The GEEKOM IT15: Intel’s AI Push — But Is 32GB RAM Enough?

GEEKOM has unveiled the IT15 — a compact, future-proof mini PC built around Intel’s Core Ultra 9 285H Arrow Lake processor. Marketed as “Intel’s answer to AMD’s Strix Halo mini PCs,” it features 99 TOPS of AI performance and an Arc 140T GPU. But when you look past the “99 TOPS” marketing claims, there are some concerning limitations — particularly the 32GB RAM cap that makes this machine unsuitable for serious local LLM workloads.

Processor: Intel Core Ultra 9 285H — High Numbers, Platform Limitations

The IT15 is powered by Intel’s most powerful mobile AI processor:

• Core Ultra 9 285H Arrow Lake processor
• Intel Arc 140T GPU for graphics and AI acceleration
• 99 TOPS AI performance for on-device AI workloads
• DDR5-5600 memory support

Intel claims 99 TOPS of NPU performance, which sounds impressive on paper. However, Intel’s NPU architecture works differently than AMD’s — the 99 TOPS figure includes multiple AI engines working together, but in practice, local LLM inference benefits more from AMD’s unified 50-80 TOPS NPUs that are optimized for transformer models.

The Real-World Difference: 32GB RAM vs The 128GB Standard

The IT15 supports up to 32GB of DDR5-5600 memory — adequate for most workloads, though less than some AMD competitors’ 128GB. This is a fundamental limitation of Intel’s mobile platform, which typically supports less memory than AMD’s Strix Halo.

For AI professionals: Running unquantized 70B parameter models requires 40-48GB of RAM minimum. With only 32GB max, the IT15 cannot run these models natively — you’re limited to quantized 8B-14B models or must rely on cloud inference. This is a dealbreaker for serious local AI work.

Storage and Connectivity: Good Specs, Missing Features

The IT15 comes with a pre-installed 2TB SSD and M.2 NVMe slots with PCIe 4.0 support. Connectivity includes Wi-Fi 7, USB4 for 40Gbps data transfer, and comprehensive port selection.

But notably: Like many Intel mini PCs, the IT15 lacks an OCuLink port, which is a significant omission for users who want eGPU expansion capability. AMD-based Strix Halo mini PCs like the FAVM FA-EX9 and Minisforum AI X1 Pro include OCuLink as a standard feature.

Pricing and Value: How Does It Compare?

The GEEKOM IT15 is estimated to cost around $1,700–$2,000 USD (approximately £1,350 GBP, €1,560 EUR, or AU$2,650 AUD) for the 32GB/2TB configuration. But at this price point, you’re getting significantly less RAM and no OCuLink expansion compared to AMD alternatives.

vs. FAVM FA-EX9 / GEEKOM A7 Max (Ryzen AI Max+ 395, 128GB RAM, ~$1,800–$2,200 USD):

• The AMD alternatives offer 4x the RAM (128GB vs 32GB) — critical for unquantized LLM workloads
• The AMD alternatives include OCuLink for eGPU expansion; the IT15 does not
• The AMD NPU (50 TOPS) is more optimized for local LLM inference than Intel’s 99 TOPS figure

vs. Minisforum AI X1 Pro (Ryzen AI 9 HX 370, 80 TOPS NPU, 64GB RAM, ~$1,600–$2,000 USD):

• The AI X1 Pro offers 64GB RAM vs the IT15’s 32GB limit
• The AI X1 Pro’s 80 TOPS NPU is significantly better optimized for local LLM inference
• The AI X1 Pro includes OCuLink + dual USB4 for expandability

The Verdict: Intel’s AI Push or Missed Opportunity?

The GEEKOM IT15 offers impressive hardware on paper with Core Ultra 9 285H, Arc 140T GPU, and 99 TOPS NPU. However, the 32GB RAM limitation makes it unsuitable for serious local AI workloads.

The good:

• Intel’s Arc 140T GPU offers good integrated graphics performance
• Pre-installed 2TB SSD with PCIe 4.0 support
• Wi-Fi 7 and USB4 for modern connectivity
• Intel enthusiasts who prefer Intel ecosystem

The bad:

• 32GB RAM maximum — cannot run unquantized 70B+ LLMs natively
• Lacks an OCuLink port for eGPU expansion
• Intel’s “99 TOPS” figure is misleading compared to AMD’s optimized NPU performance
• Priced similarly to AMD alternatives but with significantly less RAM and no OCuLink

Who should buy it: Intel enthusiasts, users who need good integrated graphics for creative work, or those who don’t plan to run large local LLMs.

Who should avoid it: AI professionals or power users who need 128GB RAM for unquantized models, or users who want OCuLink expansion capability.

Rating: 6/10 — Good integrated graphics and modern connectivity, but the 32GB RAM limit and lack of OCuLink make it a poor choice for serious local AI work. It’s an Intel alternative, not the best value.