Pretty good chance Apple is going to point at GPU cores for those who have massive embarrasingly parallel compute needs. Current VRAM sizes are already at 40GB range ( A100 ). in 2-4 years 40GB could be well into the affordable for many in classic Mac Pro budget range.
Honestly, if Apple does this, I think it'll be the dual GPU issue all over again. Them trying to skate to a puck they think exists, but then doesn't materialize. Especially when Apple also has to push the boulder up the hill with Metal for compute.
All the more reason Apple is better off giving us some idea what the end goal of an "8,1" is in advance, IMO.
Not baffling when they currently don't have a publically facing answer. The DTK is a A12Z that has no substantive PCIe I/O bandwidth comparable to an iMac processor let along the Mac Pros. That kind of PCIe bandwidth is probably two years out.
That's my whole point though. It's not a good look.
GPU they pragmatically did with the Maya demo that was 'good enough'. That is on a two year old GPU. When the A14X lands that will only get at least incrementally better. At the Mac Pro user heavy leaning on GPU horsepower that answer those is with the PCIe facilty not the Apple GPU.... and as pointed out above that is year(s) away.
Where their GPU work lands is important compared to Intel iGPUs. If they can make Intel look bad.
Great.
Not what my argument was about though. My argument is that they should be talking about the PCIe facility
which the A12Z doesn't need because it isn't a desktop processor. They should at least say something to the effect of: "Yes, we are making sure that folks using higher end GPUs and other PCIe I/O have a way forward, we just aren't ready to show that off yet."
But even for the 27" iMac, they could really benefit from having PCIe lanes for a dGPU option.
In short, the Mac Pro isn't the leading , primary target behind the Apple Slicon design philosophy. Apple isn't going out to attack the 250-300W zone of host processors. They presented a chart to say they are trying to move away and out of that space. Not build something for it. More performance out of a already generous core count budget with constrained power consumption is their announced path. Yeah that will put a cap on embarrassingly parallel host core peformance but Apple is probably going to be "happy enough" to live in that cap. ( 20-28 cores isn't the "norm" of systems bought now. So if the current 8-16 core buyers move up to a Apple Silicon SoC with 28-32 cores in the future that will be core count upgrade. That the 28 core folks "roll off the top" of the Mac Pro range ... probably not a big loss in units or earnings after BOM expenses . )
Yes, I saw and understood the chart. I would be happy to have comparable performance for lower wattage. I'm not asking them to deliver a 200+W processor. What I'm saying is that going from a 4/4 design to a 28/4 design isn't just "slap more cores" and call it a day. Adding desktop-class PCIe and memory controllers requires extra die space as well. It all adds more complexities as the die size grows. It's a substantial jump from the types of SoCs they've made to date, and it'd just be nice for them to demonstrate that they intend to actually land there, and that they have a good shot at doing it.
I mean, Apple's basically claiming that they can go from the A12Z to workstation class SoCs in 24 months. If they actually can.
Great. But this isn't like the Intel transition where the whole lineup they wanted pre-existed, and Intel's roadmap was not exactly a secret. Here we've only got Apple's word that they'll deliver. Apple's staking a lot on this, and I'd rather not see it become another AirPower issue where they announced too quickly.
a bigger contributor is that they were not working on it in earnest until 2017. That getting the iMac Pro out the door was higher priority. Same issue likely in play here too. Getting mobile "Apple Silicon" out the door first is probably priority. Something for the iMac Pro / Mac Pro space on the "get to it eventually" list.
It still took them 2 years from announcing they were going to change direction from the trash can to putting out the demo at WWDC last year.
That's the lag we were talking about.
I'm not even arguing against what you mention about the 2017 start date. But it's not a 2 year cycle to get something out the door unless you are hell bent on customizing the engineering design like Apple did with the 7,1. In some ways, it was worth it, but in this case, it was clearly perfect being the enemy of the good adding time to the project.
