• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

Info TOP 20 of the World's Most Powerful CPU Cores - IPC/PPC comparison

Page 12 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Carfax83

Diamond Member
Nov 1, 2010
5,926
702
126
I think the major part of why so many people are bothered by how Richie Rich frames the entire IPC argument, is how it often ignores other aspects of CPU performance which are equally as important. I don't doubt for one second that Intel and AMD could have made a lower clocked, very wide CPU with high IPC if they wanted to. But clockspeed is also important to overall performance, especially in heavier workloads; and so is SIMD, SMT, caches etcetera.

You see the same thing in GPUs. For many years GPUs were stuck in sub ghz clock speeds, and rose slowly over time. Now they're hitting close to, and over 2ghz. And this is for 3D rendering which is often said to be "embarrassingly parallel."

What matters more than anything is the overall performance of a CPU. I guess we'll see which strategy wins out in the next few years.
 

Doug S

Senior member
Feb 8, 2020
356
510
96
What matters more than anything is the overall performance of a CPU. I guess we'll see which strategy wins out in the next few years.
Well the numbers for the A14 show it beating the best of both Intel & AMD in single core GB5. I know, I know, people will say "but GB5 is (somehow) biased for ARM and against x86". If they can put that A14 in an iPad (and presumably the iPhone 12) then they'll have something at least as fast in the Macs coming out in a few months. They'll be able to run a much wider variety of benchmarks, I think that will make things uncomfortable for the x86 fanboys.

But I agree 100% that IPC is pretty irrelevant. Apple can't juice A14 to 5 GHz anymore than Intel can double their IPC by targeting 3 GHz, so the only realistic comparison is head to head in raw performance and ranking by IPC is stupid.
 

Thunder 57

Golden Member
Aug 19, 2007
1,546
1,452
136
Well the numbers for the A14 show it beating the best of both Intel & AMD in single core GB5. I know, I know, people will say "but GB5 is (somehow) biased for ARM and against x86". If they can put that A14 in an iPad (and presumably the iPhone 12) then they'll have something at least as fast in the Macs coming out in a few months. They'll be able to run a much wider variety of benchmarks, I think that will make things uncomfortable for the x86 fanboys.

But I agree 100% that IPC is pretty irrelevant. Apple can't juice A14 to 5 GHz anymore than Intel can double their IPC by targeting 3 GHz, so the only realistic comparison is head to head in raw performance and ranking by IPC is stupid.
I don't think anybody is a "fanboy" of anything other than performance and price. Well, unfortunately that isn't true. Some people really love AMD or Intel or ARM or x86. I think most normal people though don't think that way.
 

DrMrLordX

Lifer
Apr 27, 2000
16,494
5,474
136
Why is this bogus? Looks pretty accurate to me.
It relies solely on GB5 (though he was using SPEC earlier, which is arguably worse). It doesn't take into account anything mentioned by @Carfax83 .

@Doug S

If Apple can challenge something like a 9900k with a low-power laptop version of A14 in absolute performance then kudos to them, and I await the benchmarks proving it. I can only lament that Apple's CPU designs are trapped within their software ecosystem. MacOS isn't quite as bad, but still. Bleh. At least we can run some proper benchmarks like Blender on A14.
 

Thala

Golden Member
Nov 12, 2014
1,096
430
136
It relies solely on GB5 (though he was using SPEC earlier, which is arguably worse). It doesn't take into account anything mentioned by @Carfax83 .
So what? It is an perf/clock table using a benchmark, which has no architecture specific optimizations.

Sure, your could create tables with other metrics or other benchmarks - but this table is not. It just shows how much instruction level parallelism a certain architecture can extract from the code per cycle using architecture neutral source code and with Geekbench 5 is even using the same compiler (CLANG).
This does not make the table bogus - as i said, the information collected looks pretty much accurate.
 
  • Like
Reactions: Lodix

Hitman928

Platinum Member
Apr 15, 2012
2,915
2,454
136
So what? It is an perf/clock table using a benchmark, which has no architecture specific optimizations.

Sure, your could create tables with other metrics or other benchmarks - but this table is not. It just shows how much instruction level parallelism a certain architecture can extract from the code per cycle using architecture neutral source code and with Geekbench 5 is even using the same compiler (CLANG).
This does not make the table bogus - as i said, the information collected looks pretty much accurate.
If you actually read back through the thread, there are many valid criticisms of his table that he completely ignores. Not the least of which he chose his AMD and Intel examples carefully and refuses to update his table with any GB5 results for Intel and AMD that show significantly higher points per clock.
 

Richie Rich

Senior member
Jul 28, 2019
470
227
76
I don't doubt for one second that Intel and AMD could have made a lower clocked, very wide CPU with high IPC if they wanted to. But clockspeed is also important to overall performance, especially in heavier workloads; and so is SIMD, SMT, caches etcetera.
So Intel can develop CPU with double IPC to Comet Lake for 3 GHz but they don't want to? Jezeez, wake up man. Do you remember when Intel tried to enter tablet and smart phone market with Atom cores? What a disaster it was! In-order 2xALU core with half IPC with terrible power consumption in compare to ARMs. AMD Bobcat/Jaguar was similar disaster (2xALU OoO).

  • Double IPC means ..... half speed for the same performance.
  • Half speed means ..... 4x lower power consumption

Intel and AMD would love to have CPU with double IPC like Apple has but they don't know how to do that:
  1. Rising clocks and more cores is the easiest way to increase perf
  2. Widening the SIMD vector units is also pretty simple
  3. Increasing IPC is very hard way to increase perf
  4. Increasing IPC while using less transistors and power is even harder

ARMs delivering higher IPC while using less power (iso-process, iso-freq):
  • Cortex A72 vs A57 .... +15% IPC and -20% power
  • Cortex A73 vs A72 .... +10% IPC and -20% power
  • Cortex A77 vs A76 ........... +20% IPC and +17% transistors (IPC per transistor increased)
  • Intel Ice Lake vs Coffee ... +18% IPC but +38% transistors (IPC per transistor DECREASED)


But I agree 100% that IPC is pretty irrelevant.
No offense but everybody who doubt IPC importance don't understand CPU design a tiny bit. Don't worry you are not alone.


Apple can't juice A14 to 5 GHz anymore than Intel can double their IPC by targeting 3 GHz
  • Apple IPC increase (A7 Cyclone 2013 up to A14 2020) .............. 2.7x
  • Apple clock increase (A7 1.3GHz up to A14 2.8 GHz) ................. 2.2x


  • Intel IPC increase (Sandy Bridge 2011 up to Comet Lake 2020) ............. 1.34x
  • Intel clock increase (Sandy Bridge 5 GHz up to Comet Lake 5.3 GHz) .... 1.06x

Tell me who is stagnating with delivering clock speed for decade? Intel and AMD. ARM chips are delivering pretty huge frequency jumps every single node improvement due to their natural efficiency (not thermally limited). Every ARM can go to 4 GHz pretty easy when it is manufactured on HP process instead of HD version.

7nm Cortex A72 runs at 4 GHz with chiplets on interposer:
This single proof-of-concept chiplet system successfully demonstrates the key technologies for building an HPC System-On-Chip (SoC) with Arm-based cores operating at 4GHz in a 7nm FinFET process.

A72_A57_A15_efficiency_comparison.png

A73_IPC_uplift.jpg

A73_power_eff_uplift.jpg
 

Thala

Golden Member
Nov 12, 2014
1,096
430
136
If you actually read back through the thread, there are many valid criticisms of his table that he completely ignores. Not the least of which he chose his AMD and Intel examples carefully and refuses to update his table with any GB5 results for Intel and AMD that show significantly higher points per clock.
So your claim is the numbers are not accurate? Is there any particular entry you have better information? Just to be clear, it is important that the author carefully chose the results, as there are too many bogus or over-clocked uploads in Geekbench. Ideally you take the numbers from a trusted review site.
 
Last edited:

Hitman928

Platinum Member
Apr 15, 2012
2,915
2,454
136
So your claim is the numbers are not accurate? Is there any particular entry you have better information? Just to be clear, it is important that the author carefully chose the results, as there are too many bogus or over-clocked uploads in Geekbench. Ideally you take the numbers from a trusted review site.
I'm not going to re-hash this debate again, it's all in the thread.


Edit: Also, just to point out, overclocked numbers don't help when the table is ppc. In fact, usually overclocks hurt ppc.
 

Richie Rich

Senior member
Jul 28, 2019
470
227
76
If you actually read back through the thread, there are many valid criticisms of his table that he completely ignores. Not the least of which he chose his AMD and Intel examples carefully and refuses to update his table with any GB5 results for Intel and AMD that show significantly higher points per clock.
People wanted to cheat. I remember one guy downclocked his x86 CPU to 2.6 GHz similar to Apple, probably boosted uncore and DDR mem to maximum and claimed score higher about +20% than my table. This is cheating. Apple core can also provide higher IPC at 1 GHz. GB database is full of tweaked and OC'ed systems with wrongly reported frequency. I try to use GB results from reviews as much as possible to get reasonable numbers. I stated this table is chart of IPC under normal maximum performance to avoid down clocking (cheating).

I added also 32-bit A7, A15 which is just for fun and rough comparison as these results are recalculated from SPEC2000 because GB5 is 64-bit only. But for some people like me is useful to see how the CPU performance evolves in time at iso-proces and iso-frequency.

This IPC table helped me to understand where the insane IPC of Apple cores comes from. It was not magic but rather a lot of super hard work year by year. Apple started from zero with Jim Keller and went to hero. At least as inpiration that anybody can create star-up company like PA-semi or Nuvia and kick Intel's giant ass.
 
  • Like
Reactions: Lodix

Hitman928

Platinum Member
Apr 15, 2012
2,915
2,454
136
People wanted to cheat. I remember one guy downclocked his x86 CPU to 2.6 GHz similar to Apple, probably boosted uncore and DDR mem to maximum and claimed score higher about +20% than my table. This is cheating. Apple core can also provide higher IPC at 1 GHz. GB database is full of tweaked and OC'ed systems with wrongly reported frequency. I try to use GB results from reviews as much as possible to get reasonable numbers. I stated this table is chart of IPC under normal maximum performance to avoid down clocking (cheating).

I added also 32-bit A7, A15 which is just for fun and rough comparison as these results are recalculated from SPEC2000 because GB5 is 64-bit only. But for some people like me is useful to see how the CPU performance evolves in time at iso-proces and iso-frequency.

This IPC table helped me to understand where the insane IPC of Apple cores comes from. It was not magic but rather a lot of super hard work year by year. Apple started from zero with Jim Keller and went to hero. At least as inpiration that anybody can create star-up company like PA-semi or Nuvia and kick Intel's giant ass.
No, there are plenty of valid examples of much higher ppc for Intel and AMD cores on GB5 and you've been given several of them, you just choose to ignore them.
 
  • Like
Reactions: lobz

xblax

Member
Feb 20, 2017
46
52
61
So your claim is the numbers are not accurate? Is there any particular entry you have better information? Just to be clear, it is important that the author carefully chose the results, as there are too many bogus or over-clocked uploads in Geekbench. Ideally you take the numbers from a trusted review site.
Not saying that the results in the table are invalid, but in general Geekbench results really cannot be trusted. I get 8% increase in Geekbench 5 Single Core with my Ryzen 2600 by just going from Windows to Linux (stock settings). I believe it was even worse before the latest Windows scheduler updates.

Windows: https://browser.geekbench.com/v5/cpu/4017338
Linux: https://browser.geekbench.com/v5/cpu/4017796

Browsing through Ryzen 2600 results I guess i'm easily in the top 5% percentile although not much was done to achieve that.
 
  • Like
Reactions: Tlh97

Richie Rich

Senior member
Jul 28, 2019
470
227
76
Not saying that the results in the table are invalid, but in general Geekbench results really cannot be trusted. I get 8% increase in Geekbench 5 Single Core with my Ryzen 2600 by just going from Windows to Linux (stock settings). I believe it was even worse before the latest Windows scheduler updates.

Windows: https://browser.geekbench.com/v5/cpu/4017338
Linux: https://browser.geekbench.com/v5/cpu/4017796

Browsing through Ryzen 2600 results I guess i'm easily in the top 5% percentile although not much was done to achieve that.
Fair point. There is scheduler task shuffling and OS boost control in play. Nothing's perfect.

Problem is that people tried to say those Apple A13's +82% IPC advantage is due to different OS. Which is non-sense. I accept that Win10 can hurt performance by 8% in compare to Linux like in your example. But I'm not gonna accept that iOS is able to unleash 80% of performance from nowhere.

Those 82% of performance per GHz of Apple's A13 clearly comes from huge core:
  • A13: 6xALUs ......... Zen2: 4xALU
  • A13: 6xShifter ........Zen2: ? but Ice Lake has 2xShifter
  • A13: 2xDIV ............. Zen2: 1xDIV
  • A13: 2xMUL ........... Zen2: 1xMUL, Ice Lake 2xMUL
  • A13: 3xbranch ....... Zen2: 2xbranch
You can see most execution units are doubled, some of them tripled (shifters). Apple has the most brutal uarch I've ever seen. It's not only 6-wide but probably 6xALU+3xBranch = 9-wide uarch. No wonder A14 Firestorm can deliver double IPC in compare to Zen2 and Intel Comet Lake (9-wide is 2.25x wider integer core in compare to 4-wide AMD and Intel). Similar to Cortex X1 using separate Branch ports config 4xALU+2xBranch:

A78-X1-ALU.png


Zen2:
Zen2_ALU_AGU_FPU.png

Ice Lake / Sunny Cove:
SunnyCove_backend_ALU.jpg


Apple A12 Vortex, similar to A13:
6xALU_uarch_A11-A12_comparison.png
 
Last edited:
  • Like
Reactions: Lodix

name99

Senior member
Sep 11, 2010
253
226
116
I think the major part of why so many people are bothered by how Richie Rich frames the entire IPC argument, is how it often ignores other aspects of CPU performance which are equally as important. I don't doubt for one second that Intel and AMD could have made a lower clocked, very wide CPU with high IPC if they wanted to. But clockspeed is also important to overall performance, especially in heavier workloads; and so is SIMD, SMT, caches etcetera.

You see the same thing in GPUs. For many years GPUs were stuck in sub ghz clock speeds, and rose slowly over time. Now they're hitting close to, and over 2ghz. And this is for 3D rendering which is often said to be "embarrassingly parallel."

What matters more than anything is the overall performance of a CPU. I guess we'll see which strategy wins out in the next few years.
The reason why IPC is a more important metric than frequency FOR UNDERSTANDING THE FUTURE has been beaten to death. (Hint IPC grows with more transistors and so with future nodes, whereas it's very difficult to get more frequency out of newer processors.)

If you want to refute this reality, explain why it is incorrect, why you believe future nodes will result in impressive frequency gains even though no-one in the fab space claims this.
Don't insult the rest of by implying that we are all idiots who are too dumb to understand the concept of multiplying IPC by frequency to get actual performance.
 

name99

Senior member
Sep 11, 2010
253
226
116
Those 82% of performance per GHz of Apple's A13 clearly comes from huge core:
  • A13: 6xALUs ......... Zen2: 4xALU
  • A13: 6xShifter ........Zen2: ? but Ice Lake has 2xShifter
  • A13: 2xDIV ............. Zen2: 1xDIV
  • A13: 2xMUL ........... Zen2: 1xMUL, Ice Lake 2xMUL
  • A13: 3xbranch ....... Zen2: 2xbranch
Be careful with loose language. The Apple cores has lots of transistors, yes. But it is only "huge" relative to ARM cores. It is still smaller than Intel and AMD cores.
Why?
Because those use fast transistors which are a lot larger than the high efficiency transistors used by Apple. There's a reason Apple hits 2 to 3x the density of Intel on what's supposed to be essentially the same process.

Basically Apple is making all the sensible choices based on what today's tech offers: transistors don't get much faster, so don't rely on that; but they do keep getting smaller, so do rely on being able to provide ever more logic for each new core.
 

Doug S

Senior member
Feb 8, 2020
356
510
96
It relies solely on GB5 (though he was using SPEC earlier, which is arguably worse). It doesn't take into account anything mentioned by @Carfax83 .

@Doug S

If Apple can challenge something like a 9900k with a low-power laptop version of A14 in absolute performance then kudos to them, and I await the benchmarks proving it. I can only lament that Apple's CPU designs are trapped within their software ecosystem. MacOS isn't quite as bad, but still. Bleh. At least we can run some proper benchmarks like Blender on A14.
In what world is BLENDER a better benchmark than SPEC? :rolleyes:
 
  • Haha
  • Like
Reactions: Lodix and Zucker2k

Thala

Golden Member
Nov 12, 2014
1,096
430
136
No, there are plenty of valid examples of much higher ppc for Intel and AMD cores on GB5 and you've been given several of them, you just choose to ignore them.
How do you know these examples are valid, if they do not come from a review, where the test settings are clearly described? Sounds to me as if the OP has a very good reason to ignore such reports.
I believe, that the requirement, that the numbers have to come from a trusted review site should be the minimum standard to be used here.
 
Last edited:
  • Like
Reactions: Lodix

Carfax83

Diamond Member
Nov 1, 2010
5,926
702
126
Well the numbers for the A14 show it beating the best of both Intel & AMD in single core GB5. I know, I know, people will say "but GB5 is (somehow) biased for ARM and against x86". If they can put that A14 in an iPad (and presumably the iPhone 12) then they'll have something at least as fast in the Macs coming out in a few months. They'll be able to run a much wider variety of benchmarks, I think that will make things uncomfortable for the x86 fanboys.
I don't know much about Geekbench, but a quick perusal of the scores on their website shows that there are Intel Tiger Lake score which breach the 1600 point barrier for single core, which is higher than the A14 score reported in the Apple 14 thread.

I've personally only ever ran Geekbench once. The benchmark didn't even get my old 6900K to boost into turbo mode. The highest clock speed it ran at was a little over 3ghz if I recall, and mine is overclocked to 4.3ghz, so it barely stimulated my CPU. Also, it didn't even know that my CPU had a quad channel memory interface. It reported dual channel.

That was some time ago so maybe it's gotten better, but that was a terrible first impression.
 

Thala

Golden Member
Nov 12, 2014
1,096
430
136
I don't know much about Geekbench, but a quick perusal of the scores on their website shows that there are Intel Tiger Lake score which breach the 1600 point barrier for single core, which is higher than the A14 score reported in the Apple 14 thread.
That is because anyone can submit scores using settings way outside of specification. In fact the highest scores are always coming from modders trying to push the limits to the extreme.
 

Hitman928

Platinum Member
Apr 15, 2012
2,915
2,454
136
How do you know these examples are valid, if they do not come from a review, where the test settings are clearly described? Sounds to me as if the OP has a very good reason to ignore such reports.
I believe, that the requirement, that the numbers have to come from a trusted review site should be the minimum standard to be used here.
Reviewers rarely run geekbench for desktop processors because no one cares about geekbench on desktop machines. It gets used in laptop reviews more, but it's still pretty rare. Also, how sure are you that his desktop CPU numbers come from trusted reviews? Or any of his numbers for that matter.

As far as the numbers I posted, it's very easy to see that the numbers I've posted are legitimate when they are in OEM machines, are clearly not overclocked, and are consistent over a large number of tests. That's better than a single number, whether its from a reviewer or not.
 

Carfax83

Diamond Member
Nov 1, 2010
5,926
702
126
Intel and AMD would love to have CPU with double IPC like Apple has but they don't know how to do that:
This comment needs some additional context. Intel stagnated IPC wise mostly because of the disastrous 10nm process node which caused severe delays. Also, the lack of competition from AMD didn't do them any favors either. Their market domination made them complacent. If Intel had nailed the 10nm process, we would probably be on the verge of their 7nm CPUs, which are rumored to provide very large IPC increases.

As for AMD, they just caught up with Intel with Zen 2 so they were way behind. Looks like they are on the right path now though, so I look forward to seeing what Zen 3 and Zen 4 can offer.
 

Thala

Golden Member
Nov 12, 2014
1,096
430
136
As far as the numbers I posted, it's very easy to see that the numbers I've posted are legitimate when they are in OEM machines, are clearly not overclocked, and are consistent over a large number of tests.
It is not easy to see, because Geekbench database entries do not give enough information and in many cases this information is not even reliable.
Your other argument, that if a number is not coming from a review, because the reviewers do not use Geekbench i can absolutely follow. Such entries should at least marked in the table and the methodology how the number is obtained should be described. I do consider using the median as relatively reliable - but calculating the median is quite a bit more work.
 
  • Like
Reactions: Tlh97 and Lodix

Carfax83

Diamond Member
Nov 1, 2010
5,926
702
126
The reason why IPC is a more important metric than frequency FOR UNDERSTANDING THE FUTURE has been beaten to death. (Hint IPC grows with more transistors and so with future nodes, whereas it's very difficult to get more frequency out of newer processors.)
I understand that, but my point was that Intel and AMD prefer to have a higher operating frequencies compared to Apple. Intel and AMD both seem to be targeting around 5ghz with their CPUs. If they didn't think frequency was important, why didn't they design a CPU with lower clock speeds and higher IPC over the years? Same for IBM.

Different markets comes to mind.

If you want to refute this reality, explain why it is incorrect, why you believe future nodes will result in impressive frequency gains even though no-one in the fab space claims this.
It's not incorrect, it's just a different design paradigm. You can have your cake and eat it too. High frequency doesn't mean foregoing high IPC entirely.

Don't insult the rest of by implying that we are all idiots who are too dumb to understand the concept of multiplying IPC by frequency to get actual performance.
Insults weren't my intention. Most of you guys know way more about this subject than myself as a lot of you work in the IT industry, and I don't. I'm just a hardware enthusiast, but I follow the industry news and trends very closely.
 
  • Like
Reactions: Tlh97

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
20,812
8,997
136
This comment needs some additional context. Intel stagnated IPC wise mostly because of the disastrous 10nm process node which caused severe delays. Also, the lack of competition from AMD didn't do them any favors either. Their market domination made them complacent. If Intel had nailed the 10nm process, we would probably be on the verge of their 7nm CPUs, which are rumored to provide very large IPC increases.

As for AMD, they just caught up with Intel with Zen 2 so they were way behind. Looks like they are on the right path now though, so I look forward to seeing what Zen 3 and Zen 4 can offer.
In addition to that.... I object to this thread, as a phone, a laptop, a desktop, a server and HEDT, all have very different targets. a phone needs to be the most efficient and have the least number of cores (actually can have more cores, but power is still the biggest consideration). Usually also run at low ghz to save power. A laptop is not far off of that but needs more cores. Next comes a desktop, that needs more single core speed and cores. Next HEDT, more cores again and more speed. And lastly, a server needs as many cores as it can get, and to run as fast as possible while using low power.

Every one of these targets require a different design. For example, a phone CPU can not run at 5 ghz and be power efficient and cool. This thread does not take ANY of this into account, only what geekbench5 says and divided by ghz. Which is stupid, since as I said, you can't get a phone to run at 5 ghz no matter what (in the foreseeable future).

The only thing that makes sense, is compare
arm to arm
phone to phone
laptop to laptop
desktop to desktop
HEDT to HEDT
server to server.

Now almost any of these categories can be arm or some other technology, but they need to be compared to others that are designed for the same purpose. Do you ever see a reviewer compare a phone CPU to a server ? or a desktop to a server ? of a phone to a desktop, etc.... Nobody with a brain will compare one category to another (with the rare exception that the same CPU fits 2 different categories).

But this thread does
 
Last edited:

ASK THE COMMUNITY