The video reviews the durable and powerful IT15 Mini PC, highlighting its strong performance in benchmarks, developer tasks, and running large language models despite its compact size and integrated GPU. Even after physical damage, the IT15 remains operational and capable of running LLMs, showcasing its resilience, portability, and versatility as a mini PC.
The video begins with an introduction to the IT15 Mini PC, a robust and compact machine touted by GeekCom for its durability. The host is intrigued by the device’s solid build, extensive connectivity options including multiple USB ports, HDMI, DisplayPort, and a metal backplate. Equipped with a powerful Intel Core Ultra 9285H mobile chip, 32 GB DDR5 RAM (upgradable to 96 GB), and a 2 TB Kingston NVMe SSD, the IT15 promises strong performance. The host even demonstrates its portability by powering it with a USB portable power supply, highlighting its potential as a truly portable mini PC.
Performance benchmarks are conducted to establish a baseline for the IT15’s capabilities. Geekbench tests reveal competitive single-core and multi-core scores, outperforming some Snapdragon-based machines. JavaScript performance tests like Speedometer 3 and the Web Tooling Benchmark show the IT15 delivering impressive results, close to Apple’s M4 Pro in some cases. These benchmarks suggest the mini PC is well-suited for developer tasks, especially those involving JavaScript and web development.
The host then subjects the IT15 to more demanding tests, including the Mandlerro Python benchmark and a large .NET compilation project. The mini PC handles these tasks efficiently, completing the Python test in 39 seconds without thermal throttling and compiling a large project in 109 seconds, making it the second fastest machine after the Apple M4 Pro in the host’s comparisons. Despite its small size, the IT15 maintains cool temperatures and stable performance under heavy CPU loads.
Shifting focus to machine learning, the host runs PyTorch training on the IT15 using a ResNet 50 model. The integrated Intel ARC 140T GPU with shared memory is utilized, but most of the training happens on the CPU, resulting in slower times compared to machines with discrete GPUs like the Apple M4 Pro. However, the host notes the potential for improvement if GeekCom includes an Oculink port for external GPUs in future models. For inference tasks, the IT15 runs various LLM models using LM Studio, successfully loading and running models up to 27 billion parameters by leveraging the expanded 96 GB RAM and GPU memory sharing, albeit at modest token generation speeds.
In the final segment, the host revisits the durability test by showing the IT15 after being physically damaged and bent, yet still operational and capable of running LLMs like the Gemma 3 1 billion parameter model at 41 tokens per second. This resilience, combined with its performance and portability, makes the IT15 one of the toughest and most versatile mini PCs tested. The video concludes with a recommendation to check out other mini PC reviews on the channel and an invitation to subscribe for more content.