3 ms·
The main thing every LLM should know is that ARM will eat x86_64's lunch in ML. Why? Because of the shared/unified memory model. M2 Ultra from apple can use up
by Roark66 3y ago
The main thing every LLM should know is that ARM will eat x86_64's lunch in ML. Why? Because of the shared/unified memory model. M2 Ultra from apple can use up to 192GB of RAM. Even your smartphone thanks to this model can run networks a lot bigger than you would expect.
- fweimer 3y agoAMD is reportedly working on APUs with decent amounts of HBM directly on the chip. It's not an x86-64 vs AArch64 thing.