4 ms·
Running LLMs on Macs is still terribly slow. They simply lack the optimizations other platforms have. An RTX 6000 pro Blackwell is a pretty good card
by mountainriver 5mo ago
Running LLMs on Macs is still terribly slow. They simply lack the optimizations other platforms have.
An RTX 6000 pro Blackwell is a pretty good card
- speedgoose 5mo agoA M3 ultra mac Studio can run models that do not fit in similarly priced computers with multiple Nvidia GPUs. And it will use a lot less electricity while still having good enough performance. Except the pre-filing perfs that are quite poor on the M3.
- ttoinou 5mo agoM5 pro 48GB should be good and future proof
- thefounder 5mo agoIf you buy Mac get at least 256GB ram otherwise just buy a bunch of nvidia cards. It really does not make sense otherwise if you are looking for performance / $. The mac (studio) is unique as it has more ram than the alternatives(I.e consumer nvidia cards or spark stuff) so it can fit bigger models but otherwise its performance is worse.
- ttoinou 5mo agoI’d rather use a 128GB mac which is super flexible than always configuring a desktop with a huge GPU. Unless you’re doing others things with that gpu such as gaming, video, video AI etc.
- plutokras 5mo agoLast time I looked into it, I realized how severely it's limited by memory bandwidth. Only the M5 Ultra compares to a dedicated graphics card, but it still falls short.
- ttoinou 5mo agoM5 Pro is not that expensive and allows for model below 35GB to be used which is a lot of models. And has a boost with neural engines, its not just the memory bandwidth that makes the speed