Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
If the XBox-X has a 12 TFLOP GPU on its SoC, what stops Apple doing the same?
- Thread starter johngwheeler
- Start date
- Sort by reaction score
The problem that I have with this debate centring on CPU and GPUs in the ASi system is the neglect to recognise the large number of specialised sub-processors in the CPU that speeds up the most demanding operations a conventional CPU encounters - often with a factor of 100 to 1000 times faster. Especially when directly supporting the APIs in the MacOS. CPU and even GPU performance might not be the most important factor - the efficient use of specialised sub-processors will probably be the most important aspect of the ASi.
These are still niche cases. Yes, image classification will run much faster, which is undeniably cool, but a lot of users will care about system or gaming performance as well. Well-integrated domain specific accelerators in Apple Silicon are definitely going to be a major selling point, but they are not a replacement for solid base performance.
Of course the base performance is important - but in the case of applications running in MacOS most of them are calling core routines that are directly optimised to use these specialised processor cores - speeding up all apps with no additional developer effort. This will probably deliver higher functional performance than the CPU performance differences might indicate. There is a reason why the iOS devices has such high performance in demanding applications and processes.
I think the XBox Series S would be the better comparison to the potential Apple Silicon Macs than the Series X. Putting aside the fact that the Series S looks like Microsoft ripped off the design of their own Adaptive Controller for that console, Microsoft decided to use a GPU with 1/3 the power of the Series X (4 TFLOPS). The die size for the Series S is 197.05mm^2 while the Series X die is 360.45mm^2.
Yield issues in defects or in operating speeds?
Sony and Microsoft took two different approaches to how "hard" they run their AMD based designs. Sony has a smaller GPU ( 36 compute units at base 2.23GHz and variable speed) and Microsoft has a bigger GPU ( 52 compute units at 1.8GHz at a locked clock). Similar on the CPU cores : Sony 3.5Ghz and Xbox 3.5-3.8GHz .
Microsoft has a wider and slower approach and Sony is leverage higher clock speeds on a 'narrower' GPU. Sony chip costs a bit less if the physical defect rate is the same. Microsoft could be sacrificing increased die cost for more stable chips.
If it is actual physical defects are the same (on the same TSMC process) then Microsoft probably has bigger problems.
Bringing the discussion back to the Mac this is exactly why Apple is probably not doing a die about as large as these on 5nm any time soon. Something relatively incrementally bigger than the A14 but not quite this big. The completed transition of all of the Macs over to Apple Silicon is probably years rather than a handful of months.
The farther away from entry level Mac laptops ( which is higher than mainstream PCs ) the longer it will probably take.
It moved from 7nm to 5nm . Very simply it could boost clock significantly higher just on the fab improvements.
Likely also got a bump in cache size levels due to more "room" for incrementally bigger caches. Again almost all comes along for the ride with the process bump.
There may be other stuff, but they can get more than a decent bump just from those two. It probably isn't just one magic bullet or some radical new invention.
Apple is calling out out more so to reinforce the notion that " Nobody in smartphone space is going to catch us this iteration". They moved again. The more telling iteration will be when they next do a iteration and there is no new fab process to iterate on. Apple is probably running out of 'low hanging fruit' because not constrained on die sizes limits.
Metal API has a new constant for 7th generation Apple GPU as of the latest API update (which most certainly refers to the A14), and it seems that A14 GPU supports at least one feature (barycentric fragment coordinates) that was previously only restricted to Intel and AMD GPUs. But as Apple didn't update the Metal feature tables or provided any new information about this new GPU, we'll have to wait and see. They usually release short tech notes as new hardware comes up, so I'd expect the information to be released before the iPad becomes available.