Thank you to all. I fully understand Apple not working with NVIDIA. That has a long history.
But on the CPU Side. Even RYZEN 3900X I guess the RAM is limited to 128GB Dual Channel.
Yes, the RAM is one consideration. But it’s not just amount either. It wouldn’t support ECC, error correcting memory. Furthermore, Ryzen is a consumer platform chip, where Xeon W is a workstation chip. That means more thoroughly verified reliability which is important for the sectors the Mac Pro is targeted at.
Then there’s PCIe. Now I don’t have the lane numbers in my head, but whilst Ryzen offers more PCIe lanes than regular Intel Core chips, the Xeons offer even more. To compare you need to go to Threadripper or EPYC.
Then there’s the rest of the stack. We don’t know how easy it’d be to support things like the T2 chip with an AMD chipset. And the software stack; whilst an AMD chip could easily be made to work with macOS, it wouldn’t have the same characteristics, and a lot of optimisation work wouldn’t be relevant on AMD. If you look to WIndows even they have issue with this, reworking the scheduler a lot lately to better support AMD. Now of course Apple should be working on that regardless, like how they prepared OS X for Intel before the switch from PowerPC earlier, but still.
It’s also a matter of these things being years in the making and AMD being the underdog by a significant margin for a long time before releasing the original Ryzen. Whilst they of course put the current chips in at the time of reveal, they build the spec a lot prior based on platform characteristics that’ll likely stay constant, like the CPU-chipset connection Intel uses and TDP numbers.
Regarding GPUs, it’s already been stated about the Nvidi—Apple relationship, but I’ll add that Nvidia GPUs in fact aren’t superior at all. They are better at certain tasks, but for raw compute, the Vegas are in fact bloody amazing. They are typically constrained by their geometry units, so a lot of their shaders will idle during things like games, but throw matrix multiplication at them and they’ll push through it better than an equivalent Nvidia GPU. And for the Mac Pro market, I think that’s more important actually. Plus they have insane memory bandwidth with their large HBM2 modules.
And the Navi GPUs AMD recently brought out are very competitive with Nvidia on the geometry front as well, offering better price to performance, so if your workload is very geometry dependent, you’ll soon be able to get Navi GPUs for the Mac Pro as well in form of the W5700X. Mind it’ll probably be slower than the Vega for compute, but should offer good geometry performance.
