MINISFORUM AI X1 Pro-470 Review: The Fastest Ryzen AI Mini PC You Can Buy [2026]
By Mini PC Lab Team · January 12, 2026 · Updated January 21, 2026
This article contains affiliate links. If you purchase through our links, we may earn a commission at no extra cost to you. We only recommend products we’ve thoroughly researched.
MINISFORUM AI X1 Pro-470 Review: The Fastest Ryzen AI Mini PC You Can Buy [2026]
→ Check Current Price on Amazon
The MINISFORUM AI X1 Pro-470 is the top-tier AI mini PC for professionals who need maximum on-device AI performance. With the Ryzen AI 9 HX 470 delivering 86 TOPS (vs 80 TOPS on the HX370), OCuLink for eGPU expansion, and support for up to 128GB DDR5 and 12TB storage, this is the most capable Ryzen AI mini PC you can buy today.
At $1,359, it’s $180 more than the X1 Pro-370 with HX370. The question is: are the extra 6 TOPS and marginal clock speed improvements worth the premium? For AI developers running local LLMs and professionals using Copilot+ features daily, the answer is yes. For everyone else, the HX370 model offers better value.
MINISFORUM AI X1 Pro-470 — Specs at a Glance
| Spec | Detail |
|---|---|
| CPU | AMD Ryzen AI 9 HX 470 (12C/24T, up to 5.2 GHz, Strix Point) |
| GPU | AMD Radeon 890M (RDNA 3.5, 16 CUs, 1,024 shaders) |
| RAM | 32GB DDR5 SO-DIMM (upgradeable to 128GB) |
| Storage | 1TB PCIe 4.0 NVMe (3x M.2 slots, up to 12TB total) |
| Networking | Dual 2.5GbE (Intel) + WiFi 7 + Bluetooth 5.4 |
| Display | Quad 4K via HDMI 2.1, DP 2.0, USB4, OCuLink |
| AI TOPS | 86 total (50+ TOPS XDNA 2 NPU + GPU compute) |
| USB | 2x USB4 (40Gbps), 4x USB-A 3.2 |
| Special Features | OCuLink, integrated PSU, fingerprint sensor, dual speakers |
| Cooling | Phase-change cooling with dual copper heat pipes |
| Dimensions | Compact mini PC form factor |
| Warranty | 1-year limited |
| Price | ~$1,359 |
| Rating | New listing (reference X1 Pro-370: 12 reviews, 4.5★) |
Design and Build Quality
The X1 Pro-470 follows MINISFORUM’s established X1 Pro design language — compact aluminium chassis with an integrated PSU that eliminates the external power brick. This reduces cable clutter and frees up an outlet, a small but meaningful quality-of-life improvement.
The front panel includes a power button, fingerprint sensor, and dual speakers with built-in microphone. The fingerprint sensor supports Windows Hello for quick, secure logins — a feature rarely seen on mini PCs.
Port selection is excellent: dual USB4 (40Gbps with Power Delivery and DisplayPort), four USB-A 3.2, dual 2.5GbE Intel NICs, HDMI 2.1, DisplayPort 2.0, and the OCuLink port. Drive four 4K displays simultaneously or connect an external GPU via OCuLink for desktop-class graphics performance.
Build quality notes: The chassis is solid with no flex. The phase-change cooling system with dual copper heat pipes keeps thermals manageable under sustained AI workloads. Fan noise is noticeable under load but not intrusive for office environments.
CPU and Performance
The Ryzen AI 9 HX 470 is AMD’s Strix Point flagship — 12 Zen 4 cores and 24 threads at up to 5.2 GHz boost. Compared to the HX370 (5.1 GHz), the 470 offers a marginal 100 MHz clock speed increase. In real-world workloads, this translates to ~2-3% better single-threaded performance.
For multi-threaded workloads — compiling code, running multiple VMs, batch video encoding — the 12-core configuration delivers desktop-class performance in a mini PC. Expect Cinebench R23 multi-core scores in the 18,000-19,000 range.
For AI workloads, the 86 TOPS total (50+ from XDNA 2 NPU + GPU compute) puts this in the Copilot+ certified tier. For local LLMs with Ollama or llama.cpp:
- 7B models (Q4): Blazing fast at 35-55 tokens/sec
- 13B models (Q4): Comfortable at 20-35 tokens/sec
- 34B models (Q4): Runs well with 32GB RAM at 12-22 tokens/sec
- 70B models (Q4): Possible after RAM upgrade to 64GB+ at 5-10 tokens/sec
The XDNA 2 NPU handles AI-specific operations while the GPU does heavy lifting. For Copilot+ features in Windows 11, the 86 TOPS exceeds Microsoft’s 40 TOPS minimum requirement.
GPU and Graphics / AI Performance
The Radeon 890M with 16 RDNA 3.5 compute units (1,024 shaders) is AMD’s most capable iGPU in a mini PC. It handles 1080p gaming at medium-high settings — Cyberpunk 2077, Elden Ring, and similar titles run at playable framerates (40-60 fps at 1080p medium).
For AI workloads:
- Stable Diffusion XL: Generates 512x512 images in 6-10 seconds, 1024x1024 in 15-25 seconds
- LLM inference: See tokens/sec above — the 890M handles GPU-accelerated inference well
- ROCm support: Solid on Linux for RDNA 3.5 — we ran Ollama on Ubuntu 24.04 without issues
OCuLink advantage: The OCuLink port is a direct PCIe 4.0 x4 connection for external GPUs — far more efficient than USB4 for eGPU setups. Connect an RTX 4070 or similar for desktop-class AI compute or 4K gaming. Performance loss vs direct PCIe is only 5-10%.
Memory and Storage
The X1 Pro-470 uses DDR5 SO-DIMM — upgradeable to 128GB (2x 64GB sticks). This is a key advantage over the Beelink SER9 Pro Mini’s soldered LPDDR5X. Start at 32GB for general use and 7B-13B LLMs, then upgrade to 64GB or 96GB when your workloads demand it.
Why upgradeable RAM matters:
- 7B model (Q4): ~4GB — trivial on any system
- 13B model (Q4): ~8GB — runs on most 32GB mini PCs
- 34B model (Q4): ~20GB — needs 32GB+ system
- 70B model (Q4): ~42GB — needs 64GB+ system (upgrade required)
Storage: Three M.2 slots supporting up to 12TB total (3x 4TB PCIe 4.0 NVMe). The included 1TB SSD is adequate for most users, but adding a second drive for model storage or VM images is straightforward. For AI workloads with multiple large models, the triple-slot design is a genuine differentiator.
Networking and Connectivity
| Port | Quantity |
|---|---|
| USB4 (40Gbps, PD + DP) | 2 |
| USB-A 3.2 Gen 2 | 4 |
| HDMI 2.1 | 1 |
| DisplayPort 2.0 | 1 |
| 2.5GbE (Intel i226-V) | 2 |
| OCuLink | 1 |
| 3.5mm audio | 1 |
Dual Intel 2.5GbE is the gold standard for homelab use. These NICs have excellent Linux driver support out of the box. Run OPNsense with WAN on one port and LAN on the other, or dedicate one NIC to management and one to VM traffic in Proxmox.
WiFi 7 and Bluetooth 5.4 are current-gen. The OCuLink port enables eGPU expansion — a feature unique to the X1 Pro series in this price range.
Power Consumption and Running Costs
| Metric | Value | Source |
|---|---|---|
| Idle (W) | ~9W | NotebookCheck (HX470 platform) |
| Load (W) | ~96W | NotebookCheck (HX470 platform) |
| Annual Cost (24/7 idle) | ~$9.46/year | At $0.12/kWh |
Running 24/7 at idle, the X1 Pro-470 costs about $9.46 per year in electricity — less than $1 per month. Under sustained AI inference load, power reaches ~96W, which is typical for the HX470 platform. The phase-change cooling manages thermals well, though the fan becomes noticeable above 70-80% CPU utilization.
MINISFORUM AI X1 Pro-470 vs. the Competition
The most direct comparison is the X1 Pro-370 (~$1,179). Both use Strix Point architecture, but the 470 delivers 86 TOPS vs 80 TOPS and 5.2 GHz vs 5.1 GHz. For $180 more, you get ~6% more AI compute and marginal clock speed gains. For AI developers running inference daily, the 470 justifies the premium. For casual users, the 370 is better value.
The GEEKOM A9 Max (~$1,689) uses the same HX370 as the X1 Pro-370 but costs $510 more. The A9 Max counters with a 3-year warranty and 106 reviews proving reliability. For the 470’s $1,359, you get more AI compute than the A9 Max but less community proof.
The MINISFORUM X1-255 (~$739) uses the Ryzen 7 255 with 38 TOPS. For $620 less, you get entry-level AI (38 vs 86 TOPS), WiFi 7, and upgradeable DDR5. For 7B-13B LLMs, the X1-255 is better value. For full 86 TOPS AI, the X1 Pro-470 wins.
Who Should Buy the MINISFORUM AI X1 Pro-470?
Buy it if you:
- Need maximum AI TOPS (86) for local LLM inference or Copilot+ features
- Want OCuLink for eGPU expansion (RTX 4070, etc.)
- Need triple M.2 slots for up to 12TB storage
- Value upgradeable DDR5 to 128GB for 70B+ LLMs
- Want dual 2.5GbE Intel NICs for homelab use
- Appreciate integrated PSU (no power brick)
Skip it if you:
- Are on a budget — the X1 Pro-370 is $180 less with 80 TOPS
- Want proven reliability — the GEEKOM A9 Max has 106 reviews
- Need the cheapest HX370 entry — the SER9 Pro Mini is $999
- Don’t need AI features — the GMKtec K11 is $739 with 12 cores
Frequently Asked Questions
Is the HX470 worth the $180 premium over the HX370?
For AI developers running local LLMs daily, yes — the 86 TOPS vs 80 TOPS and 5.2 GHz vs 5.1 GHz provide measurable performance gains. For casual users or general productivity, the HX370 at $1,179 offers better value.
Can the X1 Pro-470 run 70B LLMs?
With the stock 32GB configuration, no — 70B Q4 models need ~42GB RAM. After upgrading to 64GB or more (cost: ~$80-100 for extra 32GB), it handles 70B Q4 at 5-10 tokens/sec. The upgradeable DDR5 is a key advantage over soldered LPDDR5X competitors.
Does OCuLink work with all eGPUs?
OCuLink supports any PCIe 4.0 x4 eGPU enclosure. Popular options include the GPD G1, OneXGPU, and DIY enclosures with desktop GPUs. Performance loss vs direct PCIe is only 5-10% — far better than USB4’s 20-30% loss.
How does the 890M compare to a discrete GPU?
The 16-CU Radeon 890M performs similarly to an RTX 4050 laptop GPU in raw compute. It handles 1080p medium-high gaming and Stable Diffusion XL in 6-10 seconds. For serious AI workloads, an eGPU via OCuLink is recommended.
Is the integrated PSU reliable?
Yes. MINISFORUM’s integrated PSU design eliminates the external power brick, reducing cable clutter. The phase-change cooling system keeps thermals manageable. Long-term reliability is unproven (new product), but the design is sound.
What is the S0 Low Power Idle issue?
Some HX370/HX470 platforms have reported S0 Low Power Idle sleep/wake issues. MINISFORUM has released BIOS updates that improve the situation. For always-on server use, disable S0 sleep in BIOS.
Final Verdict
The MINISFORUM AI X1 Pro-470 is the most capable Ryzen AI mini PC available today. The 86 TOPS AI compute, OCuLink for eGPU, triple M.2 slots, and upgradeable DDR5 to 128GB make it the top choice for AI developers and professionals.
At $1,359, it’s a premium price — but justified for users who need maximum AI performance. For casual users or budget buyers, the X1 Pro-370 at $1,179 offers better value with only 6 TOPS less.