There will still probably be a few Macs with an AMD GPU until Apple ultimately gets to where they're going. We've already seen references to future AMD GPUs in the OS X code, so they've at least thought about it.
The M1 iMacs? AMD's drivers are allot more portable than I would have thought. I wonder if they produce byte code that's run through a hardware abstraction layer.There will still probably be a few Macs with an AMD GPU until Apple ultimately gets to where they're going. We've already seen references to future AMD GPUs in the OS X code, so they've at least thought about it.
I looked back at some reviews of the M1 GPU and if they doubled the cores and allowed for a higher TDP they could probably match an RX 580 or 5500 XT without too much trouble, but they do sell iMacs with 57xx Navi GPUs and those would be hard to beat with an SoC.
There will still probably be a few Macs with an AMD GPU until Apple ultimately gets to where they're going. We've already seen references to future AMD GPUs in the OS X code, so they've at least thought about it.
I looked back at some reviews of the M1 GPU and if they doubled the cores and allowed for a higher TDP they could probably match an RX 580 or 5500 XT without too much trouble, but they do sell iMacs with 57xx Navi GPUs and those would be hard to beat with an SoC.
There will still probably be a few Macs with an AMD GPU until Apple ultimately gets to where they're going. We've already seen references to future AMD GPUs in the OS X code, so they've at least thought about it.
I'm also not sure Apple wants to make a discrete GPU either. I don't think the volume is there for them to justify the extra costs even though they do have their own IP to use and when their SoC / APUs cover 99% of their customers.
They probably do also ha e the option of just not caring about that market though. It isn't as though anyone is buying a Mac as a gaming rig, so if they only get passable performance for casual gaming that's probably good enough for them.
Apple may not care all that much about gaming performance in the Mac but they absolutely care about GPU workstation performance, at least on the Macbook Pro and Mac Pro.
There's no reason for them to make a discrete GPU when they will already have GPU cores in each Jade-C chiplet. They just need to have enough GPU cores so that whatever the maximum number of chiplets has sufficient GPU cores to at beat the performance of the Radeon Pro W6000 - and ideally all Nvidia workstation GPU offerings as well.
They will need sufficient memory bandwidth for that, what exactly that looks like we'll have to see, as they have multiple directions they could go.
Now I suppose if it was easier to design the solution meeting those performance goals as a PCIe card rather than integrated, they could do so. I mean, the Mac Pro starts at $5000 so it isn't like a low volume part would be a problem. They could make for a fraction of what AMD/Nvidia workstation GPUs retail for. Other than possibly cooling, I can't think of any reason why having a discrete GPU would be better and there are several reasons why they would prefer an integrated GPU.
Apple actually has a very good reason to make discrete GPUs: They can stuff more cores alongside fast memory (HBM, for example) and sell those solutions at a premium to customers that require that kind of performance.
Why can't they do that with integrated cores? Nothing stops them from putting HBM controllers on Jade-C. Silicon area is cheap, putting stuff on the die you won't use in every configuration is no big deal - a lot cheaper than whole new mask set for a discrete GPU.
Silicon area is NOT cheap. Competitive designs from AMD/NVIDIA are very large. There is no getting around this.
Not every customer needs a large GPU, and you are talking about significant amounts of die area.
It is better (cheaper) to have a standardized design that can be used across multiple SKUs.
A powerful GPU chip could easily be 4-6X the size of the rumored CPU chip. That means an equivalent increase in cost.
The phys on HBM take up allot of space though - this would be tough on a highly integrated SoC with lots of I/O. In any case, really interesting times ahead with Apple's move into larger and more powerful designs.You think the addition of unused HBM controllers on Jade-C would be "NOT cheap"? I doubt it is even a single sq mm in size. That would add few dozen pennies to the per chip pricing.
You think the addition of unused HBM controllers on Jade-C would be "NOT cheap"? I doubt it is even a single sq mm in size. That would add few dozen pennies to the per chip pricing.
Geekbench 5 CPU
A15 0% IPC change
Faster from 2.99->3.23GHz
Efficiency is worse at that higher clock
ST 8% faster at 17% more power
MT 15% faster at 28% more power
MT power similar to s888
A15 vs A12X/Z: ST+56%, MT+7%
In low power mode A15 more efficient than A14
(big cores 1.3->1.4GHz at slightly lower power, little cores pull 18.4% higher power than A14 for 14% higher score)
ST 8% faster at 9% less power
MT 12% faster at 2% more power
I don't believe they tested the little cores on their own sadly, so we don't know if the 0% IPC improvement also applies there.This is disappointing to say the least.
0% IPC increase is a disaster, what on earth has their design team been doing for the past 12 months? I can only assume this is related to the brain drain from NUVIA a couple of years ago.
Pushing up clock speeds at the expense of Perf/W is a real NetBurst way of thinking, and I quite frankly expected better from Apple. I sincerely hope this is a weak year and Apple has much stronger designs in the pipeline.
Also, @uzzi38 - Could you please confirm whether the 0% IPC increase applies to the big cores only, or the big cores and little cores?
Hmm, I thought Apple was up to 3 design teams (which makes sense for yearly releases). I would have blamed COVID for the delays, offhand. So what happened, an entire design team just walked? To another company or to create a new startup?12 months is a very agressive timeline (nobody else adheres to it really). It looks pretty clear that they missed their target dates for the new cores, due to the massivebrain drain, and simply reshuffled the old ones (on possibly a slightly improved N5P process).
@CatMerc Mentioned on twitter that the original design team (that left) should still have been able to a lay the groundwork for the next CPU cores, so not having any new designs with improved IPC in work at all isn't all that likely. I do believe though that future IPC gains will be much smaller
12 months is a very agressive timeline (nobody else adheres to it really). It looks pretty clear that they missed their target dates for the new cores, due to the massivebrain drain, and simply reshuffled the old ones (on possibly a slightly improved N5P process).
@CatMerc Mentioned on twitter that the original design team (that left) should still have been able to a lay the groundwork for the next CPU cores, so not having any new designs with improved IPC in work at all isn't all that likely. I do believe though that future IPC gains will be much smaller
Except they did increase die area with A15. By like 40% too iirc.Or, more likely, no node shrink means it is much harder to squeeze more IPC out of the design without increasing die area (which they don't want to do right now due to supply issues).
Or, more likely, no node shrink means it is much harder to squeeze more IPC out of the design without increasing die area (which they don't want to do right now due to supply issues).
I guess A15 has longer pipeline design so that it can run higher clocks, and pros and cons, highest clocks while lower efficiency, but when it comes to power-saving mode A15 can roughly maintain the performance with much lower power which result in much higher efficiency compared to A14. All these changes base on 5nm+.