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

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

DrMrLordX

Lifer
Apr 27, 2000
15,742
4,706
136
Heading over to the Geekbench site and looking at top performers, the A13 doesn't even show up as a top performer. If one browses the specific processors such as the a13 and the 3700x the comparison becomes absurd. I don't know much about particulars compared to many here but it seems to me it's bit like golf. If one goes about it right any drunk can beat the best pro out there with the right course and handicap. In the real world, not so much. I understand what you are saying but AFAIK there's not anyone using the A13 because it's faster than everything else.
As far as fixed-function hardware goes, I was being mostly pedantic. You are, of course, correct. Mind telling @Richie Rich for me?
 

Richie Rich

Senior member
Jul 28, 2019
421
196
76
Great finding!

OP, can you please update the World's Most Powerful PPC CPU Top? We have all these new exciting entries!
I see your point but to avoid PPC vs. clock fluctuation lets consider only the highest clocked versions from each core. There is two good reasons for that:
  • It's realistic - as we aim to get max performance from each urach
  • Max clocks is the hard limit given by process and uarch (and I try to compare uarchs among each other). Lower clocks has no limit actually.
I've put that into initial post.
EDIT:
For Coercitiv and others to stop endless discusion about PPC frequency scaling: To have fair and clean comparison I will use only the top (high clocked) version from each core.
If you will continue ignore this rule I will consider that as a inflamatory behavior and you will be reported.


You don't make the rules for a thread. We have posted forum guidelines on what is, and what isn't allowed.
If you post something, users have the right to debate/challenge the claim.

AT Mod Usandthem
 
Last edited by a moderator:

Hitman928

Platinum Member
Apr 15, 2012
2,505
1,638
136
I see your point but to avoid PPC vs. clock fluctuation I will use only high clocked versions from each core. I've put that into initial post.

If you will continue ignore this rule I will consider that as a inflamatory behavior and you will be reported.
Define high clocked because the examples I gave are all much higher clocked than A13 and are running at their boost target frequencies.
 
  • Love
  • Like
Reactions: myocardia and Tlh97

Thunder 57

Golden Member
Aug 19, 2007
1,404
1,178
136
If you will continue ignore this rule I will consider that as a inflamatory behavior and you will be reported.
Inflammatory? Don't you realize you are the one causing trouble? I'm surprised you haven't reported me. Or have you? Also, you don't get to make the rules. Go away, no one wants you here as can be seen by replies to this post.


User insults are not allowed. Stick to debating the data,
or add the user to your ignore list.

AT Mod Usandthem
 
Last edited by a moderator:
  • Love
  • Like
Reactions: myocardia and Tlh97

coercitiv

Diamond Member
Jan 24, 2014
3,725
3,500
136
If you will continue ignore this rule I will consider that as a inflamatory behavior and you will be reported.
You are free to do as you wish.

Meanwhile I will continue to point out that arbitrarily seeking performance targets to subject them to normalization for the purpose of refactoring the numbers again into high performance estimates is one very crooked way of looking at CPU architectures. You're not interested in how Apple CPUs perform in phones. What you really want to talk about is how an imaginary high performance CPU built by Apple/ARM will perform relative to x86 cores. You want the big 8-16+ core 3-4Ghz Apple CPU to prove you right, but can't wait a few years for actual HPC mature silicon. So what can be done now? PPC can be done, to show the potential of a core while ignoring massive multi-core architecture changes, memory subsystem scaling, cost control and power efficiency drops with both high clocks and massive data transfers inside and outside the SOC.

You want PPC? It's fine, I applaud your initiative. It's a great time to start having this discussion, but we will have to do it fair and square, with all publicly available data and without arbitrary requirements such as "highest clocked entry only". If we are to directly compare an architecture built for sub 3GHz operation and one built for anything between 2 and 5Ghz depending on form factor and power/performance targets, then each architecture gets the best case scenario within parameters it was built for.

Having said that, please update the Ice Lake numbers with those provided by @Hitman928 - score of 1461 @ ~3880 Mhz for a PPC of 375+. The CPU is operating near the same frequency as the one in the World's Most Powerful PPC CPU Top, but with ~17% better PPC.
 

aigomorla

Cases and Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
18,043
850
126
Why isn't the asgard computer core included in this pointless comparison, lets put in the Hal9000 while we are on it, or Mother from Aliens? :D

If were going to do pointless comparisons lets put in pointless processors.

Your mixing different platforms with different arch.
Your not going to go get a solid reliable comparison.
This is why most reputable tech sites avoid doing comparisons for what you state.
And if apple's SoC processor was that powerful, the mac server would be on them, yet last time i recall, they still went (insert intel bell melody here).

But they will probably go (insert EYPC MOAR COARS meme here) soon if not next year.
 
Last edited:
  • Love
Reactions: myocardia

lobz

Golden Member
Feb 10, 2017
1,194
1,156
106
I see your point but to avoid PPC vs. clock fluctuation lets consider only the highest clocked versions from each core. There is two good reasons for that:
  • It's realistic - as we aim to get max performance from each urach
  • Max clocks is the hard limit given by process and uarch (and I try to compare uarchs among each other). Lower clocks has no limit actually.
I've put that into initial post.

If you will continue ignore this rule I will consider that as a inflamatory behavior and you will be reported.


You don't make the rules for a thread. We have posted forum guidelines on what is, and what isn't allowed.
If you post something, users have the right to debate/challenge the claim.

AT Mod Usandthem
You could, however, create your own forum called Tim's Kitchen and claim Apple's IPC supremacy without the danger of anyone challenging your flawed logic.
 

soresu

Golden Member
Dec 19, 2014
1,179
443
136
SpecFP2006 is missing I see.

In that benchmark A77 (SD 865) exceeds A11 by 13%, with SpecInt2006 falling behind by just under 10%. While using considerably less power I might add. Overall I would say that ARM Ltd are not so far behind as Apple would like people to think.

 
Last edited:
  • Like
Reactions: myocardia and Tlh97

soresu

Golden Member
Dec 19, 2014
1,179
443
136
OK, just did some basic math based on Anandtech's Spec2006 scores and energy usage.

A13 vs Cortex-A77 perf in Specint2006 = 1.578x
A13 vs Cortex-A77 power in Specint2006 = 1.623x
A13 vs Cortex-A77 perf in Specfp2006 = 1.356x
A13 vs Cortex-A77 power in Specfp2006 = 1.616x

So, yes more raw performance in both fp and int, but a little more power for int, and a lot more for fp. Overall I'd say A77 wins on perf/watt.
 

lobz

Golden Member
Feb 10, 2017
1,194
1,156
106
OK, just did some basic math based on Anandtech's Spec2006 scores and energy usage.

A13 vs Cortex-A77 perf in Specint2006 = 1.578x
A13 vs Cortex-A77 power in Specint2006 = 1.623x
A13 vs Cortex-A77 perf in Specfp2006 = 1.356x
A13 vs Cortex-A77 power in Specfp2006 = 1.616x

So, yes more raw performance in both fp and int, but a little more power for int, and a lot more for fp. Overall I'd say A77 wins on perf/watt.
Any more of this nonsense and he may report you.....

Keep the discussion on the data, and not the user.

AT Mod Usandthem
 
Last edited by a moderator:
  • Like
Reactions: myocardia

UsandThem

Elite Member
Super Moderator
May 4, 2000
12,911
3,487
146
In case anyone is wondering, I've already been in this thread
four times in only two pages of responses.

Going forward, I'd advise all users to keep the discussion
focused, and without insults.

AT Mod Usandthem
 

Richie Rich

Senior member
Jul 28, 2019
421
196
76
You are free to do as you wish.

Meanwhile I will continue to point out that arbitrarily seeking performance targets to subject them to normalization for the purpose of refactoring the numbers again into high performance estimates is one very crooked way of looking at CPU architectures. You're not interested in how Apple CPUs perform in phones. What you really want to talk about is how an imaginary high performance CPU built by Apple/ARM will perform relative to x86 cores. You want the big 8-16+ core 3-4Ghz Apple CPU to prove you right, but can't wait a few years for actual HPC mature silicon. So what can be done now? PPC can be done, to show the potential of a core while ignoring massive multi-core architecture changes, memory subsystem scaling, cost control and power efficiency drops with both high clocks and massive data transfers inside and outside the SOC.

You want PPC? It's fine, I applaud your initiative. It's a great time to start having this discussion, but we will have to do it fair and square, with all publicly available data and without arbitrary requirements such as "highest clocked entry only". If we are to directly compare an architecture built for sub 3GHz operation and one built for anything between 2 and 5Ghz depending on form factor and power/performance targets, then each architecture gets the best case scenario within parameters it was built for.

Having said that, please update the Ice Lake numbers with those provided by @Hitman928 - score of 1461 @ ~3880 Mhz for a PPC of 375+. The CPU is operating near the same frequency as the one in the World's Most Powerful PPC CPU Top, but with ~17% better PPC.
  1. I see your point but you cannot mix two different set of PPC results together, each based on different key. My chart is based on PPC delivered at maximum performance at given platform and constraints (because we seek for max performance).
  2. However feel free to create your own table with chart based on your key. That would be also interesting to read so please do that.

OK, just did some basic math based on Anandtech's Spec2006 scores and energy usage.

A13 vs Cortex-A77 perf in Specint2006 = 1.578x
A13 vs Cortex-A77 power in Specint2006 = 1.623x
A13 vs Cortex-A77 perf in Specfp2006 = 1.356x
A13 vs Cortex-A77 power in Specfp2006 = 1.616x

So, yes more raw performance in both fp and int, but a little more power for int, and a lot more for fp. Overall I'd say A77 wins on perf/watt.
No doubt that A77 is in some ways brilliant design. I like the fact that ARM spent 17% more transistors while gained 20% int IPC gain and 35% FPU IPC gain and kept the best performance per watt ratio. But IMHO when A13 would be clocked at lower speed to achieve same performance as A77 (lets say about 1.4x lower to have balanced INT and FPU speed = 2.65 / 1.4 = 1.9 Ghz) this A13@1.9 GHz would have power consumption much lower than A77 (approximately 1.4^3 = 2.7x lower than A13@2.65 Ghz). When comparing those two CPUs at iso-performance A13@1.9=A77@2.84Ghz then power efficiency will be 2.7 / 1.6 = 1.7x more power efficiency for A13. But this is more server comparison than consumer CPU comparison. No doubt A77 would be No. 1 in TOP10 performance per watt chart.
 

Thunder 57

Golden Member
Aug 19, 2007
1,404
1,178
136
First, I am sorry for causing trouble earlier. I posted in haste and without thinking properly, and that rarely ends well.

Now I have a couple of questions.

1) Why do we care so much about PPC? They don't clock anywhere near each other. PPC is one part of the overall picture. If the Athlon 64 only clocked at 1GHz, it would've been a failure, regardless of PPC. IMHO, the argument you should be making is performance per watt.

2) If you haven't seen it already, check out DrMrLordX's post. The short version of it is that no one would call a "review" using only SPEC and GB data as a review. He posted a link to a proper review.

People want to see a wide sample of data, preferably including software they run. So one CPU may be quickest in SPEC. That's cool I guess, if you like to run SPEC all day. How about seeing a CPU be considerably faster in Handbrake? Maybe you do a good amount of transcoding. Great! Now you know you'll be saving some time. So why limit your results to SPEC and GB?
 

naukkis

Senior member
Jun 5, 2002
331
159
116
1) Why do we care so much about PPC? They don't clock anywhere near each other. PPC is one part of the overall picture. If the Athlon 64 only clocked at 1GHz, it would've been a failure, regardless of PPC. IMHO, the argument you should be making is performance per watt.
Great! Now you know you'll be saving some time. So why limit your results to SPEC and GB?
As didn't Athlon64 clock to near P4 but still performed well. A13 PPC is so good that SOC in phone could rival best use all power they could desktop CPU's. That's about same kind of advantage that A64 had over P4.
 

coercitiv

Diamond Member
Jan 24, 2014
3,725
3,500
136
IMHO, the argument you should be making is performance per watt.
He can't argue for performance per watt while also aiming to prove a theoretical Apple high performance CPU will be superior. The goal is best ST performance in the world, and this PPC classification is great because it "normalizes" performance and then allows the reader to dream of unicorn CPUs.

Notice how he keeps insisting on maximizing clocks within platform constraints, basically comparing x86 chips pushed to their power/thermal limits to ARM chips running on passive cooling frequencies. It's the same discussion we had in the previous threads, only this time he blatantly refuses to even update his World Top with valid results such as the EPYC score provided by me or the Ice Lake score provided by @Hitman928.

He chooses the metric - PPC - but even then feels uncomfortable when Ice Lake jumps from ~320 PPC to ~375 PPC at virtually the same clocks. That's the level of insecurity.
 

naukkis

Senior member
Jun 5, 2002
331
159
116
A13 is also better than in those Richie's results, it's a phone chip and won't sustain it's maximum clocks in heavy stress tests.
 

lobz

Golden Member
Feb 10, 2017
1,194
1,156
106
  1. I see your point but you cannot mix two different set of PPC results together, each based on different key. My chart is based on PPC delivered at maximum performance at given platform and constraints (because we seek for max performance).
  2. However feel free to create your own table with chart based on your key. That would be also interesting to read so please do that
This is deliberate provocation from your side. Anyone trying to debate your 'data points' is just a unwitting participant in this charade. Your chart is not based on anything else than hand-picked results that fit your agenda. Even witeken's self-praise is a more objective forum topic than this.
 

Richie Rich

Senior member
Jul 28, 2019
421
196
76
GeekBench5.1 comparison from 6/22/2020:
  • added Cortex X1 and A78 performance projections from Andrei here
  • 2020 awaiting new Apple A14 Firestorm core and Zen3 core
Updated:

Pos
Man
CPU
Core
Year
ISA
GB5 Score
GHz
PPC (score/GHz)
Relative
1​
Apple​
A13​
Lightning​
2019​
ARMv8​
1332​
2.65​
502.64​
100%​
2​
Apple​
A12​
Vortex​
2018​
ARMv8​
1116​
2.53​
441.11​
88%​
3​
Apple​
A11​
Monsoon​
2017​
ARMv8​
933​
2.39​
390.38​
78%​
4​
ARM Cortex​
X1​
Hera​
2020​
ARMv8​
1055​
3.00​
351.67​
70%​
5​
Apple​
A10​
Hurricane​
2016​
ARMv8​
770​
2.34​
329.06​
65%​
6​
Intel​
1065G7​
Icelake​
2019​
x86-64​
1252​
3.90​
321.03​
64%​
7​
Apple​
A9​
Twister​
2015​
ARMv8​
564​
1.85​
304.86​
61%​
8​
ARM Cortex​
A78​
Hercules​
2020​
ARMv8​
868​
3.00​
289.33​
58%​
9​
AMD​
3950X​
Zen 2​
2019​
x86-64​
1317​
4.60​
286.30​
57%​
10​
ARM Cortex​
A77​
Deimos​
2019​
ARMv8​
812​
2.84​
285.92​
57%​
11​
Intel​
9900K​
Skylake​
2018​
x86-64​
1384​
5.00​
276.80​
55%​
12​
AMD​
1800X​
Zen 1​
2017​
x86-64​
1073​
3.90​
275.13​
55%​
13​
ARM Cortex​
A76​
Enyo​
2018​
ARMv8​
720​
2.84​
253.52​
50%​
14​
Intel​
4770K​
Haswell​
2013​
x86-64​
966​
3.90​
247.69​
49%​
15​
Apple​
A8​
Typhoon​
2014​
ARMv8​
323​
1.40​
230.71​
46%​
16​
Intel​
3770K​
Ivy Bridge​
2012​
x86-64​
764​
3.50​
218.29​
43%​
17​
Apple​
A7​
Cyclone​
2013​
ARMv8​
270​
1.30​
207.69​
41%​
18​
Intel​
2700K​
Sandy Bridge​
2011​
x86-64​
723​
3.50​
206.57​
41%​
19​
Intel​
E6600​
Core2​
2006​
x86-64​
338​
2.40​
140.83​
28%​
20​
AMD​
Phenom 965 BE​
K10.5​
2006​
x86-64​
496​
3.70​
134.05​
27%​
21​
AMD​
Athlon 64 X2 3800+​
K8​
2005​
x86-64​
207​
2.00​
103.50​
21%​
22​
Intel​
Pentium D​
P4​
2005​
x86-64​
228​
3.40​
67.06​
13%​
 

ASK THE COMMUNITY