The video compares the cost-effectiveness and performance of a used Mac Studio Ultra versus a new AMD Ryzen AIAX Plus 395 mini PC for AI workloads, concluding that the AMD mini PC offers better value with competitive performance and warranty support. Mac users needing macOS should choose between a used M1 Ultra for slower prompt handling or a new M5 Max for faster responsiveness, while existing M1 Ultra owners are advised to keep their devices.
The video examines whether buying a used Mac Studio Ultra with 128 GB of memory is still a cost-effective option for local AI workloads compared to purchasing a new AMD Ryzen AIAX Plus 395 mini PC with similar specs. Despite listings advertising used Mac Studios for around $2,000, actual available units on platforms like eBay and refurbishers are priced between $4,000 and $5,000, making them more expensive than many new alternatives. The new AMD mini PC starts at $3,299 with 128 GB RAM and 2 TB storage, offering a more affordable and readily available option for most buyers.
Performance comparisons reveal that the Mac Studio Ultra writes text faster on smaller AI models, roughly twice as fast as the AMD mini PC in community benchmarks. However, when running larger models like GPOSS 12B, which require more memory and compute, the speed advantage of the Mac shrinks significantly to about a sixth faster, with prompt reading speeds nearly equal between the two machines. This narrowing performance gap makes the AMD mini PC a more attractive option given its lower price and newer hardware.
For users who specifically need macOS, the choice is between a used M1 Ultra or a new M5 Max Mac Studio. The new M5 Max offers faster prompt reading speeds—about 2.7 times faster than the M1 Ultra—and comparable writing speeds, making it a better buy for those who prioritize prompt responsiveness and want Apple’s warranty and return policies. The used M1 Ultra remains a viable option if the primary workload involves waiting on generated answers rather than prompt processing.
Memory allocation and software support also influence the decision. The AMD mini PC under Linux can allocate up to 120 GB of memory to the GPU, while under Windows, it is limited to 96 GB. The Mac OS reserves some memory for system tasks but can be adjusted via terminal commands. Software-wise, the AMD box runs best on Linux with ROCm support, while macOS uses Apple’s MLX framework. Neither machine supports CUDA-only tools, so users dependent on CUDA must look elsewhere.
In conclusion, the video recommends buying the new AMD mini PC for most users due to its lower price, warranty, and competitive performance on large AI models. Mac users should consider a used M1 Ultra if they mainly wait on output or invest in a new M5 Max for better prompt handling and official support. Existing M1 Ultra owners are advised to keep their machines, as switching to the AMD mini PC will not significantly improve performance. Buyers should verify stock availability and seller reliability before purchasing.