ASRock AI BOX-A395 Packs Ryzen AI Max+ Power for Edge AI

ASRock AI BOX-A395 (without handle)

ASRock AI BOX-A395 targets developers and engineers who want high-performance local AI computing without adding discrete accelerators. Built around the AMD Ryzen AI Max+ 395 (Strix Halo), this compact box PC combines CPU, GPU, and NPU resources inside a 200 × 100 × 232 mm aluminum chassis. It focuses on edge AI workloads, engineering simulations, and content creation tasks where latency and data privacy matter.

ASRock positions the ASRock AI BOX-A395 as a workstation-class edge system capable of running AI inference locally. The platform integrates up to 16 Zen 5 CPU cores (32 threads), AMD Radeon 8060S graphics, and an AMD XDNA 2 NPU delivering up to 50 TOPS. This setup allows developers to handle LLM workloads and GPU-heavy applications on a single machine while maintaining a compact footprint and efficient power design.

We have previously covered several box PCs, including AAEON BOXER-8742AI, Jetway B420UADN1, MSI MS-C926, and ASUS IoT EBS-P300. You can explore those as well if you’re interested in similar box PC solutions.

AI BOX A395 (with handle)

Features of the ASRock AI BOX-A395:

• Processor: AMD Ryzen AI Max+ 395 (Strix Halo), 16C/32T, up to 5.1GHz
• AI Acceleration: AMD XDNA 2 NPU delivering up to 50 TOPS
• Memory: Onboard LPDDR5X-8000, up to 128GB
• Graphics: AMD Radeon 8060S iGPU
• Display: Quad display support
  • 2× HDMI 2.1 (up to 8K @ 60Hz)
  • DP 2.1 via USB4 / Type-C
• Storage:
  • 2× M.2 Key M (PCIe Gen4 x4)
  • RAID 0 / RAID 1 support
• Networking:
  • 1× 10GbE (Marvell AQC113)
  • 1× 2.5GbE (Realtek RTL8125BG)
• Wireless: Wi-Fi 7 (802.11be) + Bluetooth 5.4 via M.2 Key E
• USB:
  • 2× USB4 (DP 2.1, 5V/3A)
  • 1× USB 3.2 Gen2 Type-C
  • 2× USB 3.2 Gen2
  • 2× USB 2.0
• Audio: Realtek ALC897 HD Audio (Mic-in, Line-out)
• Security: TPM 2.0 onboard
• BIOS: Dual BIOS with failover protection
• Power: 400W AC input (100–240V), AT/ATX modes
• Cooling: Fanned system
• Weight: 2.8 kg

AI BOX-A395 specifications

The ASRock AI BOX-A395 supports Windows 11 and Linux, which makes it flexible for both enterprise deployments and development environments. Developers can run local AI workloads using frameworks like llama.cpp and deploy quantized models such as Llama, Mistral, DeepSeek, Qwen, Gemma, and Phi. The unified memory architecture allows large models to utilize system memory efficiently, with up to 96GB accessible for GPU workloads. This design removes the dependency on cloud inference and enables faster, private AI processing at the edge.

AI BOX-A395 dimensions

ASRock Industrial has not disclosed official pricing for the ASRock AI BOX-A395 yet. That alone tells you the target audience—this is not for casual buyers. Expect enterprise-level pricing depending on configuration and region. You can check availability or request a quote through the ASRock Industrial official website or authorized distributors.

Images used courtesy of ASRock