This 128GB Mini PC Has a Performance Dial You Can

The Acemagic M1A Pro+ looks like someone designed a mini PC while heavily influenced by gaming hardware aesthetics—complete with an RGB dial on the front that actually serves a purpose beyond looking cool in YouTube thumbnails.

That dial controls three performance modes: 70 watts for silent operation, 100 watts for balanced performance, and 140 watts when you need everything this AMD Ryzen AI Max+ 395 system can deliver. The RGB changes color with each mode shift. More importantly, you can switch between them while the system is running. No BIOS menu. No reboot. Just turn the dial.

ServeTheHome tested this feature under load—running both a CPU stress test and loading a 20B parameter AI model into memory. Power consumption dropped from 163 watts to 87 watts with a simple dial adjustment. Noise levels fell from 42-43 dBA to the 36-38 dBA range. "The cool thing is if I needed something to be quieter because I was on a call or something like that, you flip this thing, give it a couple seconds, and it's at a lower power state," the reviewer notes.

What You're Actually Paying For

The M1A Pro+ costs well over $2,000, but the headline feature isn't the RGB or the tank-like chassis design. It's the 128GB of unified memory that can be split between CPU and GPU tasks. This matters for running large language models locally—the kind that don't fit in systems with conventional memory configurations.

The DRAM shortage means pricing is volatile. Systems that sold for just over $2,000 a few months ago now command higher prices, and memory costs show no signs of stabilizing through 2026-2027. The reviewer observes that what seemed expensive six months ago now looks prescient: "128 GB for $2,000 or an entire system with 128 GB for $2,000 doesn't seem so bad in hindsight."

Inside, you'll find AMD's 16-core, 32-thread Zen 5 processor paired with a Radeon 860S GPU—roughly equivalent to an RTX 4060 in terms of gaming performance. But gaming isn't the target use case here. The real application is running AI models that require substantial memory allocation.

The Actual Design Choices

Acemagic built this system around a heat pipe cooler with storage mounted vertically on side panels that slide off without tools. Pop the panel, and you have immediate access to M.2 slots. The main drive is a 2TB NVMe SSD from a brand called "Z HiT AI SSD"—which is certainly a choice in naming conventions.

There are multiple M.2 slots available, including PCIe Gen 4x4 slots that some users will inevitably use for network adapters or Oculink connections to external GPUs. The system ships with MediaTek Wi-Fi 7 and dual Realtek 2.5GbE network ports—not the 5GbE or 10GbE found on competing systems like the Framework Desktop or Minisforum MS-S1 Max, but adequate for most use cases. Those two ports support SMB Direct multi-channel for around 4+ Gbps without additional configuration.

The front panel includes two USB 3.2 Gen 2 Type-A ports, an SD card reader, and what appears to be a USB-C port with a lightning-style icon that isn't the standard Thunderbolt logo. The reviewer's assessment: "My best guess on what that is is that they're trying to show that they have a Thunderbolt port, but they didn't actually go through the Thunderbolt certification."

One notable omission: no rear USB-C port. Six USB Type-A ports total, but if you want high-speed Type-C connectivity, you're limited to that questionably-certified front port.

The AI Angle That Actually Matters

The M1A Pro+ ships with Windows 11 Pro, but Linux users can reconfigure memory allocation between CPU and GPU for better AI performance. The typical split is 32GB for system memory and 96GB for GPU memory, which allows running substantial language models locally.

For users new to local AI deployment, the reviewer suggests a straightforward path: install updated drivers, download LM Studio, and load something like GPT-4 OSS 120B. LM Studio can run as a server endpoint, meaning you can use this system as a backend for other machines running interfaces like OpenClaw.

"Now I know a lot of folks are going to be like hey AI slop it's all over it's ruining all my feeds," the reviewer acknowledges. "But on the flip side if you were to go back say 6 months that was a lot of what people were doing. But in the last I don't know 3, 4, 5 months somewhere in there, we've started to see these larger models and especially with the agentic AI."

The argument here centers on model quality versus speed. Smaller quantized models run faster but produce more errors in long workflows. Larger models running slower deliver higher quality output—which matters when you're setting up an agentic workflow and leaving it to run for hours.

How It Compares to NVIDIA's Platform

ServeTheHome has tested multiple NVIDIA GB10 systems and several AMD Ryzen AI Max+ 395 configurations. The GB10 platform scales better for multi-system deployments, but the AMD option costs less and potentially uses less power. The NVIDIA DGX Spark jumped from $4,000 to $4,700 due to the DRAM shortage.

For desktop use—Windows environment, LM Studio for AI models, occasional gaming—the reviewer leans toward the AMD platform. "If I wanted to use this as a desktop and like I wanted to run Windows and I also wanted to run LM Studio for my AI models to just make my life really easy and perhaps I also maybe every once in a while I want to play a game on it or something like that, I definitely would pick the AMD Ryzen AI Max Plus 395 right now."

Performance sits between the GMKtec EVO-X2 and the Minisforum MS-S1 Max, which runs at higher TDP and consequently uses more power. The performance dial doesn't double output by jumping from 70 watts to 140 watts, but the incremental gains exist without requiring system restarts.

The Production Unit Question

Early units had a motherboard stability issue when the SSD was installed in its default slot—a problem that only manifested under high GPU loads. Acemagic disclosed this to ServeTheHome and claims production units have the fix. Whether that's actually resolved in shipping hardware remains a data point to watch.

The 300-watt power brick deserves mention because it's comically oversized—complete with its own rubber feet. The reviewer wishes for USB-C power delivery instead, though that's probably unrealistic for a system that can pull 163 watts under load.

What's interesting about the M1A Pro+ isn't any single feature but rather the combination: accessible internal design, flexible memory configuration, and that performance dial that actually changes system behavior in real-time. Whether those features justify the premium over competing 128GB systems depends on how much you value tool-less access and on-the-fly performance adjustment. The DRAM shortage means prices will likely keep rising, making current pricing potentially the best you'll see for a while.

—Rachel "Rach" Kovacs