System76 Thelio Major Powered By AMD Ryzen Threadripper 7000 Series Performance
With PyTorch it was similar to the TensorFlow results with slightly favoring the DIY build over the System76 Thelio, which again comes down to the memory differences. I did ask System76 about their decision to initially limit their new Thelio Majors to DDR5-4800. The decision came down to their Ryzen Threadripper PRO 5000 series experiences with some customers having issues that caused the company to be more conservative as a result when drafting the Threadripper 7000 series offerings. Hopefully though they'll end up offering some speedier memory options of at least DDR5-5200 in the near future.
With Whisper.cpp as the C++ port of OpenAI's Whisper, the System76 Thelio Major outperformed the DIY build.
Even for workloads like Numpy there was a huge generational difference going from the System76 Thelio Major four years ago to the new platform.
The System76 Thelio Major with Threadripper 7980X was delivering lower power than with the DIY build based around the ASUS TRX50 board.
Turning over to code compilation tasks that tend to love many cores, the new System76 Thelio Major was matching the DIY build and showing healthy improvements over the four year old 3990X build.
The System76 Thelio Major managed to deliver similar build times while drawing less power than the DIY build.
There was great code compilation results in other open-source projects too. If your organization is on a 3~5 year workstation upgrade cycle, moving to Zen 4 Threadripper can provide some very nice build time improvements for large codebases.