If you are talking about AI, it doesn't really matter.
the AMD AI max, the 395 is so expensive you could have built a 5070 TI desktop PC that would destroy the AI max in AI (of course, you may be missing out on the battery / mobility, but the price is about the same).
I don't know how the 32gb model performs because every single reviewer I have seen has the 64gb version or higher, but I assume the 32gb model has the same problem that you can't load an LLM model using 100% of the allocated space, so instead of running a 24gb model, you can only run a 16gb or something.
That sounds bad but if the AI max could run a 24gb model, it would only run it at around ~7tk/s (use smaller model, it runs faster, so a 16gb model would be 10tk/s). That's 3x faster than running the AI on a regular CPU (similar speeds as a 4060 TI 16gb), but a 5070 TI runs like 30tk/s in theory IF it could store 24gb (more like 40-50tk/s with a 16gb model, and offloading it to a regular CPU to reach 24gb would be like 15tk/s).
The 395 is a very strong CPU however, it's the closest thing to a threadripper in a laptop (without the PCIE lanes), and it has an APU similar to a 4060 (but the 385 on the framework mini PC has 80-90% of the APU performance for a lot cheaper).
>>105860219
that's AMD phoenix
AMD AI 370 is strix point
AMD AI max is strix halo
