Apple Silicon A14X Bionic specs and results

[AppleFeller]

The leaked benchmark is not reliable. It's just a rumor number.

Geekbench is a joke measurement anyhow.

Why is it a joke? What is a good benchmark for you? Real world use? We will see that on Tuesday and it will excel all the same. The leaked numbers are entirely inline with A14 single core performance we already know is real, and multicore improvement is inline with historical jump apple makes from each generation of Ax series chips.

 

[Erehy Dobon]

Look man, the PC guys use something like 8-16 benchmarks for a real assessment (TomsHardware, AnandTech, etc.) and very few of those are synthetic benchmarks. The PC folks don't even use Geekbench and many of them are pretty explicit about an individual benchmark's shortcomings.

Real world use is ultimately the best.

When the Windows PC guys review graphics cards, they typically run 8-12 gaming benchmarks. Some are canned built-in benchmarks like Shadows of the Tomb Raider. Others are scripted play. A few sites will average the FPS from their tested games. There's also the quality-performance dichotomy. It's not just about FPS. If a GeForce RTX 2070 SUPER can feature hardware ray-tracing and a Radeon RX 5700XT cannot while both get similar FPS, there's qualitative difference.

There's also the notion of power consumption, cooling system noise, etc.

That's just a simple example of why simple benchmarks are dumb.

I can run the most advanced CPU benchmarks on my custom build PC but guess what? I rarely use my PC in that manner. The closest I ever get is Handbrake ripping.

By contrast, PC game play with AAA titles will tax the GPU but with different demands on the various subsystems of the graphics card. It's not just about core load. There are a ton of game metrics that aren't captured by a benchmark (load times, latency, gameplay quality, etc.).

Mac benchmarks are especially lame. I think a large part of this is that the people who write these benchmarks (mostly PC developers) don't understand some of the Mac advantages that are more difficult to quantify.

For sure, no one benchmark is comprehensive which is why the PC guys have 10-20 page reviews on a given CPU or GPU.

I'm not claiming that Windows PC are inherently superior but in terms of the benchmarking and review process, there are simply more eyes on the Windows PC hardware process than the Mac hardware process. This leads to some very critical analysis. If the reviewer slips up, he/she can count of an army of review readers to unvalidate his/her analysis within hours.

In the Mac world, we see regular published writers spout inaccuracies that never get corrected.

 

[leman]

Why would they add low power cores to the 16” chip? As I understand it they are intended for background processing of light tasks, and there’s only so much of that work to go around on any running system.

Low power cores are a fundamental feature of Apple SoCs and they are tremendously useful. There is a lot of background activity going on most of the time - backups, email fetches, indexing, analysis etc. - just look at your activity monitor. Efficiency cores will provide a tremendous boost for battery life abs responsiveness of the Macs.

At the same time, I don’t expect Apple to ever ship more than 4 efficiency cores, that seems to be a sweet spot for dealing with non-real-time tasks on any machine.

 

[leman]

Why is it a joke? What is a good benchmark for you? Real world use? We will see that on Tuesday and it will excel all the same. The leaked numbers are entirely inline with A14 single core performance we already know is real, and multicore improvement is inline with historical jump apple makes from each generation of Ax series chips.

Let’s not make it too complicated. The point is - there is no geekbench entry and no screenshot. These are just some numbers someone put in their website and someone else copied. There is no proof these numbers were ever real. They are in realistic ballpark sure - I was posting similar numbers half a year ago when extrapolating A14 performance- but we should not discuss them as if they were a real deal.

 

[aednichols]

Low power cores are a fundamental feature of Apple SoCs and they are tremendously useful.

Acknowledged.

I should have written more precisely, "Why would they add [4] low power cores to the 16” chip [on top of the base 4]?" which was the claim in the post I replied to.

 

[leman]

Acknowledged.

I should have written more precisely, "Why would they add [4] low power cores to the 16” chip [on top of the base 4]?" which was the claim in the post I replied to.

Ah, yes, here I agree with you. 8 efficiency cores in a 16” don’t make any sense to me. My guess is 12 performance + 4 efficiency cores - this will outperform any desktop Intel or AMD CPU.

 

[anthony13]

Wow... I really wanted to get in on this conversation but ya‘ll are preaching above my paygrade. Hope my brand new iMac from august isn’t out of date too soon (my software wont be compatible for years).

 

[aednichols]

Wow... I really wanted to get in on this conversation but ya‘ll are preaching above my paygrade. Hope my brand new iMac from august isn’t out of date too soon (my software wont be compatible for years).

You'll be fine, the config in your signature is a monster.

 

[machinesworking]

Ah, yes, here I agree with you. 8 efficiency cores in a 16” don’t make any sense to me. My guess is 12 performance + 4 efficiency cores - this will outperform any desktop Intel or AMD CPU.

Man, I really hope you're right, and the 16" comes out blazing with 12 performance + 4 efficiency cores. I lucked out for years with a sweat spot 2012 2.7ghz Retina MBP that kept up with most of the laptops Apple put out until about two years ago, but in general it's beyond time to upgrade. I'm just hoping the rumor is right and Apple announces and puts out the 15-16" model this year.

 

[leman]

Man, I really hope you're right, and the 16" comes out blazing with 12 performance + 4 efficiency cores. I lucked out for years with a sweat spot 2012 2.7ghz Retina MBP that kept up with most of the laptops Apple put out until about two years ago, but in general it's beyond time to upgrade. I'm just hoping the rumor is right and Apple announces and puts out the 15-16" model this year.

I won’t expect one before summer 2021, but anything can happen.

 

[aednichols]

I won’t expect one before summer 2021, but anything can happen.

I become eligible for a new 15/16" work laptop in April 2021. Can easily wait a few more months to summer.

My current 2016 15" is very hot & loud, has the butterfly keyboard, and 16 GB is not enough for the extreme multi-tasking I do.

 

[iPadified]

Look man, the PC guys use something like 8-16 benchmarks for a real assessment (TomsHardware, AnandTech, etc.) and very few of those are synthetic benchmarks. The PC folks don't even use Geekbench and many of them are pretty explicit about an individual benchmark's shortcomings.

Real world use is ultimately the best.

When the Windows PC guys review graphics cards, they typically run 8-12 gaming benchmarks. Some are canned built-in benchmarks like Shadows of the Tomb Raider. Others are scripted play. A few sites will average the FPS from their tested games. There's also the quality-performance dichotomy. It's not just about FPS. If a GeForce RTX 2070 SUPER can feature hardware ray-tracing and a Radeon RX 5700XT cannot while both get similar FPS, there's qualitative difference.

There's also the notion of power consumption, cooling system noise, etc.

That's just a simple example of why simple benchmarks are dumb.

I can run the most advanced CPU benchmarks on my custom build PC but guess what? I rarely use my PC in that manner. The closest I ever get is Handbrake ripping.

By contrast, PC game play with AAA titles will tax the GPU but with different demands on the various subsystems of the graphics card. It's not just about core load. There are a ton of game metrics that aren't captured by a benchmark (load times, latency, gameplay quality, etc.).

Mac benchmarks are especially lame. I think a large part of this is that the people who write these benchmarks (mostly PC developers) don't understand some of the Mac advantages that are more difficult to quantify.

For sure, no one benchmark is comprehensive which is why the PC guys have 10-20 page reviews on a given CPU or GPU.

I'm not claiming that Windows PC are inherently superior but in terms of the benchmarking and review process, there are simply more eyes on the Windows PC hardware process than the Mac hardware process. This leads to some very critical analysis. If the reviewer slips up, he/she can count of an army of review readers to unvalidate his/her analysis within hours.

In the Mac world, we see regular published writers spout inaccuracies that never get corrected.

If you compile all these Anandtech benchmarks into one number, it may not be far away from the Geekbench ranking. I also read the deep analysis but seldom gets much wiser compared to a Geekbench score. The problem with specific benchmark is the if you use an app not covered by the benchmark, you have you no clue how the machine will perform anyway.

Furthermore, the variance between specific benchmarks is large between platforms (Intel, AMD, Apple Silicon, other ARM implementations). Some platform is strong on some tests and other platforms on other tests. It nearly is meaningless to compare 20-30 pages of tests unless you buy a computer for one specific application but few of us do.

For a multipurpose machine, Geekbench is a fine approximation. For specialised workloads like handbrake, rendering or one particular game, specific tests are of course better.

 

[pshufd]

I'd like having a lot of efficiency cores in a MacBook Pro 16 as it would mean better overall battery life. The cost, of course, is more die space and expense so there's a tradeoff for Apple. I am one of those few people that would pay more for a system that runs cooler and more efficiently.

 

[leman]

I'd like having a lot of efficiency cores in a MacBook Pro 16 as it would mean better overall battery life. The cost, of course, is more die space and expense so there's a tradeoff for Apple. I am one of those few people that would pay more for a system that runs cooler and more efficiently.

It’s a balance between performance and battery life... overall, I’d say that four efficiency cores are a sweet spot. Their main purpose is to handle the low-priority background tasks. If you are running some demanding sustained multi-core work, you are probably better off with high-performance cores anyway.

 

[leman]

If you compile all these Anandtech benchmarks into one number, it may not be far away from the Geekbench ranking. I also read the deep analysis but seldom gets much wiser compared to a Geekbench score. The problem with specific benchmark is the if you use an app not covered by the benchmark, you have you no clue how the machine will perform anyway.

Furthermore, the variance between specific benchmarks is large between platforms (Intel, AMD, Apple Silicon, other ARM implementations). Some platform is strong on some tests and other platforms on other tests. It nearly is meaningless to compare 20-30 pages of tests unless you buy a computer for one specific application but few of us do.

For a multipurpose machine, Geekbench is a fine approximation. For specialised workloads like handbrake, rendering or one particular game, specific tests are of course better.

I used to be highly critical of Geekbench (still am to certain degree), but GB5 does seem to correlate well with other benchmarks (https://nuviainc.com/blog/performance-delivered-a-new-way-part-2geekbench-versus-spec)

One thing to keep in mind is that Geekbench is a bursty workload, so it will perform better (especially in single core) on CPUs with more agile turbo. This is one reason why Ice Lake performs so much better than Skylake despite having similar max. IPC.

 

[pshufd]

It’s a balance between performance and battery life... overall, I’d say that four efficiency cores are a sweet spot. Their main purpose is to handle the low-priority background tasks. If you are running some demanding sustained multi-core work, you are probably better off with high-performance cores anyway.

Power consumption goes up with frequency so having a lot of low-power cores would do more work with the same amount of power. The tradeoff is die-space. My daily workload runs about 15-20 percent of my CPU (10700) and I want it that way to keep the system cool. It has a dual-fan CPU cooler and three case fans (can put in up to eight).

I'd love to try a 16 core CPU running at very low frequencies.

 

[dmccloud]

I used to be highly critical of Geekbench (still am to certain degree), but GB5 does seem to correlate well with other benchmarks (https://nuviainc.com/blog/performance-delivered-a-new-way-part-2geekbench-versus-spec)

One thing to keep in mind is that Geekbench is a bursty workload, so it will perform better (especially in single core) on CPUs with more agile turbo. This is one reason why Ice Lake performs so much better than Skylake despite having similar max. IPC.

Sometimes I openly wonder if Intel prioritizes single-core performance just so it can brag about how their new ultra-mobile notebook processor is the fastest around (completely ignoring the fact that once they switch to multi-core the performance drops off noticeably). To me, Geekbench is a good way to measure performance relative to other systems on the market, but has little bearing on my personal workload, especially since I use my MBA in a completely different manner than my MSI gaming laptop (which is my heavy lifter).

 

[pshufd]

Sometimes I openly wonder if Intel prioritizes single-core performance just so it can brag about how their new ultra-mobile notebook processor is the fastest around (completely ignoring the fact that once they switch to multi-core the performance drops off noticeably). To me, Geekbench is a good way to measure performance relative to other systems on the market, but has little bearing on my personal workload, especially since I use my MBA in a completely different manner than my MSI gaming laptop (which is my heavy lifter).

I think that they prioritize it because it's all that they have. Well, had now with Zen 3.

I find that Geekbench 5 correlates well with relative performance of my workload which is highly threaded. I find it quite nice to have a simple benchmark that allows me to look at CPUs and systems for my use.

One thing that I'd love to add to the benchmark is a measure of efficiency for the workload. I imagine that would be pretty hard to do.

 

[Woochoo]

Neither AMD or Intel can compete with Apple Silicon because Apple "doesn't care" about CPU margins or cost per unit (as long as it's cheaper for them than buying from Intel) while Intel/AMD do: Apple's benefits doesn't come with CPU sales but with hardware/whole product with high margins. That means Apple can make bigger dies and won't care that much: to put an example, the A13 was 98mm2 while the 3600X is 74mm2. That's 25% bigger die size, thus 25% less number of CPUs per waffer (so less benefits in AMD/Intel case).

The other part of the equation is ASICs: x86 barely has any dedicated hardware aside video encoder/decoders and AVX, while ARM SoCs usually have specially Apple Silicon. That Machine Learning accelerators, NPU for inferencing, VPU and IPU for video and image processing, etc. really can make things faster and for less watts. That's why an iPad Pro with the A12X was rendering a video 25% faster than the 2018 15" MBP with a higher scoring multithread Intel than the A12X.

AMD/Intel doesn't have to compete with Apple Silicon as Apple doesn't sell any SoC, but they may have to compete soon with other ARM manufacturers like Qualcomm or Samsung if this snowballs.

 

[thingstoponder]

Intel and AMD have no response to this approach. I have a custom built desktop PC and I'm running a 65W TDP CPU. It performs wonderfully when I max everything out. It's a big power hog when it's mostly idle

Intel has a response to big.little. Their next next architecture is called Alder Lake and will be a 8+8 configuration on the desktop, with laptop chips too.

 

[thingstoponder]

The leaked benchmark is not reliable. It's just a rumor number.

Geekbench is a joke measurement anyhow.

No, geekbench isn’t a joke. It doesn’t matter how many times people on here say this, it doesn’t make it true.

The leaks however are probably nonsense. I don’t believe the CPU Monkey supposed leak of the a14x.

 

[thingstoponder]

You guys are hilarious, the A14X results that have leaked are the next iPad Pro and entry level Macbook/Air SOC with the 4 LP and 4HP core config. The 13" MacBook Pro will employ a variant with the 12 core configuration. The 16" we have no leak to go off yet but if I was to guess it will likely be a 16 core configuration with 8 LP and 8 HP cores. To clarify though that "12 core" SOC will appear in the 13" MacBook Pro.

You don’t know that. You’re just guessing like anyone else. The Bloomberg report says the first batch of 3 chips will not go above 12 cores.

Now they could all be 12 core with different GPUs, or two of them could be a one of them a 10 core. But there is no indication of a 16 core.

 

[thingstoponder]

I used to be highly critical of Geekbench (still am to certain degree), but GB5 does seem to correlate well with other benchmarks (https://nuviainc.com/blog/performance-delivered-a-new-way-part-2geekbench-versus-spec)

One thing to keep in mind is that Geekbench is a bursty workload, so it will perform better (especially in single core) on CPUs with more agile turbo. This is one reason why Ice Lake performs so much better than Skylake despite having similar max. IPC.

The main difference is Tiger Lake has AVX512 so it smokes the Skylake ones in tests that use that. If you look at the score breakdowns it’s less impressive.

 

[thingstoponder]

Neither AMD or Intel can compete with Apple Silicon because Apple "doesn't care" about CPU margins or cost per unit (as long as it's cheaper for them than buying from Intel) while Intel/AMD do: Apple's benefits doesn't come with CPU sales but with hardware/whole product with high margins. That means Apple can make bigger dies and won't care that much: to put an example, the A13 was 98mm2 while the 3600X is 74mm2. That's 25% bigger die size, thus 25% less number of CPUs per waffer (so less benefits in AMD/Intel case).

The other part of the equation is ASICs: x86 barely has any dedicated hardware aside video encoder/decoders and AVX, while ARM SoCs usually have specially Apple Silicon. That Machine Learning accelerators, NPU for inferencing, VPU and IPU for video and image processing, etc. really can make things faster and for less watts. That's why an iPad Pro with the A12X was rendering a video 25% faster than the 2018 15" MBP with a higher scoring multithread Intel than the A12X.

AMD/Intel doesn't have to compete with Apple Silicon as Apple doesn't sell any SoC, but they may have to compete soon with other ARM manufacturers like Qualcomm or Samsung if this snowballs.

You’re forgetting the IO die for the 3600x. The 74mm number is just the CPU core die. The IO die is 125mm but is on 12nm so it’s cheaper to make.

if you want a better illustration of Apple having bigger chips just look to the phones. Apples are always bigger when you consider the modem is discrete on iPhones and usually integrated in snapdragons. Your main point was correct.

 

[iPadified]

I used to be highly critical of Geekbench (still am to certain degree), but GB5 does seem to correlate well with other benchmarks (https://nuviainc.com/blog/performance-delivered-a-new-way-part-2geekbench-versus-spec)

One thing to keep in mind is that Geekbench is a bursty workload, so it will perform better (especially in single core) on CPUs with more agile turbo. This is one reason why Ice Lake performs so much better than Skylake despite having similar max. IPC.

You have a point with Geekbench short workloads. That is not good but could easily be fixed if Geekbench wanted to but I doubt the average user/journalist want to wait up to 30 minutes. Why not also report temperature and fans speed at the same time to have complete package? Why not also hijack the microphone and measure the noise levels.

At any rate, tonight is probably the most interesting event for years that Apple will have. WWDC was a teaser but tonight they will need to prove themselves. Therefore, I am surprised the iMac 24 is not in the cards. However, a 16 inch MBP, even a teaser, showing true desktop performance is a suitable substitute. The 13 inches will show excellent performance per watt but only in the low performance end which tells nothing about the potential for the iMac and MP. Info about the latter is much needed.