD's long-awaited large-monolithic APU, the spiritual successor to the Xbox-console-style system-on-chip concept brought to the PC. AMD officially unveiled the Ryzen AI Max and Ryzen AI Max PRO 300 Series on 6 January 2025 at CES, positioning Strix Halo as a workstation-class mobile platform that could collapse the CPU, dGPU, and NPU into one die. The PRO 395 debuted as the flagship PRO SKU, targeting mobile workstations and premium commercial systems where enterprise manageability matters as much as raw throughput.
</br></br>Throughout 2025, OEMs including HP, ASUS, and Framework shipped the chip in thin-and-light workstations, gaming tablets, and 4.5-litre mini desktops, demonstrating the design's versatility across form factors. The platform's ability to run large language models entirely on integrated graphics, using AMD Variable Graphics Memory to expose up to 96 GB of VRAM, made it a favourite among local-LLM enthusiasts.
In 2026 AMD followed up with the Ryzen AI Max PRO 400 Series (Ryzen AI Halo), extending the concept with a new generation of agent-class AI compute, while the PRO 395 remains in active production for the commercial lifecycle AMD promised at launch.
Improvements over Previous Generation
- Monolithic 16-core Zen 5 die replaces chiplet Dragon Range layout
- Integrated 40-CU Radeon iGPU removes need for a discrete mobile GPU in many builds
- 256-bit LPDDR5x bus doubles memory bandwidth versus 128-bit predecessors
- 50-TOPS XDNA 2 NPU adds dedicated Copilot+ AI acceleration
- AMD PRO Technologies extend enterprise security and manageability over consumer Max+ 395
Alternatives & Competitors
Should You Buy It?
Recommended for the right buyer
A premium mobile workstation or small-form-factor desktop where local LLM inference, 4K content editing, and enterprise manageability must coexist in one compact, low-part-count system.
Avoid if…
- You need a socketed, upgradable desktop platform with a discrete GPU
- Your workload is purely gaming and a discrete GPU desktop is feasible
- You require ECC UDIMM DDR5 rather than soldered LPDDR5x
- You want manual CPU overclocking via an unlocked multiplier
- Your budget targets mainstream productivity rather than workstation-class compute
Use Cases
Interesting Facts
Strix Halo is AMD's first client SoC to use a 256-bit-wide LPDDR5x memory bus, a width previously associated with workstation and server memory controllers.
Up to 96 GB of the 128 GB memory pool can be reassigned to the integrated GPU through AMD Variable Graphics Memory, letting the iGPU host large AI models that would not fit on typical discrete mobile GPUs.
The PRO 395 and the consumer Ryzen AI Max+ 395 share identical silicon, clocks, cache, and graphics; the PRO differentiation is entirely in security, manageability, and lifecycle features.
AMD rates the platform at up to 126 TOPS of combined INT8 throughput across the CPU, Radeon 8060S iGPU, and XDNA 2 NPU.
The Radeon 8060S is the largest integrated GPU AMD has shipped in a client mobile SoC, with 40 RDNA 3.5 compute units running at 2.9 GHz.
The chip exposes DisplayPort 2.1 UHBR20, supporting 8K60 or 4K240 output directly from the integrated graphics.
Early leaked gaming benchmarks placed the Radeon 8060S in the vicinity of an NVIDIA RTX 2060 mobile class of performance, a notable mark for integrated graphics.
Strix Halo uses a three-die package construction with a 4nm I/O die alongside the 4nm compute logic, despite being a monolithic-style client SoC.
The PRO 395 carries AMD Product ID 100-000001243 and is offered in both laptop and small desktop form factors, including mini PCs from OEMs such as GMKtec and Framework.
AMD demonstrated the platform running the GPT-OSS 120B model locally, underscoring the value of the large unified memory pool for on-device generative AI.
People Also Ask
What is the difference between Ryzen AI Max+ 395 and Ryzen AI Max+ PRO 395?
The silicon, core count, clocks, cache, and Radeon 8060S iGPU are identical. The PRO 395 adds AMD PRO Technologies: enterprise security (Memory Guard, Secure Processor, Secured-core PC), DASH remote manageability, and extended planned availability for fleet buyers.
How much RAM does the Ryzen AI Max+ PRO 395 support?
Up to 128 GB of LPDDR5x-8000 across a 256-bit interface. Memory is soldered and shared between CPU and iGPU, with up to 96 GB convertible to VRAM via AMD Variable Graphics Memory.
Can the Ryzen AI Max+ PRO 395 run large language models locally?
Yes. The combination of up to 128 GB of unified memory and the 256 GB/s bus allows quantised 70B-class and even 120B-parameter models to run on the integrated GPU, which is impractical on most discrete mobile GPUs.
Is the Ryzen AI Max+ PRO 395 good for gaming?
It is very capable for 1080p high and many 1440p titles thanks to the 40-CU Radeon 8060S iGPU, but it is not a replacement for a high-end discrete desktop GPU at 4K ultra settings.
What socket does the Ryzen AI Max+ PRO 395 use?
It uses AMD Socket FP11, a soldered BGA package, so it is not user-replaceable or upgradable; it ships only pre-installed in OEM systems.
Is the Ryzen AI Max+ PRO 395 unlocked for overclocking?
No. The multiplier is locked. Tuning is limited to Precision Boost Overdrive and Curve Optimizer within OEM-permitted limits.
What is the TDP of the Ryzen AI Max+ PRO 395?
The default TDP is 55W with a configurable TDP (cTDP) range of 45-120W, letting OEMs tune the chip for thin-and-light or higher-performance cooling designs.
Does the Ryzen AI Max+ PRO 395 support ECC memory?
It uses soldered LPDDR5x rather than ECC UDIMMs, but AMD lists the platform as supporting ECC memory capability through the SoC's memory controller; practical ECC behaviour depends on OEM implementation.
When was the Ryzen AI Max+ PRO 395 released?
AMD announced it on 6 January 2025 at CES, with systems expected to be available starting in Q1 2025.
What NPU TOPS does the Ryzen AI Max+ PRO 395 provide?
The XDNA 2 NPU delivers up to 50 TOPS, contributing to a 126 TOPS aggregate platform rating across CPU, iGPU, and NPU.
Frequently Asked Questions
Does the Ryzen AI Max+ PRO 395 have integrated graphics?
Yes. It integrates a Radeon 8060S with 40 RDNA 3.5 compute units running at 2.9 GHz, AMD's largest client mobile iGPU.
Can the Ryzen AI Max+ PRO 395 be overclocked?
The CPU multiplier is locked. Enthusiast tuning is limited to Precision Boost Overdrive and Curve Optimizer within OEM-permitted thermal and power limits.
What memory type does the Ryzen AI Max+ PRO 395 use?
It uses soldered LPDDR5x-8000 across a 256-bit bus, supporting up to 128 GB. It does not use socketed DDR4 or DDR5 UDIMMs.
How many PCIe lanes does the Ryzen AI Max+ PRO 395 provide?
Sixteen native PCIe 4.0 lanes from the CPU, typically used for a fast NVMe SSD plus a USB4 or external GPU link, with additional chipset-provided lanes for I/O expansion.
Is the Ryzen AI Max+ PRO 395 a desktop or laptop processor?
It is a mobile SoC built on the FP11 BGA package, but AMD lists it for both laptops and small-form-factor desktops, including mini PCs and the Framework Desktop.
What is AMD Variable Graphics Memory on the PRO 395?
It is a feature that lets the system reassign up to 96 GB of the 128 GB unified LPDDR5x pool as GPU-accessible VRAM, useful for running large AI models on the integrated Radeon 8060S.
Does the Ryzen AI Max+ PRO 395 support Windows Copilot+?
Yes. The 50-TOPS XDNA 2 NPU meets the Copilot+ performance threshold, enabling on-device AI features such as Recall, Live Captions translation, and Cocreator where supported by the OEM image.
What is the PassMark CPU Mark score of the PRO 395?
Submitted PassMark results place the CPU Mark near 51,567 with a single-thread rating around 4,082, reflecting strong 16-core throughput and competitive single-core performance.
Is the PRO 395 suitable for virtualization?
Yes. It supports AMD-V, AMD-Vi IOMMU, nested paging, and SLAT, and with up to 128 GB of memory it can comfortably run multiple VMs or containers in a compact workstation.
What cooling is recommended for the PRO 395?
A 280mm or 360mm AIO in desktop mini PCs, or a dual-fan vapor chamber in laptops, is recommended to sustain workloads near the 120W cTDP ceiling; the default 55W TDP is manageable with robust mobile cooling.