The AMD Ryzen AI Max+ PRO 395 processor boasts exceptional performance， particularly in large language model AI applications， to the extent that it can rival dedicated AI all-in-one machines. Furthermore， its outstanding energy efficiency allows this platform to be integrated into mini PCs and even handheld gaming devices.
Today， we’re taking a closer look at the HP Z2 Mini G1a， a high-performance mini PC engineered with the AMD Ryzen AI Max+ PRO 395 processor.
Exquisite Craftsmanship in a 2.9L Ultra-Compact Chassis
The HP Z2 Mini G1a features a 2.9L ultra-compact chassis design. At first glance， its exterior resembles a Network Attached Storage (NAS) device. The all-black body， coupled with a futuristic front panel design， makes this product exceptionally well-suited for enterprise users.
The initial impression upon handling this machine is one of robustness and remarkably high build quality. The rounded corners of the rectangular chassis are meticulously smoothed， and the diamond-patterned ventilation grille on the front not only enhances aesthetics but also ensures excellent airflow to the internal components.
On the upper center of the front panel， there are two distinct diamond-shaped elements. The upper one is the power button， and the lower one houses the product logo.
Below these， the classic HP fries logo is present. Users can slightly pull it out and rotate it to ensure the logo remains upright when the machine is placed horizontally.
Concerns about scratching the side panel when the machine is placed horizontally are alleviated by the inclusion of two rubber feet. This allows for flexible placement in either a horizontal or vertical orientation. Additionally， the machine weighs 2.4kg， comparable to that of a 16-inch gaming laptop.
With dimensions of 20×16.8×8.55cm， the compact size ensures it occupies minimal desk space， seamlessly integrating into any office environment while maintaining a tidy and comfortable workspace. Users can also opt to mount it behind a monitor or beneath a desk for further space-saving and a cleaner aesthetic.
For enterprise users requiring cluster computing power， the HP Z2 Mini G1a can be conveniently mounted in a rack. A 4U rack can accommodate five such units， leveraging the formidable AI computing capabilities of the AMD Ryzen AI Max+ PRO 395 to easily establish a private AI large model operating environment tailored to their business needs.
Compared to dedicated AI all-in-one machines， which can cost tens or even hundreds of thousands， the HP Z2 Mini G1a， priced around 20，000 RMB， presents a significantly more cost-effective solution.
Comprehensive Connectivity and Outstanding Expandability
For a mini PC， superior expandability is crucial， and the HP Z2 Mini G1a does not disappoint. On the left side of the chassis， there is one USB Type-A port with a 10Gbps transfer speed， capable of charging devices. The right side features a USB Type-C port with a 10Gbps transfer speed， supporting external power delivery and DP 2.1 video output， along with a 3.5mm audio jack.
The rear of the chassis serves as the primary interface hub. It includes a standard RJ45 2.5GbE Ethernet port， four USB ports (2x 10Gbps USB-A， 2x USB-A 2.0) for connecting peripherals like keyboards， mice， and portable high-speed SSDs. Additionally， there are two 40Gbps high-speed Thunderbolt 4 USB-C ports， offering extensive functionality and impressive expansion capabilities. Two MiniDP 2.1 ports are also present， and a MiniDP to DP cable is included to facilitate monitor connections.
Furthermore， the top and bottom interface areas on the far right reserve two Flex I/O bays. Users can select optional configurations such as serial ports， additional USB ports， extra 2.5G Ethernet ports， or even 10G Ethernet ports to suit their specific needs.
Other ports include the power connector and a security lock slot. Above the two Thunderbolt 4 ports， a small switch is visible. Flipping this switch according to the arrow allows the side cover to slide backward for tool-free removal， making hardware maintenance， upgrades， and replacements significantly more convenient.
Upon opening the side cover， the internal configuration is revealed. The motherboard area is covered by a large copper heatsink and two fans. The close integration between the fans and the heatsink ensures excellent thermal dissipation and power delivery efficiency.
In addition to the PC unit and video adapter cable， the HP Z2 Mini G1a also includes a set of HP wired keyboard and mouse as a standard offering for its mini PCs. This ensures users can get started immediately upon unboxing without the need to purchase additional peripherals.
Hardware Platform Theoretical Performance Evaluation
While the exterior design is impressive， many will be more interested in the performance of the HP Z2 Mini G1a， as it is equipped with AMD’s groundbreaking processor for the AI domain: the Ryzen AI Max+ PRO 395.
Specifications reveal a cTDP ranging from 45-120W， balancing performance and power consumption. It features a 16-core， 32-thread 4nm Zen 5 CPU with a peak boost frequency of 5.1GHz and a substantial 80MB of total cache. The integrated NPU boasts up to 50 TOPS of AI compute power， enabling a wide range of system-level local AI functionalities within Windows 11 AI+PC.
Crucial for intensive AI computations is the integrated Radeon 8060s GPU， boasting 40 Compute Units (CUs) based on the RDNA 3.5 architecture， 2560 stream processors， and AMD’s proprietary Infinity Cache to enhance memory bandwidth and reduce latency， a feature typically found in discrete graphics cards.
The AMD Ryzen AI Max+ PRO 395 supports up to 128GB of unified memory. Unlike traditional APUs that can only allocate half of the system RAM to VRAM， this platform supports up to 96GB of dedicated GPU memory. Let’s examine its theoretical and practical performance.
We have previously tested the performance of the Ryzen AI Max+ 395 processor. The Ryzen AI Max+ PRO 395 shares the same specifications but adds AMD PRO enterprise-level security management features. Therefore， the performance hinges on the cooling design of the chassis to facilitate higher sustained speeds. In CINEBENCH R23 and 2024 tests， the HP Z2 Mini G1a demonstrates its potential: achieving a single-core score of 1938 and a multi-core score of 37708 in R23， and 110 and 1881 respectively in the 2024 benchmark. The multi-core performance significantly surpasses our previous test results.
This improvement is attributed to the HP Z2 Mini G1a’s robust power delivery capabilities for the Ryzen AI Max+ PRO 395. In the AIDA 64 FPU CPU stress test， the processor consistently maintained a power draw around 120W， with an average core temperature of only 95.5℃. This indicates that the HP Z2 Mini G1a’s thermal management is exceptionally effective， allowing the Ryzen AI Max+ PRO 395 to operate at its full performance potential.
The Ryzen AI Max+ PRO 395 supports 128GB of LPDDR5x 8000 high-speed memory， achieving impressive read speeds of up to 119.74GB/s， write speeds of up to 211.65GB/s， and copy speeds of up to 149.46GB/s， demonstrating exceptionally fast memory performance.
Our review unit is equipped with a 1TB PCIe 4.0 SSD. Real-world sequential read speeds reached 4516.67MB/s， and sequential write speeds were 3640.92MB/s. 4K random read speeds were 45.59MB/s， and 4K random write speeds were 179.16MB/s， positioning it as a mid-range performer for PCIe 4.0 SSDs. This speed ensures smooth operation for daily productivity tasks， efficient loading of large language models， and quick access to image and video assets. HP also offers various capacity and speed configurations for the Z2 Mini G1a， with support for up to 2TB Z Turbo high-speed PCIe 4.0 SSDs.
The Ryzen AI Max+ PRO 395 integrates the Radeon 8060S integrated GPU. While inherently powerful， AMD’s design allows it to leverage system memory as VRAM to expand its effective capacity for running large-parameter models. It features 2560 stream processors and a core frequency of up to 2900MHz， making it one of the most potent integrated graphics solutions available today. In our configuration， 64GB of dedicated memory was allocated， with the ability to utilize half of the remaining 64GB system RAM， theoretically providing up to 96GB of usable VRAM for the GPU.
In 3DMark tests， the graphics scores for Time Spy and Fire Strike E reached 11418 and 14267 respectively. For an integrated graphics solution， these scores are exceptionally impressive， with theoretical graphics performance comparable to a discrete RTX 4060. However， when it comes to running large language models， even an RTX 5090 would fall short of its capabilities.
Tests conducted using UL Procyon’s Ryzen AI NPU Integer benchmark and Windows ML ONNX GPU Float 16 performance benchmark yielded impressive results， with scores of 1761 and 953 respectively. These figures confirm that the NPU and GPU AI compute capabilities of the Ryzen AI Max+ PRO 395 are at the forefront of the AI PC landscape.
Productivity and AI Application Performance Evaluation
After evaluating the theoretical performance， let’s delve into the practical application performance of the HP Z2 Mini G1a.
For engineering design software such as D5 Render， we tested both project rendering and 1080p video rendering. The former took only 1 minute and 30 seconds， while the latter completed in just 5 minutes and 57 seconds.
Previously， such professional rendering tasks would necessitate dedicated desktop workstations. Now， with the power of the Ryzen AI Max+ PRO 395 platform， even ultra-compact mini PCs like the HP Z2 Mini G1a can effortlessly handle them， a testament to the rapid advancements in technology.
Beyond its exceptional productivity， the most striking aspect of the Ryzen AI Max+ PRO 395 is its outstanding AI compute support.
Just one to two years ago， it was unimaginable to run AI on even the most high-end integrated graphics platforms. This is precisely why the Ryzen AI Max+ series represents a truly revolutionary step in local AI hardware. It provides an economical and practical low-cost solution for enterprises， developers， and individuals looking to deploy AI locally， engage in Generative AI (AIGC)， perform AI-powered content creation， conduct scientific computing， and develop software. It is compatible with a wide range of software and has passed multiple ISV certifications.
Furthermore， AMD has partnered with Amuse AI to develop Amuse， a dedicated AIGC tool for Ryzen AI PCs. This tool allows for rapid deployment of AIGC large models optimized for AMD Ryzen platforms. It supports features like text-to-image， image-to-image， and text-to-video generation. Notably， it can leverage AMD NPUs for super-resolution enhancement or independently perform image generation， making it a highly effective and user-friendly AIGC application.
We used the same text prompt to experience image generation across different large models. For less precise models， an image could be generated in as little as 5 seconds， while high-precision models took only 37.4 seconds to produce a 1024×1024 resolution image.
We also experimented with the Locomotion model for text-to-video generation， producing the GIF shown below in 30.6 seconds. The overall quality is remarkably satisfactory.
Finally， we deployed seven common large language models of varying parameter sizes using LM Studio. The GIF below is recorded at 1x speed and showcases the OpenAI GPT-OSS-120B model running at approximately 45 tokens/second， which is remarkably fast and more than sufficient for local AI assistant applications.
Since deployment is local， there are no network latency issues， no need to pay for token usage， and no concerns about data privacy breaches.
We also recorded the generation speeds for seven large models. The Qwen3-30b-A3B Mixture-of-Experts model achieved a generation speed of 61.48 tokens/s. Following closely are the Qwen2.5-omni-7b and OpenAI/GPT-OSS-120B， with generation speeds of 44.94 tokens/s and 38.57 tokens/s respectively. The Llama4-scout-17b and Qwen3-235b-A22b models performed at 15.72 tokens/s and 13.66 tokens/s， respectively， which are still well within the requirements for daily use. It’s important to note that due to single-card VRAM limitations， even high-end discrete GPUs typically only offer 24GB or 32GB of VRAM， making it difficult to deploy and smoothly run these very large parameter models. The Ryzen AI Max+ PRO 395 platform not only successfully deploys and loads them but also truly runs them efficiently.
Review Summary
HP’s mini PCs have consistently offered a sense of reliability and robustness. Their excellent stability， dependability， and performance have benefited numerous enterprise and professional users.
With the advent of the AI era， AMD’s Ryzen AI Max+ PRO 395 platform has proven to be a game-changer. Its powerful synergy with the HP Z2 Mini G1a AI Workstation provides a highly dependable hardware platform for enterprises， developers， and individuals keen on researching， learning， and utilizing AI. Users no longer need to spend tens or hundreds of thousands on dedicated AI all-in-one machines. A single HP Z2 Mini G1a equipped with the Ryzen AI Max+ PRO 395， priced around 20，000 RMB， can effectively address basic AI computing deployment needs.
