ZDNET’s hands‑on with the GEEKOM IT15 Mini PC shows a clear, practical win for power users willing to swap Windows 11 for Linux: the tiny system punches well above its weight for everyday workflows, but its AI and graphics claims need careful interpretation before you buy.
The GEEKOM IT15 is one of the newest entrants in the high‑performance mini PC category, built around Intel’s Core Ultra Series 2 platform. Vendors position the IT15 as a compact desktop replacement with modern I/O, VESA mounting, and a focus on creator and productivity workloads. Early hands‑on reviews highlight two converging stories: excellent desktop responsiveness under Linux and realistic limits for local AI/ML workloads without discrete GPU support. (geekom.co.uk)
The GEEKOM IT15 demonstrates how a thoughtful mini PC can be transformed by the right software choices: the switch to Linux unlocked a smoother, more dependable daily workflow for one experienced reviewer, while the hardware sits ready for modest AI experiments and heavy productivity tasks. Buyers should weigh the appeal of a small footprint and modern I/O against the hard limits of integrated AI/graphics acceleration — and be mindful that marketing TOPS and real‑world LLM performance are not the same thing.
Source: ZDNET I replaced Windows 11 with Linux on this mini PC, and it's already paying off for my workflow
Background
The GEEKOM IT15 is one of the newest entrants in the high‑performance mini PC category, built around Intel’s Core Ultra Series 2 platform. Vendors position the IT15 as a compact desktop replacement with modern I/O, VESA mounting, and a focus on creator and productivity workloads. Early hands‑on reviews highlight two converging stories: excellent desktop responsiveness under Linux and realistic limits for local AI/ML workloads without discrete GPU support. (geekom.co.uk)What ZDNET tested and why it matters
ZDNET’s reporter shipped the IT15 up with Windows 11, found the out‑of‑the‑box experience wanting for their use case, then installed Ubuntu Budgie (dual‑boot) and used the device as a daily driver. The result: a snappier, more reliable machine for browsing, apps, and local development; weaker but usable performance when running CPU‑bound local LLMs (because the IT15 lacks an external Nvidia GPU). Those workflow gains are the headline takeaway — Linux transformed the user experience, not some secret hardware miracle.Overview: hardware and what’s actually inside
This section lays out the IT15’s core specifications and clarifies vendor wording vs. practical realities.- CPU: Intel Core Ultra 9 285H (Arrow Lake Series 2). This is a high‑end mobile/embedded chip with 16 cores (a mixture of performance and efficient cores) and a max turbo up to around 5.4 GHz. Intel’s official spec sheet lists the chip’s NPU peak at 13 TOPS (INT8) as part of the integrated Intel AI Boost subsystem. (intel.com)
- Integrated GPU: Intel Arc 140T integrated graphics (Xe‑based), offered as the on‑board GPU for Arc‑capable Arrow Lake chips. The Arc 140T supports modern features such as ray tracing and XeSS, and Intel markets combined CPU+GPU+NPU TOPS figures for AI workloads. (liliputing.com, geekom.co.uk)
- Memory & Storage: Configurations ship commonly with 32 GB DDR5 (dual‑channel SODIMM) and a 2 TB PCIe Gen4 NVMe SSD; the platform supports up to 64 GB in vendor builds. (liliputing.com, geekom.co.uk)
- Connectivity: USB4/USB‑C, multiple USB‑A ports, HDMI outputs, 2.5 GbE Ethernet, SD card reader, Wi‑Fi 7 (Intel M.2 BE200 series in many SKUs), and Bluetooth. (geekom.co.uk, itpro.com)
- Cooling & noise: A fan‑cooled design with a compact metal frame; reviewers report solid build quality but note that the fan can get audible under sustained heavy loads. (neowin.net, itpro.com)
The Linux pivot: why Ubuntu Budgie made a difference
ZDNET’s author experienced tangible daily‑use improvements after switching to Linux. The practical gains break down into several repeatable categories:- Reduced background overhead: Linux’s lighter default service footprint and the chosen desktop (Ubuntu Budgie) produced snappier app launches and more responsive desktop animations compared with the stock Windows 11 setup on the same hardware.
- Driver and kernel behavior: In this particular unit, network and USB‑C display issues that manifested on Windows either disappeared or became manageable on Linux. That’s not universal, but it’s consistent with many accounts where specific Wi‑Fi or display chipsets receive earlier or more mature support via Linux kernel drivers.
- Control and customization: The tester appreciated the ability to tune services, control updates, and remove telemetry/telephonic background tasks — changes that yield measurable improvements in perceived responsiveness.
AI workloads and local LLMs: the real limits
The IT15 ships with an Arrow Lake Ultra 9 CPU that includes an NPU and Intel Arc integrated graphics. Marketing will trumpet TOPS numbers; real life is more nuanced.What the silicon offers (the verified facts)
- Intel’s official spec sheet lists the Core Ultra 9 285H’s NPU (Intel AI Boost) at 13 TOPS (INT8) and an overall peak system TOPS (CPU+GPU+NPU) figure that Intel markets. (intel.com)
- Independent coverage and teardown reviews note that the integrated Arc 140T iGPU has matrix/AI extensions and that Intel advertises higher combined TOPS when GPU and CPU work together. Liliputing and other outlets document that the iGPU can contribute significant lower‑precision inference throughput (vendor figures report additional TOPS from the iGPU), but the integrated GPU is still limited by memory capacity and driver ecosystem compared with discrete Nvidia GPUs. (liliputing.com, geekom.co.uk)
What that means for local LLMs
- Model size and memory: Large modern LLMs (multi‑GB parameter sizes) require substantial fast memory (GPU VRAM or a capable NPU driver stack) to achieve smooth, low‑latency inference. The IT15’s integrated Arc GPU shares system memory and offers far less dedicated VRAM than a discrete Nvidia card that creators commonly use for local LLM acceleration. This leads to slower response times and higher CPU offload when running models via Ollama or similar frameworks. ZDNET’s test (gemma3:1b model) worked but was noticeably slower than a discrete‑GPU system.
- Software and driver maturity: GPU acceleration for many popular LLM stacks is still heavily Nvidia‑centric. While Intel is improving its AI stack (OpenVINO, oneAPI), current tooling and model support for Arc + NPU acceleration remains narrower than mature CUDA ecosystems. Expect progress, but plan for a software ecosystem that’s catching up. (intel.com, liliputing.com)
Practical takeaway
If your workflow depends on running large local LLMs with low latency (e.g., local model development, fine‑tuning, or high‑throughput inference), the IT15’s integrated solution will be functional for small models and experimentation but will not match a system with a discrete GPU (high‑memory Nvidia cards). For everyday productivity, lightweight models, and developer testing, it’s fine; for serious local model inference at scale, it’s a compromise.Build quality, thermals and real‑world performance
ZDNET praised the IT15’s build: compact metal internals, precise tolerances, VESA mountable chassis, and a layout that feels premium for the form factor. Independent reviewers broadly echo the sentiment. The chassis is small (roughly 117 × 112 × 45.5 mm), light, and carefully engineered. (neowin.net, itpro.com)- Thermals: Benchmarks indicate the IT15 can sustain high performance, but sustained long runs will cause the fan to ramp and noise to increase. That’s expected for a 45 W mobile‑class SKUs in a tiny box; vendor cooling design (IceBlast 2.0, vapor or heatpipe variants depending on SKU) helps, but there are practical thermal ceilings. (geekom.co.uk, neowin.net)
- Everyday snappiness: For browsing, office apps, web development, multiple monitors, and media playback, the IT15 performs extremely well — often indistinguishable from larger desktops in subjective responsiveness under Linux testers observed.
Ports, expandability, and ergonomics
The IT15 packs a surprising amount of connectivity for its size:- Dual USB4/USB‑C ports (one supporting PD), multiple USB‑A ports, dual HDMI outputs, SD card slot, 2.5 GbE, and a 3.5 mm jack. This makes it a practical compact workstation for multi‑monitor setups and peripheral heavy desks. (neowin.net, geekom.co.uk)
Price and value: mixed signals, region dependent
ZDNET’s coverage referenced a retail price near $1,100 in their test configuration, while other outlets and GEEKOM’s product pages list varying starting prices and regional SKUs. Expect wide variation by configuration, region, and retailer:- GEEKOM lists several SKUs and marketing prices; independent reviews report retail prices ranging from roughly $679 to $999+ depending on promotions, region, and RAM/SSD options. Always check the specific configuration (RAM, SSD capacity, OS) when comparing price‑to‑value. (geekom.co.uk, itpro.com, neowin.net)
Known issues, risks and caveats
No product is perfect. The hands‑on reports and third‑party reviews call out several areas prospective buyers should weigh.- Wi‑Fi and antenna performance: Some testers (including the ZDNET author) experienced inconsistent Wi‑Fi performance on Windows that improved under Linux. This suggests driver/firmware differences or marginal antenna design; don’t assume full theoretical Wi‑Fi 7 speeds in real homes. If top wireless throughput matters, test sample units or use wired 2.5 GbE where possible.
- Audio and peripheral quirks under Linux: While the reported Linux result was very positive, small annoyances such as audio configuration or specific peripheral drivers required manual fixes in some installs. Newer hardware can still need tweaks in kernels or driver stacks.
- AI claims vs. reality: Vendor TOPS figures and marketing examples (e.g., “13‑node neural engine renders 4K concept art in real time”) should be treated as marketing scenarios. Real image generation or complex model inference times depend on model type, framework support, precision, memory, and software acceleration layers. Independent benchmarking is the true arbiter. (geekom.co.uk, intel.com)
- No discrete GPU option in the tiny chassis: While the Arc 140T iGPU is capable for many workloads, this machine is not a substitute for a desktop with a discrete high‑memory GPU for heavy rendering, large‑scale ML, or AAA gaming at high frame rates.
Who should consider the IT15 (and how to pick the right config)
The IT15 is most compelling for these buyer profiles:- Professionals who need a small desktop with robust general compute for web development, office work, media editing at modest budgets, and multiple monitors.
- Linux users or devs who value control over updates, fewer background services, and the ability to tune a system to a workflow.
- Enthusiasts who want a high‑quality mini PC that’s easy to VESA‑mount behind a monitor.
- Buyers who need a local ML workstation for large, high‑memory LLMs or heavy GPU‑bound inference work.
- Hardcore gamers seeking high‑frame‑rate 4K gaming (the integrated Arc GPU is reasonable but not desktop‑class discrete GPU level).
- Choose the U9‑285H SKU if you need the highest multi‑threaded performance and occasional AI acceleration.
- Opt for 32 GB or 64 GB DDR5 if you multitask heavily or run virtualization containers.
- Prefer a 2 TB Gen4 NVMe if you work with large local datasets or video files.
- If wireless stability is critical, validate the Wi‑Fi 7 implementation in your region or use wired 2.5 GbE. (neowin.net, geekom.co.uk)
Final analysis: strengths, weaknesses, and the verdict
Strengths- Compact, premium build with surprising connectivity and upgradeability for a mini PC. (neowin.net)
- Strong everyday performance — especially under Linux for productivity and development workloads; apps launched quickly and the desktop felt responsive in ZDNET’s tests.
- Modern platform features (USB4, HDMI 2.1, 2.5 GbE, Wi‑Fi 7) that future‑proof the unit for many use cases. (itpro.com, geekom.co.uk)
- Local AI capabilities are limited in practice compared with discrete GPU systems despite attractive TOPS marketing; software ecosystem and memory constraints matter more than headline numbers. (intel.com, liliputing.com)
- Potential driver/antenna quirks (particularly on Windows in some reports) — a reminder that real‑world networking often depends on firmware, drivers, and physical antenna design as much as chip capability.
- Fan noise under sustained loads and thermal throttling potential in the tiny chassis — acceptable trade‑offs for size but relevant for continuous high CPU workloads. (neowin.net)
- For the majority of users seeking a small, fast, and well‑connected desktop replacement — especially those comfortable on Linux — the GEEKOM IT15 is a compelling option that delivers more than its size implies. For users whose core needs include heavy local LLM inference or high‑end discrete GPU tasks, a larger chassis with a discrete GPU remains the better choice. The IT15’s marketing TOPS numbers are real hardware capabilities, but treat them as one part of the story; software maturity, memory architecture, and model sizes determine practical performance. (intel.com, itpro.com)
The GEEKOM IT15 demonstrates how a thoughtful mini PC can be transformed by the right software choices: the switch to Linux unlocked a smoother, more dependable daily workflow for one experienced reviewer, while the hardware sits ready for modest AI experiments and heavy productivity tasks. Buyers should weigh the appeal of a small footprint and modern I/O against the hard limits of integrated AI/graphics acceleration — and be mindful that marketing TOPS and real‑world LLM performance are not the same thing.
Source: ZDNET I replaced Windows 11 with Linux on this mini PC, and it's already paying off for my workflow
Last edited:
