New cores (A15)

[deconstruct60]

The voltage is determined by the physics of the materials being used and the size of the transistors.

The size of the transistors and what the fab companies are labeling them with has gone way past diverging.
TSMC's N5P has a '5' in its name but it isn't exactly '5' either.

"... In the coming weeks TSMC is set to start making chips using a performance-enhanced version of its N5 technology called N5P that promises to increase frequencies by up to 5% or reduce power consumption by up to 10% (at the same complexity). The technology offers a seamless migration path for customers without requiring significant engineering resource investment or longer design cycle time, so anyone with an N5 design can use N5P instead. For example, early adopters of N5 could re-use their IP for their N5P chips. ..."

TSMC Update: 2nm in Development, 3nm and 4nm on Track for 2022

www.anandtech.com

Since TSMC 3nm sliding to a too late to use for Fall iPhone update it would make sense for Apple to use N5P and then N4 for next year. Apple used TSMC N7P to help 'bump' the A13 past the N7 A12.
Getting the high amount of design reuse either though materials and transistors were optimized a bit helps maintain this yearly cadence. There is fine line TSMC and its users are walking where it has changed , but hasn't changed at the same time on these incremental updates.

 

[Serban55]

I like that. 32 cores would mean 104 000, on par with Radeon Pro Vega II.

are you sure? i think ive heard around 78.000 score of sustain performance, with peak of 86k
Lets say, in the worse case scenario, the 32 cores it has this 78k score...this score is on par with what Radeon/nvidia gpu?

 

[sunny5]

iPhone14,3 - Geekbench

Benchmark results for an iPhone14,3 with an ARM processor.

browser.geekbench.com

Faster than A14.

 

[Homy]

are you sure? i think ive heard around 78.000 score of sustain performance, with peak of 86k
Lets say, in the worse case scenario, the 32 cores it has this 78k score...this score is on par with what Radeon/nvidia gpu?

I'm sure. The question is if Geekbench is sure.

Metal Benchmarks - Geekbench

78K-86K is right between Radeon Pro W5700XT and Pro VII.

 

[smulji]

Edit.

 

[crazy dave]

The voltage is determined by the physics of the materials being used and the size of the transistors. It likely hasn’t changed. If you lower it too much you can’t turn off the transistors (the equation I gave is for dynamic power - the power due to work being done to process. There is also static power, caused by leaky transistors that don’t shut all the way off, that is independent of work being done). Lowering the voltage also can make the chip unreliable, by causing signal transitions to slow down - that makes them susceptible to cross-coupling to neighboring signals, and can cause wires to have the wrong value at the wrong time and break the chip.


So, the voltage doesn’t change very much except over the course of decades.

So if I’m understanding you right (and I may not be) this sounds rather dispiriting from a cpu perspective: while improvements to MT look okay, ST PPW in avalanche basically didn’t improve one jot from firestorm despite a new architecture and even worse a new node. Even if charitably only a half node upgrade over its predecessor, N5P is still supposedly better and alone, if nothing else had changed, firestorm cores manufactured on it should’ve had a better PPW than firestorm cores on the previous N5 nodes the same clock speed due to the slightly smaller core size/better lithography. Avalanche cores may not. If this logic is true, and I’m sure I must be wrong somewhere, that would make avalanche a worse core than firestorm. Which doesn’t make sense to me, so where have I gone wrong? ?

EDIT: instead surely avalanche cores must have something to recommend them right? If not voltage the only other part of your equation where there is any wiggle room is area, correct? Surely N5P would give them some advantage over N5 intrinsically in addition to any small architectural improvements? Or why not just put firestorm cores with the doubled SLC?

 

[cmaier]

So if I’m understanding you right (and I may not be) this sounds rather dispiriting from a cpu perspective: while improvements to MT look okay, ST PPW in avalanche basically didn’t improve one jot from firestorm despite a new architecture and even worse a new node. Even if charitably only a half node upgrade over its predecessor, N5P is still supposedly better and alone, if nothing else had changed, firestorm cores manufactured on it should’ve had a better PPW than firestorm cores on the previous N5 node. Avalanche cores don’t. If this logic is true, and I’m sure I must be wrong somewhere, that would make avalanche a worse core than firestorm. Which doesn’t make sense to me, so where have I gone wrong? ?

Well, i don’t think it’s all that bad. Not every new CPU has to improve performance solely by increasing computational IPC. There are two ways, at least, that the CPU portion of A15 appears to be an improvement.

1) It can maintain high IPC for longer (and thus have an effectively higher IPC) because the memory bandwidth appears to have been increased by a lot. (Due to memory controller redesign, LPDDR5, and doubling the size of the SLC, and maybe more). This makes the chip faster.

2) It probably has more clock speed headroom. That is, in a mac, we might see higher clock speeds for M2 than M1. The higher A15 vs. A14 clock speed is one indication of this, but the increased memory bandwidth likely allows the clock to go even higher (the low bandwidth of the M1 means that increasing clock speed of M1 wouldn’t do you much good, because you’d be sitting around waiting for memory accesses a lot). This makes the chip faster.

3) the leakage current probably decreased, meaning PPW probably improved slightly. Keeping in mind that most power is dynamic, not static, decreasing static Power dissipation/consumption still is a good thing. (And also helps increase clock speed, because you don’t have as much of the thermal dissipation capacity taken up by leakage current-generated heat).

 

[crazy dave]

Well, i don’t think it’s all that bad. Not every new CPU has to improve performance solely by increasing computational IPC. There are two ways, at least, that the CPU portion of A15 appears to be an improvement.

1) It can maintain high IPC for longer (and thus have an effectively higher IPC) because the memory bandwidth appears to have been increased by a lot. (Due to memory controller redesign, LPDDR5, and doubling the size of the SLC, and maybe more). This makes the chip faster.

2) It probably has more clock speed headroom. That is, in a mac, we might see higher clock speeds for M2 than M1. The higher A15 vs. A14 clock speed is one indication of this, but the increased memory bandwidth likely allows the clock to go even higher (the low bandwidth of the M1 means that increasing clock speed of M1 wouldn’t do you much good, because you’d be sitting around waiting for memory accesses a lot). This makes the chip faster.

3) the leakage current probably decreased, meaning PPW probably improved slightly. Keeping in mind that most power is dynamic, not static, decreasing static Power dissipation/consumption still is a good thing. (And also helps increase clock speed, because you don’t have as much of the thermal dissipation capacity taken up by leakage current-generated heat).

Sure 1) I get - especially for MT I bet.

2) this appears to be a direct follow on from point 1 - ie the reason why the increased bandwidth should allow ST to improve

3) This was what I was trying to get at in my edit the new node itself is supposed to get a little better PPW.

But it would mean 0 ST IPC improvements which that I can’t disagree with - the scores are the scores and the clocks are the clocks. It clearly has the same IPC as firestorm. I guess I was just holding out hope that maybe there was still a rabbit in the hat and PPW could be say be 10% better beyond what the node naturally gives them and you’re saying that’s unlikely given physics. Pesky physics.

 

[cmaier]

Sure 1) I get - especially for MT I bet.

2) this appears to be a direct follow on from point 1 - ie the reason why the increased bandwidth should allow ST to improve

3) This was what I was trying to get at in my edit the new node itself is supposed to get a little better PPW.

But it would mean 0 ST IPC improvements which that I can’t disagree with - the scores are the scores and the clocks are the clocks. It clearly has the same IPC as firestorm. I guess I was just holding out hope that maybe there was still a rabbit in the hat and PPW could be say be 10% better beyond what the node naturally gives them and you’re saying that’s unlikely given physics. Pesky physics.

I’m sure that even if you were to rip the firestorms out of M1 and replace them with avalanche and change nothing else, you would see an IPC improvement in certain circumstances. I’m sure that avalanche is not identical to firestorm, and I’m reasonably sure the changes are not entirely limited to critical path timing improvements. These changes might just not be reflected in the benchmarks we’ve seen so far. That said, all that really matters is if the overall chip is faster in actual use, and whether most of the IPC improvement comes from rejiggering ALUs and instruction scheduling logic or whether it comes from better TLBs, caches, and the MMC, doesn’t really matter to the user.

 

[Sydde]

To replace the Intel CPUs in the higher end Macs (in particular the desktop Macs), Apple is going to have to invest a lot more time and money.

Wait, what desktops? There are 2 notebooks, Air and Pro. Then there are 3 desktops, Mini, iMac and Pro. The Mini and the new iMac have already gone AS. Intel versions of iMac and Mini are still available from Apple, alongside the Mac Pro, but Pro is the only one that does not have an AS version. It is not even clear whether they will introduce a larger iMac any time soon, so Pro looks like the only "desktop" that we are waiting on. I am expecting a different form factor, that you cannot grate cheese with.

 

[reallynotnick]

Wait, what desktops? There are 2 notebooks, Air and Pro. Then there are 3 desktops, Mini, iMac and Pro. The Mini and the new iMac have already gone AS. Intel versions of iMac and Mini are still available from Apple, alongside the Mac Pro, but Pro is the only one that does not have an AS version. It is not even clear whether they will introduce a larger iMac any time soon, so Pro looks like the only "desktop" that we are waiting on. I am expecting a different form factor, that you cannot grate cheese with.

Those Intel versions of the iMac and Mac Mini are literally the higher end versions. That's like saying Apple has transitioned the MacBook Pro to AS since there is already one model.

 

[Sydde]

Those Intel versions of the iMac and Mac Mini are literally the higher end versions.

Except, if you go to 27" iMac on the Apple site and click on tech specs, it looks like you can still get a 21.5" 2.3->3.6Ghz dual core i5, which can arrive at my door tomorrow. Is that higher end?

 

[Joelist]

I fully expect we will have an Apple Event probably next month for MacBook Pros and likely the iMac Pro. The SOC will either be "M1X" (Firestorm based) or "M2X" (Avalanche Based) - either way there will be more cores in both the CPU and GPU sections among other things.

 

[cmaier]

I fully expect we will have an Apple Event probably next month for MacBook Pros and likely the iMac Pro. The SOC will either be "M1X" (Firestorm based) or "M2X" (Avalanche Based) - either way there will be more cores in both the CPU and GPU sections among other things.

Since what we all expected in these new macs was additional memory bandwidth, and since avalanche and firestorm are pretty similar in performance, seems like it no longer matters. If we get firestorm with the new MMU, SLC and ram, it probably would run very very close in performance to the same chip with avalanche.

 

[quarkysg]

I fully expect we will have an Apple Event probably next month for MacBook Pros and likely the iMac Pro.

I would think next month's event will introduce the AS 27" iMac replacement, together with the MacBook Pros. If there's going to be an AS iMac Pro, it will likely be announced together wit the the AS Mac Pro.

 

[Serban55]

I fully expect we will have an Apple Event probably next month for MacBook Pros and likely the iMac Pro

forget about the bigger imac...that comes,probably, in the next spring event along side with the new Macbook air
Next month just hope for 3 macs (14" Mbp, 16" Mbp and mac mini)

 

[Pressure]

Considering that single M1 core can read at 58GB/s, write at ~33GB/s (source: Anandtech) Apple need to add more memory bandwidth. Considering more GPU clusters,more CPU cores, they need to add a lot of bandwith. Seems to me like they are going to make similar CPU like in Xbox/PS5 with much less heat.

That's the reason Apple invest a lot in cache. The M1 big core cluster has access to a private 128KB L1D cache, 12MB L2 cache, 16MB system-level cache (shared by entire system, so CPU clusters, GPU, Neural Engine etc) before reaching RAM.

They improved the cache again in the new cores, for example doubling the system-level cache to 32MB.

 

[Pressure]

Also, during the event they boasted about A15's ability to hardware encode/decode ProRes video. A feature that could put the Mac Pro's afterburner-card to shame and I definitely expect it in the next M-Soc for macs. So that's another reason I think they'll have A15-based chips.

The new Apple Silicon SoC will surely use the Avalanche and Blizzard cores used in the A15 SoC.

Think of them as building blocks, so they will definitely reuse a lot of the blocks, especially that nice video hardware decode and encode block with support for ProRes.

I wouldn't assume it handles ProRes better than the AfterBurner card for all ProRes formats and resolutions.

 

[MrGunny94]

I finally did some reading on the A15, and I wasn't suprised by the cache increase! I'm glad they are going this route.

 

[dugbug]

I wonder if they will offer the 24” iMac with an m2(x) along with the new MacBook pros. They already come with a fan in the higher end one

 

[Serban55]

I wonder if they will offer the 24” iMac with an m2(x) along with the new MacBook pros. They already come with a fan in the higher end one

24" base has 1 fan already, and the 8 gpu cores has 2 fans..so even the base one has a fan
The next 24" imac , probably next year will come with the M2 or what it will be called the SoC from the next generation Macbook air. The more complex SoC it is reserved for the bigger imac i suppose

 

[deconstruct60]

forget about the bigger imac...that comes,probably, in the next spring event along side with the new Macbook air
Next month just hope for 3 macs (14" Mbp, 16" Mbp and mac mini)

The bigger iMac is probably "Winter" more than "Spring". At this point the iMac may be held up on wafer shortage. For example if it takes 2 dies while the MBP 14" and 16" take own die then Apple can sell twice as many Macs with a pause on iMacs to ship more MBP.

When the initial demand bubble subsides for the iPhone 13 and new MBP they should be more wafers to "double up" on the bigger iMac. The bubble would trail off faster if the new chip is on N5 (which the iPhones are exiting) rather than new wave (N5P). However, the base supply of wafers of both is constrained.

Needing an additional year to construct the design of the larger iMac after finishing off the smaller screen iMac in early 2021 seems excessive . One rumor said Apple had pulled resources off of the bigger iMac industrial design to get the smaller iMac out the door. Once the smaller one shipped they should have been able to catch up by applying all the resources loaded up on smaller iMac back to the bigger one. By this December it seems more likley that the bigger iMac will be more constrained on parts availability, rather than "Apple still trying to figure out what to do" issues.

Apple could drag their feet so that the bigger iMac waits to come out alongside a refreshed M2 smaller screen iMac. Competitive wise it doesn't make much sense. The iMac is squatting on Gen10 Intel CPUs (that were released in Q2 '20 while Apple released later ). By January, Intel and AMD desktops CPUs are going pragmatically two generations beyond that. Desktop Intel will be on Gen 12 (Alder Lake) and AMD will have their 3D L3 cache augments out.

Even worse would be a MBP 16" SoC powered Mini (e.g., 10 CPU - 32 GPU Mini ) out there and the iMac was still stuck back into the 2020 on Intel. if the Mini gets bumped up in horsepower then the larger iMac should be bumped up also. Both of those can slide just past the "pain" the holiday demand to create product before the end of the year is causing.

Waiting for the smaller screen iMac to get the M2 and release the iMacs as a super tightly coupled pair wouldn't make much sense . It would be something else if the current iMac was competitive, but it is loosing traction pretty fast. ( there are fewer and fewer x86_64 macs being sold. Once all the laptops switch over to M-series, the Mac product space is solidly on the M-series in both volume and new software development. )

 

[deconstruct60]

I wonder if they will offer the 24” iMac with an m2(x) along with the new MacBook pros. They already come with a fan in the higher end one

the MBA , more affordable MBP 13, and Mini are far more likely candidates for M2's than the 24" iMac. Apple has shrunk the logic board in the 24" iMac so smaller it is unlikely that a "X" ( substantively larger die and 2-3 more RAM stacks ) would fit. The constraint that the principle logic board has to entirely fit into about half of the chin is a dual edge sword. Apple gets to radically thin out the thickness , but also highly limiting both thermals and logic board space.

The M2 arrival will pretty depend upon how successful the iPhone 13 (and iPad Mini ) are in the market this Fall. If Apple can't make enough A15's to satisfy demand, then they probably would compound that problem with making more bigger die M2. That would be even less wafers available for the iPhone ( Apple can make more A15s off the same wafer than they could M2's ).

Even worse if the MBP 14" and 15" are soaking up possible M2 wafers and Apple is using them to fill the gaps around the A15 production. There will be even fewer "leftovers" for a couple of quarters.

I highly doubt Apple wants to get folks hooked on some exactly 12 month cadence for Mac updates that they are hooked to with iPhones. That is actually not really helpful for them. Every aspect of new fab tech doesn't come out only in May-July every year.

 

[Boil]

I highly doubt Apple wants to get folks hooked on some exactly 12 month cadence for Mac updates that they are hooked to with iPhones. That is actually not really helpful for them. Every aspect of new fab tech doesn't come out only in May-July every year.

I certainly hope not, a new computer every year may be fine for business / corporate users, but for the average Joe consumer a new computer every three to five years is reasonable...?

 

[ADGrant]

Wait, what desktops? There are 2 notebooks, Air and Pro. Then there are 3 desktops, Mini, iMac and Pro. The Mini and the new iMac have already gone AS. Intel versions of iMac and Mini are still available from Apple, alongside the Mac Pro, but Pro is the only one that does not have an AS version. It is not even clear whether they will introduce a larger iMac any time soon, so Pro looks like the only "desktop" that we are waiting on. I am expecting a different form factor, that you cannot grate cheese with.

There are four notebooks right now, MBA, M1 MBP 13", Intel MBP 13" and Intel MBP 16". The M1 models replaced the Intel MBA and the low end two port single fan 13" Intel MBP. I expect the high end 13" MBP and 16" MBP will be replaced next. I have no idea what they will call the enhanced SoC but it will need to support more RAM and have a more powerful GPU.

The 24" M1 iMac replaced the 21" Intel iMac. It is in no way a replacement for the current 27' iMac which (aside from the 5k screen) can be configured with up to 128GB of RAM and a CPU and GPU that are faster than the ones in the base Mac Pro model. We may not see a larger Apple Silicon iMac this year but I think we will see one before the transition is completed.

The Mac Pro of course can be configured with much more RAM (1.5 TB), a 28 core CPU and dual 6800X video cards with 64GB of video RAM each. It will be interesting to see what Apple offers as a replacement.

Most people also expect higher end Apple Silicon Mac minis. The Intel Mac minis are extremely weak in the GPU department and don't have particularly fast Intel CPUs but they can be configured with a decent amount of RAM and have decent I/O. The current mini chassis could certainly house a much more powerful Apple Silicon SoC than the M1.