Nvidia has stepped into the processor market with its RTX Spark, but at first glance, it’s clearly behind Apple Silicon by a considerable margin.
Comparing silicon instead of comparing performance at a price point is the most rookie mistake.
Most people use cheap, inefficient generic hardware for their day to day tasks. Good enough is just that. Apple silicon is so much worse than a couple B200 with EPYC, but it doesn’t have to compete with that.
If Nvidia is cheap and good enough, it is better than apple. If you need more than default ram amounts, it will perform as apple is ludicrously expensive.
Strange enough, with current crazy RAM prices, Apple’s RAM is now reasonably priced. You’re still getting absolutely fucked on storage though. Anyone who owns an Apple product should probably also own a NAS.
Fair enough, but M3 Pro out of the gate isn’t awful. The glaring standout for me is 300GB/s RAM bandwidth, pitiful for something supposedly targeted at AI applications. 5090 does 1.7TB/s, and RAM bandwidth is the bottleneck for LLM inference. M3 (and M5) Pro is 150GB/s, but M5 Max gets 460 or 614, smoking RTX Spark.
We haven’t seen prices yet, if they were good they’d be shouting about it now. The phrase ‘wounded bull’ comes to mind, but hey, apple.
Also GTX Spark software support has apparently been… lackluster. Reasonable chance that’ll improve, but I wouldn’t bet the farm (especially if they let AI loose on the drivers).
Your numbers are slightly wrong I think, M5 pro is 307 or half M5 max. Also not sure where you fit the 460 number. But yeah still a fair bit below dgpus.
Apple M5 under memory. Different variants of the Max have different numbers of memory channels.
Ah yes indeed, would be nice if Apple just published a table with specs normally lol.
