Acemagic, a company better known for budget mini PCs, dropped a surprise on July 25: the F9A, a compact Windows machine that pairs AMD’s flagship Ryzen AI Max+ 395 processor with up to 128GB of unified memory and an OCuLink port for external graphics. Pricing and a release date remain unknown, but the hardware list alone makes the F9A one of the most aggressively spec’d mini PCs ever announced.

What Acemagic Built

The F9A’s chassis measures 158 × 158 × 85 mm, roughly 2.1 liters, according to a detailed breakdown by IT Home. Inside, it’s built around AMD’s Ryzen AI Max+ 395 “Strix Halo” processor — a 16-core, 32-thread Zen 5 chip that can boost to 5.1 GHz and includes 64MB of L3 cache. That’s the same silicon AMD positions for premium mobile workstations, not sub-$500 office boxes.

But the real story is memory. The F9A uses soldered LPDDR5X-8000 RAM — fast but not upgradable after purchase — in capacities reportedly up to 128GB. In the Ryzen AI Max platform, CPU, integrated Radeon 8060S graphics, and the XDNA 2 NPU all share that memory pool. AMD says the platform can allocate up to 96GB of that as graphics memory, which transforms what an integrated GPU can do with large AI models or creative datasets.

Storage and connectivity fill out the package:
- Two M.2 slots for PCIe 4.0 x4 NVMe SSDs
- Two USB4 Type-C ports (40 Gbps each)
- Three USB Type-A ports (10 Gbps)
- HDMI 2.1
- Dual 2.5 GbE Ethernet ports
- Wi-Fi 7 and Bluetooth 5.4
- SD 4.0/UHS-II card reader
- 3.5 mm audio jack
- OCuLink 4i (PCIe 4.0 x4) for external GPUs
- A physical Copilot button, microphone array, and dual 2W speakers

Notebookcheck first reported the full specs, and IT Home confirmed many of those details independently.

Why 128GB of Unified Memory Is a Big Deal

Most mini PCs top out at 64GB of socketed DDR5. The F9A’s 128GB ceiling, combined with AMD’s unified memory architecture, means developers can run larger local language models, edit high-resolution video timelines without proxy files, or keep multiple virtual machines running — all without a discrete GPU. The Radeon 8060S iGPU has 40 RDNA 3.5 compute units, which is enough for light gaming or accelerated rendering, but the real value for AI is that the GPU can address a massive chunk of that unified pool. That bypasses the usual 4GB–8GB VRAM ceiling that strangles integrated graphics on smaller memory budgets.

Acemagic advertises 126 TOPS of total AI performance. The breakdown: the dedicated NPU contributes up to 50 TOPS, with the rest coming from the CPU and GPU. That’s enough to qualify the F9A as a Copilot+ PC, though Microsoft’s newer features often require a dedicated NPU of 40 TOPS or more. In practice, the F9A gives developers three different compute engines (CPU, GPU, NPU) to experiment with under Windows and Linux.

Connectivity: One Port Changes the Mini PC Game

OCuLink is the standout. Unlike USB4 eGPU setups that tunnel PCIe through a Thunderbolt controller, OCuLink gives an external GPU a direct PCIe 4.0 x4 link. It’s not as fast as a desktop’s x16 slot — OCuLink bandwidth here is roughly 8 GB/s versus 32 GB/s for an x16 Gen4 slot — but it’s far better than no expansion path at all. You could run a mid-range graphics card in an external dock and get playable 1440p gaming or accelerate Blender renders without replacing the whole PC.

The dual 2.5 GbE ports and Wi-Fi 7 turn the F9A into a legitimate network-attached processing node. Creative pros will appreciate the UHS-II SD reader for offloading high-bitrate video; IT admins can use one Ethernet port for management and another for a dedicated storage VLAN. The two USB4 ports also support DP alt mode, so multi-monitor setups are straightforward.

What It Means for You

Home users and students: For everyday browsing, office apps, and media, the F9A is wildly overkill. A $500 system with a Ryzen 5 or Intel Core i5 will feel just as snappy. The F9A only makes sense if you regularly edit 4K video, compile large codebases, or run local AI tools. If you don’t, wait for lower-tier Ryzen AI Max configurations — or buy a cheaper mini PC.

Power users and creators: If you’re juggling Premiere Pro, Blender, and local LLM experiments on one machine, the 128GB model is genuinely intriguing. The unified memory means you can open a 70-billion-parameter language model in llama.cpp and still have RAM left for other tasks. The integrated GPU won’t match a discrete RTX 4060 for pure 3D rendering speed, but the ability to keep large workloads entirely in memory can reduce toctime.

IT professionals and developers: The F9A could serve as a compact build server, a local AI sandbox for a development team, or a secure offline data-processing node. Dual 2.5GbE enables high-speed NAS access, and OCuLink means you can attach a GPU for CUDA or ROCm work when needed. The soldered RAM, however, locks you into a capacity choice upfront — so plan carefully. For virtualized lab environments, 128GB is plenty, but if you need ECC memory, this platform doesn’t support it.

AI enthusiasts: This is the F9A’s strongest niche. Models that are too large for a 24GB RTX 4090 can run (slowly) on the F9A’s 128GB unified pool, provided the software supports AMD’s ROCm stack or DirectML. It’s a far cheaper path to large-memory inference than a Mac Studio, though NVIDIA’s CUDA ecosystem remains more mature. Expect a tinkering experience — AMD’s AI software is improving rapidly on Windows and Linux, but it still requires comfort with command-line tools.

The Trade-offs and Unknowns

Price is a black box. Acemagic hasn’t shared a cent, and that matters enormously. The Ryzen AI Max+ 395 is a premium chip; 128GB of LPDDR5X isn’t cheap; and the connectivity suite adds cost. If the 128GB model lands at $1,200, it’s a bargain against a Mac Studio. If it’s $2,500, many will just build a small ITX system with a discrete GPU. Without a price, the F9A is a brilliant spec sheet in search of a business case.

Cooling is unproven. In a 2.1-liter box, a 16-core processor that AMD rates for up to 120W configurable TDP will put serious demands on the thermal solution. IT Home notes a 140W peak performance release figure, but that’s likely a short-burst number. Sustained all-core workloads like code compilation or AI inference could cause throttling or loud fan noise. No reviewer has tested the hardware, so any claim about quiet operation is speculation.

Memory is soldered. You cannot upgrade later. If you’re unsure between 64GB and 128GB, err on the higher side if local AI or heavy VMs are in your future. For general productivity, 32GB is already generous. This is a buy-once decision.

Security history requires caution. In 2024, Tom’s Hardware reported that some Acemagic mini PCs shipped with malware preinstalled. Acemagic acknowledged the issue and said it had resolved it with stricter audits and clean installations. That incident doesn’t automatically taint the F9A, but it’s a reminder to always perform a clean Windows install from Microsoft’s official media, update all firmware from the manufacturer’s site, and run a full antivirus scan on any new PC — especially one handling sensitive code or data.

Software ecosystem maturity. The XDNA 2 NPU is supported in Windows Studio Effects and some third-party apps, but the broad “AI PC” software wave is still building. The 126 TOPS number looks great on paper, but real-world acceleration will vary by application. Right now, the F9A’s AI muscle is most useful if you’re comfortable running open-source tools like LM Studio, Ollama, or Stable Diffusion with DirectML backends.

How We Got Here

AMD announced the Ryzen AI Max series at CES in January, describing it as a unified memory platform for “Strix Halo” mobile workstations. Early adopters like HP and ASUS have released laptops using the chip, but Acemagic’s desktop mini PC is among the first to bring it to a standalone box. The F9A follows a wave of Chinese mini PC makers pushing upmarket — GMKtec, Minisforum, and Beelink all now offer high-core-count Ryzen systems with OCuLink ports. Acemagic is betting that by embracing the platform’s maximum memory and connectivity, it can stand out in a crowded field.

What to Do Now

Until Acemagic publishes pricing, releases final specs for all configurations, and independent reviewers test thermals and acoustics, the F9A is a wait-and-see product. If you’re actively shopping for a compact AI workstation today, consider your alternatives:
- Apple Mac Studio M2 Ultra offers 192GB unified memory and a mature AI software stack, but it costs over $5,000 configured and locks you into macOS.
- Minisforum MS-01 or other Intel-based mini PCs with dedicated A2000/A4000 GPUs give you CUDA support in a small form factor, but at lower memory capacity.
- DIY ITX builds with a Ryzen 9 7950X and an RTX 4060 can hit similar price points with socketed RAM and full x16 GPU slots, though they’ll be larger and draw more power.

If you do buy an F9A later:
1. Immediately perform a clean Windows installation using Microsoft’s Media Creation Tool.
2. Download drivers only from AMD’s official support page and Acemagic’s site.
3. Verify BIOS and firmware are up to date.
4. Run Windows Security or a trusted third-party scanner before connecting sensitive accounts.
5. If using an OCuLink eGPU, check community forums for compatible docks — not all enclosures handle the link speed reliably.

Outlook

Acemagic’s F9A sketches a future where a tiny box on your desk can replace both a mid-tower workstation and a cloud AI subscription. The hardware is forward-looking; 128GB of unified memory, an OCuLink port, and Wi-Fi 7 in a 2-liter chassis is genuinely new. But the missing price and the unproven cooling mean that judgment day is still months away. If Acemagic gets the fundamentals right — aggressive pricing, stable firmware, and a clean software image — the F9A could become the de facto recommendation for portable AI development. If not, it’ll be another promising spec sheet in a hobbyist’s forum thread.