The BEST AI Mini PC That One Upped Apple? AMD Strix Halo Beelink GTR9 Pro

This AI-powered mini PC, the Beink GTR9 Pro, boasts an AMD Ryzen AI Max Plus 395 processor, 128 GB of LP DDR5X memory, and dual 10 Gb networking. It offers a more affordable alternative to high-end systems like the Apple Mac Studio M3 Ultra, featuring a compact design inspired by the Mac Studio but with its own distinct advantages, including superior networking and a built-in fingerprint reader.

• The Beink GTR9 Pro features an AMD Ryzen AI Max Plus 395 processor, 128 GB of LP DDR5X memory, and dual 10 Gb networking.
• Its design is inspired by the Apple Mac Studio but is smaller, offering features like a fingerprint reader and built-in SD card slot.
• The system can allocate up to 96 GB of its 128 GB memory to the GPU, enabling the running of large AI models like GPT OSS 120B at impressive speeds (around 31 tokens per second).
• Power consumption is efficient, around 120-126 watts during AI inference, with noise levels around 36-37 dBA at idle and 39-41 dBA under load.
• Key lessons learned include the processor's capability, the benefits of the large memory for AI, and the need for fixes regarding preloaded software and network card firmware.

Beink GTR9 Pro Introduction and Comparison [00:00]

• The Beink GTR9 Pro is presented as an AI powerhouse with 128 GB of memory, powered by an AMD Ryzen AI processor.
• It features dual 10 Gb networking and dual USB 4 ports.
• The design is noted to be similar to the Apple Mac Studio but in a smaller form factor.
• A comparison is made to the $10,000 Apple Mac Studio M3 Ultra, highlighting the Beink GTR9 Pro as a more affordable system under $2,000.
• Multiple AMD Ryzen AI Max Plus 395 systems are discussed, with the presenter suggesting it might be the best system for many users.
• Early testing revealed some interesting findings about these systems.

This 128 gigabyte AIPC is an absolute AI powerhouse.

Chassis and Design [01:14]

• The chassis of the Beink GTR9 Pro is compared to the Apple Mac Studio, noting that Beink took inspiration from it.
• The Beink system is significantly smaller than the Mac Studio.
• Another system, the GTI machine, is visibly larger, with this trade-off attributed to its more powerful nature.
• Beink's design direction is appreciated.
• A comparison is made to the GMK tech system, which uses an external power adapter, making the Beink's internal power solution more elegant.
• The Beink also offers better networking compared to the GMK tech.
• The Framework desktop is discussed, noting it costs more and comes in pieces requiring assembly, unlike the Beink.
• The Beink is perceived as slightly smaller than the Framework desktop.
• The Beink provides better networking compared to the Framework.

Beink definitely took inspiration from the Apple Mac Studio.

Features and Connectivity [02:51]

• The system includes a power button that doubles as a fingerprint reader for Windows Hello.
• An audio jack is present on the front.
• An SD card slot is built-in, unlike the Framework desktop which requires an optional module.
• A USB Type-C port is available, identified as USB 3.2 Gen 2 (10 Gbps), not USB 4.
• A four-microphone array is featured on the front for noise-canceling capabilities, an improvement in production quality from Beink.
• On the back, there's a headset jack and two USB 4 ports (40 Gbps).
• DisplayPort and HDMI outputs are included, offering additional display connectivity options beyond USB 4.
• Two USB 3.2 Gen 2 ports are also present and are helpfully labeled by Beink.

Thank you, Beink, for doing this.

Networking Capabilities [04:35]

• A significant feature is the LAN configuration, utilizing the Intel E610 dual 10 Gb networking solution, considered a high-end setup.
• This is superior to many other mini PCs that come with 5 Gbps or even 10 Gbps Realtek NICs.
• The Intel adapter is described as new and capable of running at slower speeds if needed.
• The networking performance is noted to be faster than the 10 Gig Mac Studio.

This is probably one of the best, if not the best nick that we've seen in a mini PC thus far.

Internal Hardware: CPU, GPU, and Memory [05:24]

• The core of the system is the AMD Ryzen AI Max Plus 395, a 16-core, 32-thread processor based on the Zen 5 architecture.
• The integrated AMD Radeon 8060S graphics are highlighted for performance, with AMD using a relatively large GPU IP.
• The GPU is compared to a desktop 4060 or a 4070 mobile, significantly better than integrated graphics but not top-tier like an Nvidia RTX 5090.
• A key differentiator is the use of LP DDR5X memory instead of SO-DIMMs for DDR5.
• The system comes with 128 GB of LP DDR5X memory, allowing for faster speeds but making it non-upgradable.
• This memory configuration provides higher memory bandwidth, benefiting both the CPU and the GPU, crucial for AI workloads which can be memory bandwidth bound.
• While not matching the M3 Ultra's bandwidth, it exceeds that of lower-end mini PCs.

AMD basically said, "Hey, we're going to take our big processor and we're going to go put a relatively large GPU IP."

Storage and Cooling [07:19]

• The system houses two SSDs, though access to them could be easier.
• It has a decent cooling solution for running two SSDs.
• The primary SSD is a 2 TB M.2 NVMe SSD, which is preferred over a 1 TB option.
• AI development often requires substantial storage for downloading large models.
• A second M.2 slot allows for expansion, with the possibility of adding a 4 TB drive.
• An alternative storage strategy involves using network storage (NAS) for models, leveraging the fast networking to transfer data to the system.

If you have actually done a lot of work with AI development, something that you will find very quickly is that you need a decent amount of storage just to go and download large models.

Additional Components and Cooling System [08:20]

• The system includes internal speakers, though they are not considered great.
• It features an internal power supply, which takes up space but is appreciated for its integration.
• A significant portion of the system's design is dedicated to cooling the processor, which is rated at 140W.
• The larger chassis allows for effective cooling at a relatively low noise level, which is considered a great advantage.

I think that having a larger chassis allows Beink to do cooling at a relatively low noise level.

Performance: CPU and GPU [08:57]

• The AMD Ryzen AI Max Plus 395 with 16 cores and 32 threads offers excellent CPU performance, benefiting from the Zen 5 architecture.
• It's considered a great CPU for most tasks, short of extreme workloads that would require something like a Threadripper.
• GPU performance is also noted as performing "darn well."
• Benchmarks like Geekbench and Geekbench AI were used for comparison.
• A key aspect of performance is the memory allocation between the CPU and GPU.

Memory Allocation for AI Workloads [09:38]

• With 128 GB of LP DDR5X memory, users can allocate significant portions to the GPU.
• The tested system had 32 GB for system memory and 96 GB for GPU memory.
• This 96 GB allocation is compared to the memory capacity of an Nvidia RTX Pro 6000 Blackwell edition, although without the same memory bandwidth or compute power.
• The large GPU memory capacity enables running larger AI models.
• The GPT OSS 120B model was successfully run on the system, achieving speeds of around 31 tokens per second.
• Running larger models leads to higher quality outputs compared to smaller models.
• The system's ability to handle large models is attributed to the ample GPU memory.
• This system offers a good foundation for local AI inference when combined with a decent GPU.

That 96 GB of memory allows you to do is run larger models.

Power Consumption and Noise Levels [12:06]

• During AI inference, the system's power consumption is around 120-126 watts.
• This is considerably lower than a high-end GPU for an entire system.
• Noise levels are very low, approximately 36-37 dBA normally, and rise to only 39-41 dBA when running demanding workloads.
• The system is described as quiet at idle and acceptable when running.

This thing is running at about 36 37 dBA normally in our testing running these.

Key Lessons Learned and Potential Issues [12:43]

• The AMD Ryzen AI Max Plus 395 is a great processor with 128 GB of memory, offering a large memory footprint and a capable GPU.
• The ability to allocate memory to the GPU is a significant advantage for AI tasks.
• The price point of under $2,000 is noted, being more than some competitors but less than others like the Framework.
• The system is a complete, well-packaged solution with good connectivity.
• A significant issue encountered was preloaded "junkware" on the Windows 11 Pro installation, which was difficult to remove and reduced trust in the system image.
• Teething issues with the Intel E610 10 Gbps networking solution were experienced, including loss of network connectivity after updates due to unoptimized firmware. A fix is reportedly in development.

First off, the AMD Ryzen AI Max Plus 395 is a great processor.

AMD for AI and Future Outlook [15:57]

• While Nvidia GPUs are still considered the superior solution for AI due to wider software support and performance, AMD solutions are improving.
• Running applications like ComfyUI is possible on AMD, but it doesn't always offer the same seamless experience as Nvidia.
• AMD is increasing its emphasis on AI software with features like ROCm 7.
• The Beink unit, and AMD solutions in general, are expected to get better over time.
• Nvidia remains ahead, but these AMD solutions are considered very good.

Nvidia is still ahead, but these AMD solutions are darn good.