Question Apple A15 announced

[jpiniero]

Seems like the CPU or GPU is not much faster than A14. Might be due to wanting lower power draw for better battery life.

The NPU did get a bump, 15.8 vs 11.8 tops.

Edit: The Pro does get an increase to 5 GPU cores from 4. Might be useful because of the 120 Hz VRR they added.

 

[jpiniero]

It is unclear whether N3 will be available in time for A16 / iPhone 14 a year from now

It won't.

There's usually an Apple event in October that focuses on their new Mac lineup, so we'll probably see any new products announced there.

Yeah. Still monolthic and N5P. Just more CPU and GPU cores, some clock tweaks. This is mostly aimed at the high end MBP. Perhaps the Mini and iMac will get the M1X as an option too.

 

[jpiniero]

iPhone 12 Mini vs iPad14,2 - Geekbench

The iPad Mini has the clock speed down to the same as the iPhone 12. So ST performance is virtually identical. The MT score is 13% faster.

 

[Mopetar]

Honestly an 8/8 core version of the M1 would be pretty beastly. The real question will be what they try to do with graphics. I don't know if they can get away from AMD at the high-end just quite yet and I doubt they want to build an SoC large enough to be able to do that.

 

[jpiniero]

Honestly an 8/8 core version of the M1 would be pretty beastly. The real question will be what they try to do with graphics. I don't know if they can get away from AMD at the high-end just quite yet and I doubt they want to build an SoC large enough to be able to do that.

I'd be quite surprised if they included a discrete GPU. Apple's high end MBP have typically used modest dGPUs so the bar is not as high as you think. The die would start to get pretty big if they say doubled the CPU and GPU cores.

 

[Doug S]

I'd be quite surprised if they included a discrete GPU. Apple's high end MBP have typically used modest dGPUs so the bar is not as high as you think. The die would start to get pretty big if they say doubled the CPU and GPU cores.

Well the Jade-C rumors are for 8+2 cores, not 8+8. That makes much more sense, especially in higher end systems with multiple Jade-C chiplets.

I think they may do more than doubling the GPU cores for the Mac Pro to match AMD/NVidia GPU performance levels. I did some back of the envelope calculations a while back with guesses about what Jade-C would include (I guessed 32 GPU cores) and it ended up around 275 mm^2.

That's a bigger die to be sure but given that everything it went in would sell for over $1000, that's hardly an issue.

 

[defferoo]

iPhone 12 Mini vs iPad14,2 - Geekbench

The iPad Mini has the clock speed down to the same as the iPhone 12. So ST performance is virtually identical. The MT score is 13% faster.

It seems like they didn't improve the P-cores at all, but improved MT either by improving the E-cores or by improving multi-threaded efficiency (potentially the doubled system level cache).

My guess is that there's so much work to do on the new Mac chips that they decided to focus on that instead of improving the CPU cores. The P-cores are already the fastest/highest IPC cores around, so they figured it would be okay for it to leave it as is this year.

 

[Ajay]

It won't.

TSMC N3 at HVM Q2 2022. Apple will happily buy up all the available wafers during the ramp up. They have done so for years. OFC, the rumor is that the next Apple CPU will use N4 - what ever that is.

Also, supply chains should be pretty close to 100% next year (barring variant xyz that eats your brain).

 

[jpiniero]

TSMC N3 at HVM Q2 2022. Apple will happily buy up all the available wafers during the ramp up. They have done so for years. OFC, the rumor is that the next Apple CPU will use N4 - what ever that is.

TSMC said second half. Too late. The iPhone 2022 is using N4.

 

[jpiniero]

I think they may do more than doubling the GPU cores for the Mac Pro to match AMD/NVidia GPU performance levels. I did some back of the envelope calculations a while back with guesses about what Jade-C would include (I guessed 32 GPU cores) and it ended up around 275 mm^2.

32 is way too much. Like I said, the comparison is with the 5600M, not with something like the 100+ W 3080 mobile.

 

[Ajay]

TSMC said second half. Too late. The iPhone 2022 is using N4.

Okay, missed that. I'll have to hunt down some info on N4, probably minor improvements in frequency vs power. Perhaps slightly denser.

 

[eek2121]

Apple is using N4 for the next iPhone.

The exact clockspeed of the A15 for iPhone Pro/Pro Max is 3233 Mhz.

There appears to be minor efficiency (perf/watt), but no IPC improvements.

Ordering mine tomorrow as soon as preorders open. The video recording stuff may simplify a lot of the work I do for one website.

 

[Doug S]

32 is way too much. Like I said, the comparison is with the 5600M, not with something like the 100+ W 3080 mobile.

Why? They don't have to enable them all in devices where power is limited like Macbook Pro. They need to have enough GPU cores for the Mac Pro to be competitive with AMD/NVidia GPUs, otherwise Apple will catch a lot of criticism for not supporting third party GPUs in a $5000+ workstation. I suppose they could do 16 cores if support 8 chiplet / 64 big cores as the Mac Pro high end.

There are possibly other ways like improving the GPU core, clocking it higher, giving it faster RAM etc. but whatever they do it WILL be competitive with third party GPUs. The fact they waited until they could ship ARM Macs that were faster than the Intel alternatives clearly shows that as their strategy. If being "sort of close" was good enough for them, the ARM Mac transition would have been completed several years ago.

 

[jpiniero]

Why? They don't have to enable them all in devices where power is limited like Macbook Pro.

The M1X's main focus is the MBP, I'm assuming.

I'm not expecting a real Mac Pro replacement for some time. They might release a product branded Mac Pro soonish but it'd be more like a beefed up Mini. Apple still hasn't released the Icelake Mac Pro yet so there's plenty of time.

 

[nxre]

I saw some estimates that the A15 die was 110mm2 which is their first one to go over 100mm2 since A10. A14 is 88mm2 and M1 is 120mm2 for comparison. A15 seems like an abnormally large die for mobile

 

[ashFTW]

Why? They don't have to enable them all in devices where power is limited like Macbook Pro. They need to have enough GPU cores for the Mac Pro to be competitive with AMD/NVidia GPUs, otherwise Apple will catch a lot of criticism for not supporting third party GPUs in a $5000+ workstation. I suppose they could do 16 cores if support 8 chiplet / 64 big cores as the Mac Pro high end.

I remember there being separate GPU dies rumored for Mac Pro. So not all GPU horsepower has to come from the jade chiplets.

 

[Mopetar]

I'd be quite surprised if they included a discrete GPU. Apple's high end MBP have typically used modest dGPUs so the bar is not as high as you think. The die would start to get pretty big if they say doubled the CPU and GPU cores.

The M1 is only 120 mm^2 and if you look at a die shot, a lot of that space is dedicated hardware for decoders, etc. as opposed to to CPU/GPU cores. Even if they did double both, it would still be smaller than 200 mm^2. Frankly, them doubling the GPU cores might even be enough for their pro notebooks. I would imagine they might add a discrete card to the desktop, just because there are going to be some users who need it and I don't know if Apple has their GPUs on quite the same level as AMD/Nvidia, at least in terms of support for some of the professional software that uses Quadro cards and the like.

 

[jpiniero]

I would imagine they might add a discrete card to the desktop, just because there are going to be some users who need it and I don't know if Apple has their GPUs on quite the same level as AMD/Nvidia, at least in terms of support for some of the professional software that uses Quadro cards and the like.

Apple only officially supports Metal. There are third party compatibility layers for Vulkan but I'd have to think any Pro App would have to use Metal and nothing else.

 

[eek2121]

Apple only officially supports Metal. There are third party compatibility layers for Vulkan but I'd have to think any Pro App would have to use Metal and nothing else.

Apple has also made every indication they are going with their own GPU solution going forward. Not that discrete GPUs won't be an option, but I imagine those discrete GPUs will have an Apple logo on them instead of AMD.

EDIT: Also, iPhone preordered.

 

[Doug S]

Apple has also made every indication they are going with their own GPU solution going forward. Not that discrete GPUs won't be an option, but I imagine those discrete GPUs will have an Apple logo on them instead of AMD.

EDIT: Also, iPhone preordered.

Not only have they basically told us they are going with their own GPU all the way through the Mac line, going with their own likely implies the market for third party cards (even if sufficiently wide PCIe slots were available at all) would be far too small for AMD and Nvidia to bother with.

I mean Nvidia ALREADY ignores the Mac Pro market, because Apple sells/supports AMD cards in the x86 Mac Pro. So there's obviously no chance they'll jump in on the new Mac Pro that comes with an Apple solution instead of an AMD solution. And if Nvidia doesn't think it is worth bothering when Apple sells Macs with something else, why would AMD not reach the same conclusion? Both have their hands full selling all the cards they can make to miners and the HPC crowd. Introduce a Mac card and the next day someone will be hacking it to figure out how to plug it into a cheap PC and mine crypto.

I have said before don't be too terribly shocked if the ARM Mac Pro does not include PCIe slots at all. I think they will include some 4x slots but I feel that's far from assured. If the GPU is integrated, the only other things you need for PCIe are add in network and storage, and unless 100GbE is a must-have for segments of the Mac Pro market (I have no idea about that, and if so they might offer an optional 100GbE fiber port like how the Mac Mini has an optional 10GbaseT port) any network < 100 Gbps or fibre channel external arrays can be connected via Thunderbolt.

 

[jpiniero]

I mean Nvidia ALREADY ignores the Mac Pro market

That is because Apple banned them. AMD's probably not interested in writing drivers without an OEM deal.

There is AFAIK no eGPU support either in ARM Macs.

 

[moinmoin]

M1 already appears to support PCIe as part of Thunderbolt's PCIe passthrough. But aside for detection it appears to be unusable so far, though this seems to be a software/driver level problem, not a hardware one.

 

[Ajay]

It’s a bit hard for me to see why AMD would even bother to do the work necessary to recompile and then test ARM compatible drivers for the relatively small Apple Mac market. Unless…..Apple drives to AMD HQ with a van stuffed with suitcases filled with bundles of $100 bills.

 

[eek2121]

If Apple GPUs become decent, they'll sell tons of Macs to people wanting to mine crypto.

 

[Mopetar]

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.

 

[fusionTi]

If AMD were still on Polaris/Vega GCN4 or a derivative, it would be a no brainer for Apple to ditch them. RDNA and RDNA2 have probably been enough progress to buy AMD some more time on Apple's platform.