Earlier quoted context omitted.
It's no secret that Apple's M2 series processors are a pretty minor improvement over the M1 series. Expect much bigger improvements in the M3 series which will likely be using TSMC's 3nm process (and then probably only minor improvements again for the M4 series). It seems that Apple's ridiculously impressive consistent year-on-year gains up the M1 were in large part possible because they were behind the state of the…
> Now that they've caught up with Intel and AMD... I'm an enormous fan of the M series. But there is an interesting consequence of a couple of fundamental design decisions. The M series has huge memory bandwidth but look at its focus on I/O. It reminds me of one of the design decisions of the Alto (memory bus was 3/2 the screen refresh rate, a mind-boggling decision for its time). The fact that the M1 can go in an ip…
AMD puts 96 cores into a 360w TDP (not counting spikes that go higher than that) at 3.6GHz. Apple could most likely fit 72 high-performance cores and 36 more efficiency cores into that same 360w TDP.
Given that AMD's chips require much higher clockspeeds to hit the same total performance (nearly 5GHz for Zen 3 to match a 3.2GHz M1), the final product from AMD would likely be quite a bit slower overall.
I believe the real reason for the Mac Pro not hitting the market is their insistence on unified memory. At that size, unified memory and controlling latencies explodes in complexity.
Even worse, the Mac Studio already appeals to most of the higher-end market meaning this $20-50k system probably doesn't have very many buyers either. They could sell such a product in the server market, but they left that market years ago and the reinvestment costs would be massive and very high-risk.