The first Apple A10x Geekbench4 score?

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Herr Kutz

Platinum Member
Jun 14, 2009
2,545
242
106
Apple has 7% of worldwide market share for PC shipments. They're roughly the size of Acer or Asus when it comes to computer shipments.

So, as I said previously, a small percentage. That combined with desktop/laptop processors not encompassing the entirety of Intel's products hardly equates Intel to being royally screwed if MACs started using ARM processors.
 

scannall

Golden Member
Jan 1, 2012
1,944
1,638
136
So, as I said previously, a small percentage. That combined with desktop/laptop processors not encompassing the entirety of Intel's products hardly equates Intel to being royally screwed if MACs started using ARM processors.
Would Intel go down the drain if Apple went their own way? Of course not. It would sting on the bottom line though, as Apple buys all high end SKU's. No Celerons or Pentiums on their purchase orders.
 

Herr Kutz

Platinum Member
Jun 14, 2009
2,545
242
106
Would Intel go down the drain if Apple went their own way? Of course not. It would sting on the bottom line though, as Apple buys all high end SKU's. No Celerons or Pentiums on their purchase orders.

It looks like we're all in agreement that "royally screwed" is a vast embellishment.
 

Drazick

Member
May 27, 2009
53
70
91
I still wonder if Apple will try to really get to Intel levels. It's unlikely their micro-architecture can reach as high frequencies as Intel, so they'd have to sacrifice IPC. But nonetheless their current chips are incredibly efficient.

Why are you so sure about this assumption?
Let me try challenge it.

Apple's CPU's are around 2 - 3 GHz which are around the numbers of Intel 2xxx generation.
Transistors wise, they have similar number of transistors (Remember to remove the GPU from both chips when doing the calculation) and since Intel's CPU has larger TDP Budget it means Apple's better manufacturing is meaningless.

Yet Apple have better IPC.
Intel, from that point, increased their speed by 30-40% and improved its IPC.

Namely, there is no physical law which will enforce Apple to reduce its IPC in order to increase speed.

Apple curve is better than Intel's curve on the most important measure, given a budget of Transistors, which one extracts better performance?

I would gamble that in the years to come Apple will both increase its speed and its IPC.
That's what history tells us, Intel did it, no reason Apple won't...
 

Valantar

Golden Member
Aug 26, 2014
1,792
508
136
From reading around a bit, it seems like the platform neutrality of GeekBench is quite controversial. For example, it counts SHA2 encryption as a CPU integer workload, despite it being handled by fixed-function hardware in most CPUs today (including Apple's). That seems like a rather silly oversight.
 

Andrei.

Senior member
Jan 26, 2015
316
386
136
From reading around a bit, it seems like the platform neutrality of GeekBench is quite controversial. For example, it counts SHA2 encryption as a CPU integer workload, despite it being handled by fixed-function hardware in most CPUs today (including Apple's). That seems like a rather silly oversight.
It would surprise you just how little these hardware accelerated blocks / instructions are actually used by software. You can make a claim against representative workloads, but you can't say it favour one platform or the other.

You can also make a point that it makes zero sense for any mobile device to do the JPEG subtest because every SoC out there has a dedicated JPEG fixed function de/encoder.
 
  • Like
Reactions: Arachnotronic

Carfax83

Diamond Member
Nov 1, 2010
6,841
1,536
136
Why are you so sure about this assumption?
Let me try challenge it.

Apple's CPU's are around 2 - 3 GHz which are around the numbers of Intel 2xxx generation.
Transistors wise, they have similar number of transistors (Remember to remove the GPU from both chips when doing the calculation) and since Intel's CPU has larger TDP Budget it means Apple's better manufacturing is meaningless.

Yet Apple have better IPC.
Intel, from that point, increased their speed by 30-40% and improved its IPC.

Namely, there is no physical law which will enforce Apple to reduce its IPC in order to increase speed.

Apple curve is better than Intel's curve on the most important measure, given a budget of Transistors, which one extracts better performance?

I would gamble that in the years to come Apple will both increase its speed and its IPC.
That's what history tells us, Intel did it, no reason Apple won't...

The comparison link that asendra posted on the previous page shows that Intel's Kaby Lake Y has better single core performance, even at a lower clock speed, which implies better IPC.
 

asendra

Member
Nov 4, 2012
156
12
81
The comparison link that asendra posted on the previous page shows that Intel's Kaby Lake Y has better single core performance, even at a lower clock speed, which implies better IPC.
Not really, even though Geekbench shows 1.6Ghz for that Kaby Lake, the real speed of it is 3.2Ghz with boost. And for what I have seen with Skylake Y CPUs, they hold the maximum boost clocks for the entire Geekbench test.
Geekbench is notoriously bad at detecting real clocks, specially with todays highly adaptive CPUs
 

StinkyPinky

Diamond Member
Jul 6, 2002
6,761
777
126
Wouldn't surprise me to see Apple move away from Intel for their laptops/desktops at some point in the future. I think their long term plan is to merge IOS and MacOS.

Interesting to note they are introducing a file manager in IOS 11....
 
  • Like
Reactions: dark zero

Drazick

Member
May 27, 2009
53
70
91
The comparison link that asendra posted on the previous page shows that Intel's Kaby Lake Y has better single core performance, even at a lower clock speed, which implies better IPC.
  1. For single thread the Intel CPU uses Turbo and the effective speed is much higher.
  2. My point was different. I said that Intel improved both Speed and IPC in the last ~6 years.
    Maybe in small figures for the IPC but the principle holds, there is no law which makes you reduce the IPC in order to increase speed of a given architecture.
I think Apple shows impressive results.
But one should note that in the past few years Intel put most of its efforts in the Vector Units.
I believe that for these kind of tests Intel still holds the crown.
 
Last edited:

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
There is definitely some significantly unbalanced compute surface in GB4, as the internal benchmarks for Cortex A75 show only 20-25% integer gains and 35% FP gains, but show a disproportionate 48% gain in GB4 - a definitive sign that it can be gamed, even though the A55 benchmarks show a more balanced gain in general compute vs GB4.

In theory this should put an A75 centered Kirin chip over 3000 in GB4, taking into account process gains since Kirin 960 to give 100-200 Mhz extra clock, plus possibly higher efficiency of LITTLE A55 giving slightly more headroom for big A75.
 

Nothingness

Platinum Member
Jul 3, 2013
2,371
713
136
Namely, there is no physical law which will enforce Apple to reduce its IPC in order to increase speed.
Increasing the IPC obviously means increasing the amount of work you have to do in a cycle, so some laws of physics will come back to haunt you: propagation delays due to increased size, increased logic depth, thermal issues and so on :)

  1. For single thread the Intel CPU uses Turbo and the effective speed is much higher.
  2. My point was different. I said that Intel improved both Speed and IPC in the last ~6 years.
    Maybe in small figures for the IPC but the principle holds, there is no law which makes you reduce the IPC in order to increase speed of a given architecture.
I think Apple shows impressive results.
But one should note that in the past few years Intel put most of its efforts in the Vector Units.
I believe that for those it still holds the crown.
Intel gained a lot by means of process, something that has slowed down a lot recently as 14nm has shown.

I'm really impressed by what Apple has achieved, but Intel also has many great engineers and micro-architects so there's no reason to think one can go way beyond Intel in terms of performance.

Also back to my original claim that Apple might have issues reaching as high frequencies as Intel without reducing IPC, I have worked in enough (successful) CPU design teams to know how hard it's to balance IPC and frequency. There's no free lunch :D
 
  • Like
Reactions: Arachnotronic

Lodix

Senior member
Jun 24, 2016
340
116
116
The funny thing is that this SOC is built at TSMC 16nmFF+ and the next Apple A11 coming inside the IPhone this year being manufactured at 10nm could match or exceed it in CPU/GPU perfomance in a smaller TDP.
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
http://browser.primatelabs.com/v4/cpu/3036382

We should see more soon.

If the frequency of 2.36 GHz is correct then the single thread
score of 3832 would significantly extend the already monumental
lead in IPC of the Apple designed cores/SOCs (edit) with much
the increase coming from the higher memory bandwidth though.
(See the link in the next post)
For reference, my 3-core A8X at 1.5 GHz gets 1823 / 4331.
 

oak8292

Member
Sep 14, 2016
81
67
91
Apple does have an engineering advantage in designing for a single OS for a single power envelope with little to no legacy, e.g. in the next iteration, A11 may quit supporting ARM 32 bit instructions. iOS 11 this fall is dropping support for 32 bit applications. The A11 won't be marketed to any 3rd party vendors who might need 32 bit support. Wouldn't this save area in silicon?

The rumors have been that the A10X is built on the TSMC 10nm process. Is there any evidence that this was changed to 16nm? Given the fairly large die size for the A10 and the increase of a third core it seems highly likely that this has been built as rumored on the '10nm' technology.
 

dark zero

Platinum Member
Jun 2, 2015
2,655
138
106
It looks like we're all in agreement that "royally screwed" is a vast embellishment.
Ok, I admit my defeat, not royally screwed, but still it would be along the Microsoft approach to Qualcomm a bigger problem for them.

And not to say that nVIDIA would see this as a chance to enter finally as a ARM PC contestant. Their Denver cores are not bad after all. And their GPU is already better than Adreno.
 

asendra

Member
Nov 4, 2012
156
12
81
Reviews dropping for the new iPad Pros. Lets see if we get a deep dive on the A10X, with SPECInt 2006 scores against intel Y CPUs. They did it for the A9X on the first iPad Pro...
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
Reviews dropping for the new iPad Pros. Lets see if we get a deep dive on the A10X, with SPECInt 2006 scores against intel Y CPUs. They did it for the A9X on the first iPad Pro...
Yeah I see there a lot of reviews but so many are just subjective, with very little objective testing.

So far, reviews like this will have to suffice.

https://www.laptopmag.com/reviews/tablets/apple-ipad-pro-10-5

There are a few tests in there but not much for CPU alone. It has Geekbench 4 at 9233 though, confirming the initial leak that it is very fast. For comparison, the 2017 Kaby Lake m3 I just ordered gets about 7000. However, the conference I'm at right now has reaffirmed I cannot function with just an iPad. I only took an iPad Air 2 this time with an external keyboard and regret not taking my laptop even though I'm not presenting or attending all that much of it.
 

TheGiant

Senior member
Jun 12, 2017
748
353
106
Are there any real world business/home usage scenarios that are CPU limited in the reviews around?
I would be very interested how the IPad Pro really works with excel, more apps open at once....

There is tons of stuff and "reviews" around the web saying the thing is fast, but the workload remains hidden..
 

Thala

Golden Member
Nov 12, 2014
1,355
653
136
Some have mentioned Keller (sorry for citing only him again, hundreds of people who have designed processors over all the years) as some figure who magically enhanced Apple's chips to this level, but it seems weird when another core developed by him too isn't up to par.

At one point Keller was designing both Ryzen and K12 in parallel. He did comment that K12 would be more efficient (e.g. faster using the same area) than Ryzen due to the advantages of ARM architecture.
So K12 would have been faster than Ryzen.
In summary it was the x86 architecture putting limitations on what Keller could achieve with Ryzen.
 

Thala

Golden Member
Nov 12, 2014
1,355
653
136
Indeed.

I still wonder if Apple will try to really get to Intel levels. It's unlikely their micro-architecture can reach as high frequencies as Intel, so they'd have to sacrifice IPC. But nonetheless their current chips are incredibly efficient.

Apple is using a low power process (high VT, low Vcc) and reaching 2.3GHz. There is no evidence that if you are using a proper high performance process and upping Vcc that 5GHz is out of reach with the current design. In fact doubling frequencies with the same design is not uncommon when switching processes in conjunction with upping Vcc to overdrive voltage levels.
 
  • Like
Reactions: Drazick and Lodix

Eug

Lifer
Mar 11, 2000
23,583
996
126
Are there any real world business/home usage scenarios that are CPU limited in the reviews around?
I would be very interested how the IPad Pro really works with excel, more apps open at once....

There is tons of stuff and "reviews" around the web saying the thing is fast, but the workload remains hidden..
Problem with Excel is not the CPU. Problem with Excel, or Numbers for that matter, is that iOS' UI is just not built for this type of functionality. The touch interface without trackpad or mouse support is annoying and awkward to use, which is one big reason I just bought a MacBook with (much) lower GeekBench scores than the iPad Pro 10.5".

I use my iPad Air 2 for media and internet consumption, and email etc. But it sucks for data entry in Excel, even with my external full size iPad keyboard.
 

TheGiant

Senior member
Jun 12, 2017
748
353
106
Problem with Excel is not the CPU. Problem with Excel, or Numbers for that matter, is that iOS' UI is just not built for this type of functionality. The touch interface without trackpad or mouse support is annoying and awkward to use, which is one big reason I just bought a MacBook with (much) lower GeekBench scores than the iPad Pro 10.5".

I use my iPad Air 2 for media and internet consumption, and email etc. But it sucks for data entry in Excel, even with my external full size iPad keyboard.

I have only ipad 4 (the first with retina) with a6x, keyboard and pen (which is of course inferior to the current line) and I have the same experience as you.
But let's put aside the user experience for classic "desktop/laptop" apps. Just performance. The whole web is filled with subjective info.
I know that it is fast for browsing web, playing youtube and other movies and other basic stuff like responding to emails. But where can I use and see the performance of the CPU.
Does anyone have the real experience ?
I wanted to compare it to the geekbench 4 score. My Surface 4 pro i5 6300U scored a while ago 3623 single and 7048 multicore result (while running antivirus and all the bloatware behind). Which is not so much slower single core but seriously slower multi core. And I consider the SP4PRO performance acceptable/nice.
 

asendra

Member
Nov 4, 2012
156
12
81
A lot of webs have done benchmarks, but all of those are more or less platform specific so you can give them the validity you want.

the only thing we can hope is that the fine guys at Anadtech will do an A10X deep dive the same way they did with A9X (but failed to do with the A10!!). With SpecInt06 scores compared to skyline/kabylake Y cpus etc
 

oak8292

Member
Sep 14, 2016
81
67
91
OK We finally have confirmation from Chipworks that the A10X was built on the TSMC 10nm process. The die size for the processor is 96.4 mm2.

http://techinsights.com/about-techinsights/overview/blog/10nm-rollout-marching-right-along/

"This is an impressive full node scale, when accounting for the extra CPU cores built into the A10X and extra IP blocks of the A10 vs. A9 family. We estimate a 45% die level scale (0.55x the area of running on the previous technology), based on our detailed floorplan analyses of the Apple A-series."

It would be nice if Anandtech could complete a timely analysis on the functionality of both the A10 and A10X sometime in the not too distant future. However, it seems the site is pretty resource constrained trying to keep up with everything AMD, Intel and others are doing.