ARM MacBook Pro I don’t want an Overpriced Netbook.

[bill-p]

I was comparing it to the 28W TDP Ice Lake in higher tier 13” MBP. Not the MBA - the iPad Pro runs circles around the MBA.

Because the Air is limited to <10W of thermal dissipation. It's not able to properly allow max sustained performance even with the iPad Pro's A12X.

This is not my statement. Refer to A12 and A13 deep dive reviews by Anandtech and others. Funny fact: a year ago I was the same opinion as you. ARM Macs? Give me a break, why does one want a dead slow toy? Then I actually looked at the data and I was shocked. Apple CPUs are an engineering marvel.

Fun fact: I thought ARM Macs are not as bad as well, until I tried to simulate the behavior of virtual memory and running multiple different processes on my iPad Pro. It's not pretty. This is from real life experience, not from "data".

And then going through the "data" just shows that Apple has come a long way, but they are not "ready" yet to replace the absolute performance offered in the 16" MacBook Pro, the iMac, and the Mac Pro. That's what I've been saying all along.

The 16” MBP debilere around 60Watt of sustained CPU performance, which allows the i9 to run above its spec. This allows the CPU to maintain clocks of around 3.2-3.4 on all cores if I remember correctly (been a while since I tested it). On single core max turbo boost it consumes around 25 watts. Yes, the GPU makes it more complicated, as the laptop in total offers around 80Watts of power delivery that needs to be shared between CPU and GPU. Note that all i9-equipped laptops suffer from this.

Use a multimeter or iStat. It's not 60W. It's 85-90W required to get 3.2GHz. Don't use Intel Power Gadget because it under-reports the wattage in an effort to show that Intel was not "lying" about their power consumption. The Core i9 9980H requires 120W for max turbo boost. There's a reason why workstation-grade laptops on the Windows side have started to ship with 130W power adapters.

Which again illustrates the problems Intel is having. You need to run a 45W CPU at 60W to get a performance gain. These chips are clocked way above their comfort point so that Intel can claim performance advances. To get most out of those chips you need to run them on a desktop.

Yes, and I never said Intel did not have a power efficiency problem. It makes sense. Their chips are still mostly on 14nm after all.

People say Intel's 14nm should be "equivalent" to 10nm elsewhere (TSMC or Global Foundries), but I think that's a flat out LIE in order to save face. Intel's CEO is the one who started this lie. It's pretty clear that Intel's 14nm is equivalent to 14nm from TSMC from a performance per watt standpoint. At least based on hard benchmark numbers, and not on marketing terms.

Why would it be a fair comparison? The i9 has 8 cores (the A12Z has 4, the low performance cores barely count), and the iPad can do substantially more work per clock - it has 50% more execution units per core than Intel CPU.

The i9 has 2.3GHz base clock. That's closer to the A12Z. Core count will only affect multi-core benchmark scores. They won't affect single-core. You know that much. If not the i9, then the more "apples to apples" would be the BTO 13" 2020 with the Core i7 1060NG7.

Since you were curious, I did some quick tests while sipping my morning tea... [shortened the rest]

Just one note, as stated above: Intel Power Gadget under-reports power consumption. The actual power consumption is typically 20 - 30W above what it's saying. This can be verified with a multimeter measuring power draw from the outlet while running a CPU-only task. I'd use iStat Menu.

But anyways, your results don't surprise me because:

1. The A12Z is still a 8-core CPU in the end. You may think the other 4 cores are not working during benchmarks, but only Apple would know the details. From an engineering standpoint, it makes no sense to only make use of 4 cores on an 8-core machine, hence my assessment that the comparison is only fair when it's the Core i9. Note that Geekbench running on the DTK says "4 cores" but running it as a native iOS app makes it say "8 cores". I don't think that's a coincidence. When the A12Z in DTK is running in "4 cores" mode, it's getting a score closer to what would happen when you limit the Core i9 9980H to the same clock speed and core count:

eperm-d995af6e2ef02771 - Geekbench 5 CPU Search - Geekbench

2. The Core i9 is still on 14nm. The A12Z is on 7nm. Power efficiency and performance numbers are indeed consistent with the fact that Intel is 2 generations behind. That may mean Apple is well ahead, but it also means Intel has 2 generations to grow from here.

And that's been my point: I'm not counting my chickens yet because we have no idea how well Apple's architecture will scale. It may look amazing here, but this may be the most optimal performance/watt configuration of the A12, so we are seeing all of the benefits and none of the drawbacks. It remains to be seen just how well Apple can make their chips scale, but... I'll say this once again: it cannot possibly be linear scaling. There is diminutive return at some point. You and many others are acting like it's going to be linear. I don't think it is.

I remain skeptical because despite being 2 generations ahead (or sure, 1 generation if I have to take wikichip at face value), Apple's lead is not truly that large. Intel is barely on 10nm. TSMC can be said to have been on 7nm+ by the time they made the A12Z. Granted, you can argue that the process node number means nothing, but the proof is in the data: Intel's numbers are more consistent with 14nm made by TSMC.

But beside all of that, my original point remains: Apple is not yet ready to produce chips that will fit the 16" MacBook Pro, the iMac, or the Mac Pro.

I can concede that they have chips ready for the MacBook Air and the 13" Pro, based on how scaling would go with another process node improvement (supposedly, 5nm from TSMC should help), but the higher-end Macs may still have to wait for a while.

See, I'm not praising Intel. They have a lot of catch-up to do at this point... being, in my eyes at least, 2 generations behind the competition. But hard data and real life experience with my own iPad Pro is telling me I shouldn't get too excited just yet.

I'll be excited about battery life, but there are a lot of cases where absolute performance (no thermal or power draw constraint) is necessary in my workflow. I don't want a long lasting device that also takes longer to finish tasks.

 

[JMacHack]

And then going through the "data" just shows that Apple has come a long way, but they are not "ready" yet to replace the absolute performance offered in the 16" MacBook Pro, the iMac, and the Mac Pro. That's what I've been saying all along.

In 2020, you're absolutely right.
But by 2022, I think they'll be able to deliver that kind of performance.

 

[leman]

Fun fact: I thought ARM Macs are not as bad as well, until I tried to simulate the behavior of virtual memory and running multiple different processes on my iPad Pro. It's not pretty. This is from real life experience, not from "data".

What do you mean by "virtual memory" in this context?

Use a multimeter or iStat. It's not 60W. It's 85-90W required to get 3.2GHz. Don't use Intel Power Gadget because it under-reports the wattage in an effort to show that Intel was not "lying" about their power consumption.

That's interesting, thanks for the hint. I will check it out.

1. The A12Z is still a 8-core CPU in the end. You may think the other 4 cores are not working during benchmarks, but only Apple would know the details.

Fair enough, but the efficiency core is only around 1/4 as fast as the performance core. Even if we assume that they give the A12Z a 25% boost (which I am skeptical about), the performance advantage of the Apple architecture is still undeniable in this particular test.

The Core i9 is still on 14nm. The A12Z is on 7nm. Power efficiency and performance numbers are indeed consistent with the fact that Intel is 2 generations behind.

It's not like 10nm Intel is doing much better. They have what, 5-10% improvement in IPC? Don't forget that we are talking about A12Z here — that chip is also two generations behind. The A13 is significantly faster. And Apple is already producing 5nm CPUs... Sure, Intel 7nm is supposed to have better transistor density than the TSMC at 5nm, but when will those CPU go in production? 2022? Later?

 

[raknor]

Because the Air is limited to <10W of thermal dissipation. It's not able to properly allow max sustained performance even with the iPad Pro's A12X.

Not it is not. Notebookcheck's test show a sustained load of 25.7Ws for the i5.

Apple MacBook Air 2020 Core i5 Review – The best MacBook you can get? – No

Notebookcheck reviews the new Apple MacBook Air with the quad-core i5 CPU. The keyboard has been improved, but the performance is still the Achilles heel of the laptop.

www.notebookcheck.net

The i3 model can sustain 24.7 Ws in the same chassis.

Apple MacBook Air 2020 Review: Is the Core i3 the better choice?

The entry-level version of the new 2020 MacBook Air is in review. Is the Core i3 processor sufficient for your everyday tasks?

www.notebookcheck.net

Fun fact: I thought ARM Macs are not as bad as well, until I tried to simulate the behavior of virtual memory and running multiple different processes on my iPad Pro. It's not pretty. This is from real life experience, not from "data".

And then going through the "data" just shows that Apple has come a long way, but they are not "ready" yet to replace the absolute performance offered in the 16" MacBook Pro, the iMac, and the Mac Pro. That's what I've been saying all along.

What virtual memory behavior did you simulate? What multiple Processes did you simulate? I would like to try the exact same thing on my iPad Pro so I can verify your results. Please share your workload details.

You are comparing Virtual Memory performance on a device with a 8W-10W SOC and 4 GB RAM and comparing it to systems that at a minimum have 45W CPUS and 8GB-16GB RAM!

Use a multimeter or iStat. It's not 60W. It's 85-90W required to get 3.2GHz. Don't use Intel Power Gadget because it under-reports the wattage in an effort to show that Intel was not "lying" about their power consumption. The Core i9 9980H requires 120W for max turbo boost. There's a reason why workstation-grade laptops on the Windows side have started to ship with 130W power adapters.

That's pathetic for a laptop processor, a fact demonstrated by AMD outperforming it on a 4800U with less than half the power draw.

Yes, and I never said Intel did not have a power efficiency problem. It makes sense. Their chips are still mostly on 14nm after all.

Yes.. but even their 10nm chips aren't that great and they are similar in process to TSMC 7nm that AMD and Apple use.

People say Intel's 14nm should be "equivalent" to 10nm elsewhere (TSMC or Global Foundries), but I think that's a flat out LIE in order to save face. Intel's CEO is the one who started this lie. It's pretty clear that Intel's 14nm is equivalent to 14nm from TSMC from a performance per watt standpoint. At least based on hard benchmark numbers, and not on marketing terms.

So everyone is lying including Intel about process technology... just wow!

The i9 has 2.3GHz base clock. That's closer to the A12Z. Core count will only affect multi-core benchmark scores. They won't affect single-core. You know that much. If not the i9, then the more "apples to apples" would be the BTO 13" 2020 with the Core i7 1060NG7.

The A12Z performs about the same as that chip being 2 years older and while consuming much less power.

MacBook Pro (13-inch Mid 2020) Benchmarks - Geekbench

iPad Pro 12.9-inch (4th generation) Benchmarks - Geekbench

browser.geekbench.com

The A13 in the iPhone is also faster than the BTO Core i7 1068NG7 in single core.

But anyways, your results don't surprise me because:

1. The A12Z is still a 8-core CPU in the end. You may think the other 4 cores are not working during benchmarks, but only Apple would know the details. From an engineering standpoint, it makes no sense to only make use of 4 cores on an 8-core machine, hence my assessment that the comparison is only fair when it's the Core i9. Note that Geekbench running on the DTK says "4 cores" but running it as a native iOS app makes it say "8 cores". I don't think that's a coincidence. When the A12Z in DTK is running in "4 cores" mode, it's getting a score closer to what would happen when you limit the Core i9 9980H to the same clock speed and core count:

eperm-d995af6e2ef02771 - Geekbench 5 CPU Search - Geekbench

Yes, that shows how good the A12Z is! If you limit it to 4 cores running x86 code (translated) it is faster (Single:836, Multi:2884) than the Core i9 9980H (Single:636, Multi:2365) at the same clock speed and core count ! I'd love to see a geekbench run where the intel chip runs the ARM version translated.

2. The Core i9 is still on 14nm. The A12Z is on 7nm. Power efficiency and performance numbers are indeed consistent with the fact that Intel is 2 generations behind. That may mean Apple is well ahead, but it also means Intel has 2 generations to grow from here.

Intel has had 5+ years to get ahead.

I remain skeptical because despite being 2 generations ahead (or sure, 1 generation if I have to take wikichip at face value), Apple's lead is not truly that large. Intel is barely on 10nm.

Intel themselves are going to be fabbing some of their IP on TSMC's 7nm process. Their 7nm process is delayed. So if
you are dreaming that Intel will somehow magically catch up to TSMC.. keep on dreaming.

But beside all of that, my original point remains: Apple is not yet ready to produce chips that will fit the 16" MacBook Pro, the iMac, or the Mac Pro.

How did you come to that conclusion? How long do you think it will be before Apple will be ready?

 

[AxiomaticRubric]

I'll be excited about battery life, but there are a lot of cases where absolute performance (no thermal or power draw constraint) is necessary in my workflow. I don't want a long lasting device that also takes longer to finish tasks.

I think the key here is that Apple will be able to optimize their proprietary ARM SoCs specifcally for macOS, so comparisons to existing chips at this point may not reflect actual performance.

This is something other companies are currently not in a position to do. Nearly complete vertical integration should prove to be a huge competitive advantage.

 

[ADGrant]

Fun fact: I thought ARM Macs are not as bad as well, until I tried to simulate the behavior of virtual memory and running multiple different processes on my iPad Pro. It's not pretty. This is from real life experience, not from "data".

And then going through the "data" just shows that Apple has come a long way, but they are not "ready" yet to replace the absolute performance offered in the 16" MacBook Pro, the iMac, and the Mac Pro. That's what I've been saying all along.

How does running multiple different processes on an iPad in anyway simulate the behavior of virtual memory on a Mac. iPad OS doesn't support virtual memory in the MacOS sense, the amount of RAM is much lower than most Macs and the flash storage performance is not the same either.

 

[AxiomaticRubric]

It will take time, but apple is definitely pushing for game development, and their hardware is already plenty capable of it. Give it a few years, and I give it 50/50 chance that Apple will be a significant gaming platform.

It’s not about hardware. It’s about marketshare.

Apple has a minority share of both mobile and PC desktop markets. Development studios have comparitively little incentive to provide games for smaller segments of these markets.

This is why gaming on the Mac has been frustrating from the very beginning. The hardware is capable and now Metal is a great graphics API, but why invest any time in it (for example) if I can create something in OpenGL and provide it to multiple major gaming platforms?

 

[leman]

The hardware is capable and now Metal is a great graphics API, but why invest any time in it (for example) if I can create something in OpenGL and provide it to multiple major gaming platforms?

Because you don’t use Metal directly, you use an established engine such as Unity (with full Metal and ARM Mac support). Or you use Vulkan + MoltenVK. The point is that if gaming capability of Macs increase significantly l, studios will have more incentive to implement games on macOS. Apples market share might be low, but so is the market share of game-capable PCs. The bulk of hardware out there uses Intel integrated graphics or something like the MX150.

 

[ADGrant]

It’s not about hardware. It’s about marketshare.

Apple has a minority share of both mobile and PC desktop markets. Development studios have comparitively little incentive to provide games for smaller segments of these markets.

This is why gaming on the Mac has been frustrating from the very beginning. The hardware is capable and now Metal is a great graphics API, but why invest any time in it (for example) if I can create something in OpenGL and provide it to multiple major gaming platforms?

Does Apple really have a minority share of the mobile gaming market by revenue. The latest legal filings by Epic Games would suggest they have a pretty significant market share.

Which major gaming platforms are still Open GL platforms, Android? The PC is really a DirectX platform. Smaller publishers will be using someone else's game engine anyway.

 

[Maximara]

If Apple was confident about performance, they would not need 2 years to transition, nor would they need to introduce more Intel-based Macs. The fact is: Apple still has to introduce Intel-based Macs. For how long, we have no idea. But the fact is: Apple knows they cannot completely usurp Intel performance at the higher end of the spectrum. The iMac, the 16" MacBook and the Mac Pro will likely have to wait until TSMC can produce 5nm chips reliably.

I don't think it is so much confidence about performance but for developers to recode to take advantage of the new design. As seen with the Windows world you really don't want to be supporting old code until the Devil and his friends starting having snowball fights. Besides something along this two year cycle was seen with the 680x0 to PowerPC and PowerPC to Intel transitions so it is effectively SOP for Apple at this stage of the game.

 

[Joelist]

You can't do a straight compare of an iPad Pro to something with four times the RAM and a bigger thermal envelope. Really you need the Apple Silicon (NOT ARM) DTK as it at least has the correct amount of RAM. However, otherwise it is completely unoptimized for the MacOS platform so it is not going to yield numbers as high as an actual AS Mac. Also Geekbench at present has to use Rosetta 2 to run - even with this the DTK is outperforming the Mac Mini and the MacBook Air. So just picture it with proper optimizations and likely more cores and it is something to look forward to.

 

[bill-p]

What do you mean by "virtual memory" in this context?

As textbook as it sounds, actually: I wrote a simple app in Xcode that spawns multiple threads... each trying to determine the prime factorization of a number by looking up a list of prime numbers and already existing factorizations. So each thread is reading a "page" in memory and swapping things in and out from storage as needed.

You can guess that this blows out of memory very fast. In a sense, it's more a benchmark of storage access performance, plus maybe some branching, than of actual CPU performance.

Then again, the A12 is an SoC, so... I/O bandwidth is one thing that is embedded in there that should be tested anyways. Let's just say... I think the results are interesting, but the same task is of course a lot faster on the Core i9, even if I limited memory usage to 4GB.

Fair enough, but the efficiency core is only around 1/4 as fast as the performance core. Even if we assume that they give the A12Z a 25% boost (which I am skeptical about), the performance advantage of the Apple architecture is still undeniable in this particular test.

I think you're probably thinking of in-kernel switcher mode of operation:

ARM big.LITTLE - Wikipedia

en.wikipedia.org

But ARM's big.LITTLE also allows for this:

ARM big.LITTLE - Wikipedia

en.wikipedia.org

And it even states that Apple has started using heterogeneous multi processing since A11... so actually, I'm guessing that means that from A11, all cores are active, and not just the high-performing ones. That would make more sense why Geekbench scores are like that.

It's not like 10nm Intel is doing much better. They have what, 5-10% improvement in IPC? Don't forget that we are talking about A12Z here — that chip is also two generations behind. The A13 is significantly faster. And Apple is already producing 5nm CPUs... Sure, Intel 7nm is supposed to have better transistor density than the TSMC at 5nm, but when will those CPU go in production? 2022? Later?

Yeah, so I do agree Intel is behind. But the whole point of my argument is: because they are behind, they have more room to grow than Apple, who's fast approaching quantum limits... because you can only shrink your die so much. There may come a point (perhaps TSMC's 3nm?) that Apple may not be able to shrink their chips anymore. That can put a damper on innovations for a while.

But beside that point, there are talks of Intel moving to TSMC for manufacturing even their CPUs. So Intel is going to jump from 14nm, "barely" 10nm, straight up to 7nm or 5nm probably much sooner than we think.

Not it is not. Notebookcheck's test show a sustained load of 25.7Ws for the i5.

Here you go:

As for your other points, I shared the workload above. I'm fairly certain you can find a multi-threaded prime number factorization source code online or write your own to benchmark the iPad in similar condition.

Run the same thing on a MacBook Pro 13" or 16". I think it'll be interesting.

Yes.. but even their 10nm chips aren't that great and they are similar in process to TSMC 7nm that AMD and Apple use.

So everyone is lying including Intel about process technology... just wow!

Sorry, I can't respond to every point you are trying to make, but I think you're misunderstanding what I wrote. I'll break it down like this to make it clearer:

1. I don't disagree A12Z is a very efficient design from Apple. It's not the absolute best performance on the market, though. Case in point, I'm sure you can agree that it's not fair to compare against an 8-core Core i9 with 64GB of RAM. But hey, that's what many Apple Silicon supporters are hoping Apple can achieve. Personally, I think that's just a bit far-fetched, but it may just be me.

2. I'm saying Intel 10nm is not "similar" to TSMC 7nm. Performance and efficiency numbers of Intel 10nm are more akin to TSMC 10nm. So benchmarks will point to the process node numbers being "right". It's just Intel CEO and maybe wikichip saying Intel 10nm is actually equal to TSMC 7nm. Here, I think this discussion is closer to the truth (and the truth is basically closer to what I wrote above):

Is Intel 10nm really denser than TSMC 7nm?

In a recent article on Wikichip they claim that Intel's 10nm process is denser than TSMC's 7nm. (https://fuse.wikichip.org/news/2408/tsmc-7nm-hd-and-hp-cells-2nd-gen-7nm-and-the-snapdragon-855-dtco/) However, I think there were 2 different Intel 10nm nodes. The first was used on Cannonlake and...

semiwiki.com

3. Intel is behind. Based on point #2 above, they're probably behind by 2 generations now, and counting. But hey, they're partnering up with TSMC. You know that much. But that's the "catch up" play. You can tell then that Intel's CPU moving to 7nm or even 5nm may happen sooner than you or many are thinking. And they were stuck on 14nm up until the first half of this year so my counter question for you now is: given the performance of the Core i9 9980HK on 14nm, how much do you think it'll improve when Intel is on 7nm or 5nm (2 or 3 process nodes ahead) with that? Will the A14X be able to rival that?

 

[leman]

As textbook as it sounds, actually: I wrote a simple app in Xcode that spawns multiple threads... each trying to determine the prime factorization of a number by looking up a list of prime numbers and already existing factorizations. So each thread is reading a "page" in memory and swapping things in and out from storage as needed.

iOS doesn’t have a backing store. How exactly do you achieve swapping? Is there any chance you can share the code somehow?

I think you're probably thinking of in-kernel switcher mode of operation:

No, I am just saying that the LITTLE cores are much slower than the BIG ones. About 4 to 5 time slower. I think you are correct in saying that all cores are involved, but utilizing the LITTLE cores won’t bring much performance advantage. Even if you subtract 25% from the A12Z scores we are seeing, it doesn’t change the result much.

 

[bill-p]

iOS doesn’t have a backing store. How exactly do you achieve swapping? Is there any chance you can share the code somehow?

Here you go:

Detailed iOS file memory mapping (mmap) - Programmer Sought

Detailed iOS file memory mapping (mmap), Programmer Sought, the best programmer technical posts sharing site.

www.programmersought.com

I think you are reading the surface documentation that stops at "no backing store", and neglects to mention you can still map a file to memory.

No, I am just saying that the LITTLE cores are much slower than the BIG ones. About 4 to 5 time slower. I think you are correct in saying that all cores are involved, but utilizing the LITTLE cores won’t bring much performance advantage. Even if you subtract 25% from the A12Z scores we are seeing, it doesn’t change the result much.

Well, it's actually in line with DTK score scaling in Geekbench 5 at least.

Assuming Rosetta 2 imposes about 30% performance penalty and it's not making use of the LITTLE cores, going native would boost the score by 30% and then by 25%, that would result in:

2800 * 1.3 * 1.25 = 4550

And that's just about what the A12Z scores in native mode, either in DTK or in an iPad Pro.

Without the 25% performance boost, the A12Z would only score about 3640. So I think it's far less impressive without the LITTLE cores.

Then again, these are all just guesses. Apple is the only one who knows for sure. But I'd treat the A12Z like an 8-core processor either way.

 

[raknor]

You can guess that this blows out of memory very fast. In a sense, it's more a benchmark of storage access performance, plus maybe some branching, than of actual CPU performance.

Then again, the A12 is an SoC, so... I/O bandwidth is one thing that is embedded in there that should be tested anyways. Let's just say... I think the results are interesting, but the same task is of course a lot faster on the Core i9, even if I limited memory usage to 4GB.

You seem to be conflating CPU, system etc almost every single time. You're not just comparing the CPU performance in your test. You are comparing a 16" MBP to a iPad Pro. The 16" MacBook pro SSD is 2800 MB/s write and 2600 MB/s read. The iPad Pro is 360 MB/s write and 1200 MB/s read. One is designed to be a 90W system the other a 10 W system.

1. I don't disagree A12Z is a very efficient design from Apple. It's not the absolute best performance on the market, though.

No one is claiming that, since the chip was never designed for that purpose.. you are conflating a lot of different metrics.

Case in point, I'm sure you can agree that it's not fair to compare against an 8-core Core i9 with 64GB of RAM. But hey, that's what many Apple Silicon supporters are hoping Apple can achieve. Personally, I think that's just a bit far-fetched, but it may just be me.

That's one thing we can agree on .. it is just you. You are mixing up random metrics again. What does the 64GB RAM have to do with comparing the A12Z micro arch performance?

It is not fair to compare a 8-Core Core i9 to an A12Z because they were built for different market segments and products. One goes in a tablet.. the other goes in 15"+ laptops. Just like it is not fair to compare any of the low powered Intel chips that go into ultra portables to a 8-Core i9.

Would you be able to put a 8-Core i9 into a iPad Pro like chassis?

Apple Silicon fans are taking Apple's Microach performance and perf/per watt and extrapolating that to a Chip that will be designed to compete with the i9.

The question you are asking, is, why is this 45W-120W Intel chip faster than the 10W chip Apple chip. Even Intel's 10-25W SoCs that are built on their 10 nm process aren't in the same level as the 14nm i9. Does that mean Intel can't make good chips or does it mean that power consumption matters when making comparisons?

2. I'm saying Intel 10nm is not "similar" to TSMC 7nm. Performance and efficiency numbers of Intel 10nm are more akin to TSMC 10nm. So benchmarks will point to the process node numbers being "right".

They are similar in density. The fact that the performance and efficiency is not the same indicates Intel's woes with the process tech more than it does the inherent nature of it.

It's just Intel CEO and maybe wikichip saying Intel 10nm is actually equal to TSMC 7nm. Here, I think this discussion is closer to the truth (and the truth is basically closer to what I wrote above):

Is Intel 10nm really denser than TSMC 7nm?

In a recent article on Wikichip they claim that Intel's 10nm process is denser than TSMC's 7nm. (https://fuse.wikichip.org/news/2408/tsmc-7nm-hd-and-hp-cells-2nd-gen-7nm-and-the-snapdragon-855-dtco/) However, I think there were 2 different Intel 10nm nodes. The first was used on Cannonlake and...

semiwiki.com

I don't think you really understood that discussion thread .. nothing in that thread proves your point. It claims Intel 10 nm might have been denser than TSMC 7 but TSMC has moved on to 7+ and 6 nm already, so Intel has lost the density advantage. Again that is comparing the original 10 nm Intel process not the 10nm+ and 10nmSF that have come out subsequently.

3. Intel is behind. Based on point #2 above, they're probably behind by 2 generations now, and counting. But hey, they're partnering up with TSMC. You know that much. But that's the "catch up" play. You can tell then that Intel's CPU moving to 7nm or even 5nm may happen sooner than you or many are thinking.

Intel only announced some part of the GPU IP will be on TSMC 7nm nodes. So there is no "catch play" for Intel CPUs, at least not publicly announced.

And they were stuck on 14nm up until the first half of this year so my counter question for you now is:

No they aren't, they are on their 3rd Gen 10nm process for some of their volume SoCs.

given the performance of the Core i9 9980HK on 14nm, how much do you think it'll improve when Intel is on 7nm or 5nm (2 or 3 process nodes ahead) with that? Will the A14X be able to rival that?

Do you really think Apple is going to put the A14X into all their Mac products for the next 2-3 year while Intel catches up on process tech and won't be making higher performance parts?

 

[fokmik]

Will a first Gen ARM based MacBook Pro Surpasss and Outperform the previous Intel based Version ? I’m gonna say No.

It May have Longer battery life, it will have WiFi 6 (intentionally left out of current gen) It May be less expensively But will it be “Faster”

Heck they can go from LPDDR4 to LPDDR5 RAM today by switching to ARM.

i feel like the link below we are getting clowned by the Swicth ti ARM, which won’t have critical support for Windows Boot Camp installations for Gaming or support for an eGPU.

based on weeks of using the mac mini dev kit...i can say for sure...that the arm macbook pro (the smaller one, 13/14")
will surpass Intel one in every way...i mean every way
- battery life better
- less heat and less noise
- better cpu performance, better iGPU performance

Now, for the 16" Mbp i guess is no wonder Apple will release the arm based next year...because i dont know if apple can (or want) in 2020 to surpass the intel one on the dGPU power if apple chose to ditch amd for gpu and go for custom made gpu

So, dont worry, if apple release an lighter Macbook/Macbook Air and/or the smaller Macbook pro they all will surpass the current Intel ones 100%
Why? 2 reasons
1) mac mini runs A12Z and that even apps under rosetta2 runs better than any Macbook air or the base 13" macbook pro
2) Apple knows that everybody will compare the arm based with the current Intel one and from this you know that apple wants everybody to say ...better better better...apple own chips are better in every way ..this is how Apple thinks and does PR

 

[Manzanito]

For apple to pull this off the new macs have to be, at the very least, on par with intel counterparts in terms of performance, and also have additional features like substantially larger battery life. Otherwise the switch makes no sense, they are already leaving behind features like windows support.

Also, you can bet they are optimizing the hell out of the os and apple’s own apps.

 

[Maximara]

1. I don't disagree A12Z is a very efficient design from Apple. It's not the absolute best performance on the market, though. Case in point, I'm sure you can agree that it's not fair to compare against an 8-core Core i9 with 64GB of RAM. But hey, that's what many Apple Silicon supporters are hoping Apple can achieve. Personally, I think that's just a bit far-fetched, but it may just be me.

Actually it depends on the design of the computer that chip is going into. The main issue is Apple was designing its computers based on what Intel was promising and when Intel didn't produce the CPU promised when promised Apple had to effectively put slow and hotter running chips then they designed for.

Put that 8-core Core i9 with 64GB of RAM in a Mac Book Air design and see how well it does. Dimes will get you docents it will be heat throttling like nobody's business and sound like a jet airplane taking off if you try to push it to its limits.

That is another reason all the benchmark tests are effectively useless - the design of the computer the CPU is in is as important as the chip itself and odds are those test are not in computers of the same design.

 

[leman]

I think you are reading the surface documentation that stops at "no backing store", and neglects to mention you can still map a file to memory.

I am aware that you can use memory-mapped files on iOS. If I understand it correctly, you are using a large memory mapped file to simulate a situation where your task does not fit into the RAM. Given the fact that iPad SSD is optimized for reads and its write performance is around 1/3 of the SSD in the MBP, I am not at all surprised at the results.

Without the 25% performance boost, the A12Z would only score about 3640. So I think it's far less impressive without the LITTLE cores.

Still 50% faster than four Coffee Lake cores at 2.4 ghz. Anyway, the performance boost from the LITTLE cores seems to be comparable to the performance boost Intel gets out of hyper-threading, so you might as well treat an i9 as a 16 core CPU I personally think that the most interesting result is the single-core one, where you see around 100% improvement on both architectures running at the same clock.

 

[raknor]

As textbook as it sounds, actually: I wrote a simple app in Xcode that spawns multiple threads... each trying to determine the prime factorization of a number by looking up a list of prime numbers and already existing factorizations. So each thread is reading a "page" in memory and swapping things in and out from storage as needed.

You can guess that this blows out of memory very fast. In a sense, it's more a benchmark of storage access performance, plus maybe some branching, than of actual CPU performance.

Then again, the A12 is an SoC, so... I/O bandwidth is one thing that is embedded in there that should be tested anyways. Let's just say... I think the results are interesting, but the same task is of course a lot faster on the Core i9, even if I limited memory usage to 4GB.

Apart form the stuff I listed above there is another problem with this test. iOS ARM64 devices have a page size of 16KB. MacOS uses 4KB pages. Having multiple threads reading a "page" (most likely 4K), from memory automatically implies a 4X the amount of memory consumed and accessed on the iPad, that combined with the slower storage easily explains the differences you observed.

Like I said you can't make such comparisons and gain any knowledge of how a Apple Silicon Mac will perform. We should be able to do a like for like companion on MacOS running on ARM and x86 once the actual systems are released.

Without the 25% performance boost, the A12Z would only score about 3640. So I think it's far less impressive without the LITTLE cores.

Then again, these are all just guesses. Apple is the only one who knows for sure. But I'd treat the A12Z like an 8-core processor either way.

Like leman said above. If you are going to consider the heterogenous BIG.Little A12Z as a 8 core processor we should also consider the Core I9 a 16 core processor, the OS and Geekbench see 16 cpus they can run threads on with hyper threading. Hyperthreading also adds a similar boost like the little cores.

As I mentioned before.

A12z 4 core (Rosetta) Single:836, Multi:2884
Core i9 9980HK 4 core disable turbo and hyper threading Single:636, Multi:2365

A12z 4+4 core Native 2.5 Ghz Single: 1118, Multi: 4631
Core i9 8 core hyperthreading enabled, no turbo: 2.4 Ghz Single 614 Multi: 5086
Core i9 8 core Native (everything enabled): Single 1110 Multi: 6944

If the A12z had 8 symmetric cores the Core I-9 wouldn't stand a chance even with turbo and hyper threading enabled.

 

[slooksterPSV]

The problem is not that ARM is inferior. It is power efficient, and that's great for laptops. The problem is that Apple's motivation is to turn the mac to a big iphone, and milk customers from the 30% fee for all app store applications. The could not care less about scientists or compatibility, so they blocked Nvidia, CUDA, Bootcamp, 32 bit applications, etc. It's very sad, because Steve Jobs wanted the Mac to be the best tool for scientists in academia, and was very successful, but Tim Cook is moving further and further away from that paradigm, converting the mac into Fischer Price laptop with an Apple app store. It's a sad direction Apple is going, and the Apple logo is no longer shining on the mac.

I have a problem with this statement. Scientists and academic studies should be using newer software. I get it NASA still has CLI systems and use COBOL but we have so much more potential with 64-bit software. OpenCL is highly used in scientific environments because CUDA is nvdia locked down. If you want to run on various GPUs CUDA is not the way to go. Use opencl and upgrade your toolset. Hell even in education students need to use newer software because it’s supported, has optimizations or features older versions may not have, and can be a move forward for tackling new areas of study.

 

[jazz1]

I hope once the ARM Apple computers come out we see a bunch of reviews Intel CPU vs. ARM CPU. Price vs. performance would be interesting. Old Mac-Mini vs. new, Old iMac vs. new, Old laptop vs. new. I know we all have different priorities, but good reviewers should be able to help us make upgrade decisions.

 

[Yurk]

Maybe, you could have waited until you knew that your applications and hardware were supported by operating system upgrades?

I bought a working eGPU and a macbook in 2017, and Apple made them obsolete in 2018 (with Mojave and Catalina). 2018>2017. Do you understand how time works? Or are you trying to undermine anyone making Apple look bad with pointless arguments?

 

[bill-p]

I am aware that you can use memory-mapped files on iOS. If I understand it correctly, you are using a large memory mapped file to simulate a situation where your task does not fit into the RAM. Given the fact that iPad SSD is optimized for reads and its write performance is around 1/3 of the SSD in the MBP, I am not at all surprised at the results.

No, results are written into smaller chunks, each containing past results of past numbers up to a file size of about 1MByte. It's not like I need to read from 0 to 1,000,000,000 in every thread...

Still 50% faster than four Coffee Lake cores at 2.4 ghz. Anyway, the performance boost from the LITTLE cores seems to be comparable to the performance boost Intel gets out of hyper-threading, so you might as well treat an i9 as a 16 core CPU I personally think that the most interesting result is the single-core one, where you see around 100% improvement on both architectures running at the same clock.

But the i9 is not 16 cores. You can see from Geekbench that it's reporting 8.

Hyper-Threading giving a performance boost of 30% means Geekbench's execution code is causing Intel to skip instructions. That means Intel is wasting processing cycles running Geekbench. That sounds like Geekbench is biased against x86 to me. That makes it even more unreliable as a test, and it means the single-core score is meaningless as a comparison.

Apart form the stuff I listed above there is another problem with this test. iOS ARM64 devices have a page size of 16KB. MacOS uses 4KB pages. Having multiple threads reading a "page" (most likely 4K), from memory automatically implies a 4X the amount of memory consumed and accessed on the iPad, that combined with the slower storage easily explains the differences you observed.

Like I said you can't make such comparisons and gain any knowledge of how a Apple Silicon Mac will perform. We should be able to do a like for like companion on MacOS running on ARM and x86 once the actual systems are released.

Well, I don't think you quite... get it. Here's a wikipedia article on page size tradeoffs. Might be worth a read.

Some highlights:

A system with a smaller page size uses more pages, requiring a page table that occupies more space.

Larger page sizes mean that a TLB cache of the same size can keep track of larger amounts of memory, which avoids the costly TLB misses.

Larger page sizes lead to large amount of wasted memory, as more potentially unused portions of memory are loaded into main memory.

Because of this, large sequential transfers are more efficient than several smaller transfers.

So point #3 is the only drawback of larger page size. However, I know for sure my threads will read almost exactly the same sized chunks of data, and will free up that chunk when they're done, so there is no waste.

In fact, Apple's approach to iOS already makes it more efficient for the slower storage solution overall. Mac OS' approach wastes more of the storage speed offered by the Mac.

Like leman said above. If you are going to consider the heterogenous BIG.Little A12Z as a 8 core processor we should also consider the Core I9 a 16 core processor, the OS and Geekbench see 16 cpus they can run threads on with hyper threading. Hyperthreading also adds a similar boost like the little cores.

The boost from Hyper-Threading just highlights the inefficiency of Geekbench's multi-core test. It's not a highlight of Apple's approach.

I'm not "going to consider" anything. The A12Z is an 8-core CPU and it is a fact that all 8 cores can be active at the same time. They are physical cores. Hyper-Threading creates virtual cores, but the exact count of physical processor cores is still 8 no matter how you want to slice it.

 

[bousozoku]

I bought a working eGPU and a MacBook in 2017, and Apple made them obsolete in 2018 (with Mojave and Catalina). 2018>2017. Do you understand how time works? Or are you trying to undermine anyone making Apple look bad with pointless arguments?

I understand how time works quite well.

I went from Atari 800 to Atari 1040 ST to IBM L40SX to Apple machines with a 68040, PowerPC 604e, G3, dual G4, and finally an Intel i7, with a Sony MSX machine using a Z80 and a Kaypro 1. Yes, I understand, and I know to take time to evaluate each situation because when you're comfortable, someone may come to disturb you.