Alex and Dan from MicroEnter build a highly compact AI workstation using a small form factor RTX Pro 4000 GPU and an ITX motherboard, overcoming challenges related to fitting powerful components into the tiny Velca 3 case. Their custom build delivers impressive AI inference speeds comparable to larger systems like the Mac Studio, demonstrating that portable, high-performance AI machines are achievable with careful component selection and persistence.
In this video, Alex and Dan from MicroEnter embark on building Alex’s smallest AI workstation yet, aiming for a compact size comparable to the Mac Studio. They start with a slightly larger Silverstone ML9 case to test component fit before transitioning to the much smaller Velca 3 case. The main challenge is fitting powerful AI-capable hardware, particularly a GPU with ample VRAM, into a small form factor. They settle on the RTX Pro 4000 Small Form Factor (SFF) GPU, which offers 24 GB of VRAM but is limited to 70W power, making it suitable for compact builds.
For the motherboard, they choose an ITX form factor MSI B850 board compatible with Ryzen AM5 CPUs, balancing cost and performance. They discuss memory options, settling on 32 GB of high-speed RAM, which is sufficient for AI workloads when paired with the GPU’s VRAM. Alex also brings a Ryzen 5 9600X processor for its efficiency and cooling advantages over higher-end CPUs. Cooling solutions include a mix of Noctua and Thermal coolers, with attention to fitting fans within the tight case constraints.
The build process proves challenging due to the compact case size and lack of included accessories like a riser cable, which is essential for GPU installation in such small cases. They encounter several setbacks, including the need to remove and reinstall components multiple times to achieve a proper fit. Despite these hurdles, they successfully assemble the workstation inside the Velca 3 case, which is only slightly larger than an iPhone in thickness and can fit into a backpack, fulfilling the portability goal.
Performance testing compares the custom-built AI workstation against Apple’s Mac Studio and Mac Mini. While the Mac Studio boasts superior memory bandwidth and can handle larger models, the RTX 4000-based build excels in prompt processing speed, outperforming the Mac Studio in that aspect. The Mac Mini, with only 16 GB RAM, lags behind both. The results highlight trade-offs between compact PC builds and high-end integrated systems, with the custom build offering competitive AI inference speeds in a portable form factor.
In conclusion, the video showcases the complexities and compromises involved in shrinking an AI workstation to a highly portable size without sacrificing too much performance. Alex and Dan demonstrate that with careful component selection and persistence, it is possible to build a powerful, backpack-sized AI machine. The project also underscores the importance of accessories like riser cables and the challenges of working with ultra-compact cases. The final build achieves a balance between size, power, and usability, making it a compelling option for mobile AI workloads.