Question Geekbench 6 released and calibrated against Core i7-12700

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

igor_kavinski

Lifer
Jul 27, 2020
28,173
19,202
146
Last edited:
  • Like
Reactions: Gideon, Kaluan and Saylick
E

Exist50

Platinum Member
Aug 18, 2016
2,452
3,106
136
richardskrad said:
OpenAI is right now the premiere AI company in silicon valley. Why are they giving all their employees Macs and not Windows laptops with Intel and AMD chips?
Click to expand...
They're also running their actual compute on x86 (+ Nvidia) platforms...

Anyway, Macs have been popular with programmers since well before the switch to Apple Silicon. They're a native Unix environment in a (typically) premium device. Apple Silicon is great, sure, and Intel/AMD face competitive pressure to close the gaps, but it hardly invalidates them overnight.
 
  • Like
Reactions: lightmanek and Captante
Markfw

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
27,284
16,123
136
Nothingness said:
Geekbench 6.1 has been released: https://www.xda-developers.com/geekbench-6-1-released/
  • Clang 16
  • SVE/AVX512-FP16
  • Fixed-point math
  • Improved multi-core performance (I guess this one will appease some... or not :) )
Click to expand...
Seeing as how my 96 core Genoa was beat by an Alderlake something CPU mul multicore, I am not even going to bother trying it, due to this comment "and multi-core scores to be raised by up to 10%.".

Now if they said by 2000% or something I might get excited.
 
  • Haha
Reactions: DisEnchantment, Glo., Tlh97 and 5 others
moinmoin

moinmoin

Diamond Member
Jun 1, 2017
5,248
8,463
136
Nothingness said:
Improved multi-core performance (I guess this one will appease some... or not :) )
Click to expand...
Yeah, that's just plain silly. Either stop pretending it's reflecting a CPUs (instead the test case's) multi-core performance, or change it to something embarrassingly parallel.

But I'm sure many people now are happy GB 6.1 improved their chip's multi-core performance...
 
  • Like
  • Love
  • Haha
Reactions: Glo., Tlh97, uzzi38 and 2 others
N

Nothingness

Diamond Member
Jul 3, 2013
3,315
2,386
136
moinmoin said:
Yeah, that's just plain silly. Either stop pretending it's reflecting a CPUs (instead the test case's) multi-core performance, or change it to something embarrassingly parallel.
Click to expand...
You almost made it sound as if multi-core performance is only measured by embarrassingly parallel tasks. If it's what people want, then they should use some rendering or parallel compilation benchmark.

Or focus on the subscores of GB that show how these tasks scale:

ASUS System Product Name - Geekbench

Benchmark results for an ASUS System Product Name with an AMD Ryzen 9 7950X3D processor.
browser.geekbench.com

Screenshot from 2023-06-09 10-00-06.png
  • clang: x14
  • ray tracer: x19
Looks like good scaling for a 16-core CPU.

Thing is we don't have enough information about bottlenecks (are these due to a benchmark issue [implementation or just inherently not scalable] or a CPU issue?).
All I could find was this: https://www.geekbench.com/doc/geekbench6-cpu-workloads.pdf and it doesn't tell anything about the MP aspects.
That's not GB specific, all closed source benchmarks suffer from that.

And another issue is that scaling changed dramatically between GB 5 and GB 6. And I personally think it's a good thing as GB 5 scaling didn't mean anything to me.

ASUS System Product Name - Geekbench

Benchmark results for an ASUS System Product Name with an AMD Ryzen 9 7950X3D processor.
browser.geekbench.com
Screenshot from 2023-06-09 10-23-27.png

moinmoin said:
But I'm sure many people now are happy GB 6.1 improved their chip's multi-core performance...
Click to expand...
:joycat:

The problem is some think a single aggregated number can represent a CPU performance. And definitely GB 6 MP number doesn't. Like any other benchmark.
 
W

whoshere

Member
Feb 28, 2020
45
99
91
itvision.altervista.org
coercitiv said:
For a simple refresh the gains are quite nice, though I hope GB7 brings a healthy generational +25% ST performance uplift! :D

View attachment 81534
Click to expand...

For a benchmark this just means that the results of the previous version cannot be compared with the results of 6.1. The title of this news piece is pure trash.

Nothingness said:
Geekbench 6.1 has been released: https://www.xda-developers.com/geekbench-6-1-released/
  • Clang 16
  • SVE/AVX512-FP16
  • Fixed-point math
  • Improved multi-core performance (I guess this one will appease some... or not :) )
Click to expand...

MT scaling is still abysmal. The core issue of GB6 being "I prefer this particular CPU with its very limited number of cores, and I don't care about any additional cores people may have" remains. And this particular CPU may very well be the basic Apple M1 (4P+4E cores).

And there's a regression, bravo John Poole! The Navigation test is now slower than in v6.0.

Generic: Geekbench 6.1.0 vs Generic: Geekbench 6.0.0 - Geekbench

browser.geekbench.com browser.geekbench.com
 
Last edited:
  • Like
Reactions: moinmoin
N

Nothingness

Diamond Member
Jul 3, 2013
3,315
2,386
136
whoshere said:
MT scaling is still abysmal. The core issue of GB6 being "I prefer this particular CPU with its very limited number of cores, and I don't care about any additional cores people may have" remains. And this particular CPU may very well be the basic Apple M1 (4P+4E cores).
Click to expand...
Yes, in real life MT scaling often is abysmal. Look at subscores for things you're interested in rather than focus on the global score; see my other post above, some of the tests scale quite well.
 
  • Like
Reactions: ikjadoon
W

whoshere

Member
Feb 28, 2020
45
99
91
itvision.altervista.org
Nothingness said:
Yes, in real life MT scaling often is abysmal. Look at subscores for things you're interested in rather than focus on the global score; see my other post above, some of the tests scale quite well.
Click to expand...

1. No one will compare subscores.
2. GB5 scaled near perfectly.
3. GB6 strongly, stubbornly, to its death even prefers very particular very limited core configurations - well, that's not a MT benchmark, that's John Poole's benchmark. Remove the "MT" mark from it and people like me will be fine.

We've had 11 pages to discuss this and the consensus has been reached. I don't understand why you wanna argue again.

I will continue to dismiss GB6 "MT" results with prejudice. Considering this peculiarity I'm not inclined to trust its ST results either. GB5 was known to demonstrate very tangible and straightforward differences between systems. GB6 turns everything upside down.
 
  • Like
Reactions: Tlh97, igor_kavinski and moinmoin
E

eek2121

Diamond Member
Aug 2, 2005
3,423
5,070
136
whoshere said:
For a benchmark this just means that the results of the previous version cannot be compared with the results of 6.1. The title of this news piece is pure trash.



MT scaling is still abysmal. The core issue of GB6 being "I prefer this particular CPU with its very limited number of cores, and I don't care about any additional cores people may have" remains. And this particular CPU may very well be the basic Apple M1 (4P+4E cores).

And there's a regression, bravo John Poole! The Navigation test is now slower than in v6.0.

Generic: Geekbench 6.1.0 vs Generic: Geekbench 6.0.0 - Geekbench

browser.geekbench.com browser.geekbench.com
Click to expand...

tl;dr it models multithreaded workloads more realistically, hence why multicore scores are lower.

Perhaps a better way would be to show both scores so users have a good idea of both theoretical and actual multicore performance.
 
  • Like
Reactions: Lodix, Tlh97, Captante and 1 other person
W

whoshere

Member
Feb 28, 2020
45
99
91
itvision.altervista.org
eek2121 said:

tl;dr it models multithreaded workloads more realistically, hence why multicore scores are lower.

Perhaps a better way would be to show both scores so users have a good idea of both theoretical and actual multicore performance.
Click to expand...
And this has been discussed here as well ad neouseum. There are a dozen ways to limit MT performance (including and not limited to memory performance; inter-core communications including inter-core latency; cache size; threads synchronization, etc), GB6 uses just one of them and calls it a day. Good. Fine. Excellent. Now is there a single mention of that on the website? In the program? Anywhere?

When you offer a universal multi-threaded benchmark (which, surprise, surprise GB5 actually was) you make sure it scales. GB6 MT almost does not scale beyond a certain number of cores.

If you're extremely interested in certain "realistic" workloads which scale poorly, you bench them yourselves. You don't go around screaming "My particular workload means 128 core EPYC systems are complete crap and a waste of money!" which is exactly what GB6 MT does, 'cause you know Apple matters!

gb6 mt.png

Now compare this to GB5 MT:

gb5 mt.png

Can you find Apple M2 Ultra, no? It's at the rock bottom.

Sources:

 
Last edited:
  • Like
  • Love
Reactions: lightmanek, Lodix, Panino Manino and 6 others
I

igor_kavinski

Lifer
Jul 27, 2020
28,173
19,202
146
GB 6.2

Apple 6S Plus
599 ST
900 something MT

Motorola Edge 30
1068 ST
3013 MT

Both on battery power.

Motorola Edge 30 (with charger)
1071 ST
3022 MT
 
D

Doug S

Diamond Member
Feb 8, 2020
3,650
6,464
136
igor_kavinski said:
GB 6.2

Apple 6S Plus
599 ST
900 something MT
Click to expand...


It is kind of amazing to me that even after the A9's massive 70% performance jump in a single year, since then Apple has managed to go a full 5x faster in only 8 years in ST performance. That's despite their rather lackluster gains the last few years.

That almost seems too good to believe...was that 599 throttling? Is yours the TSMC or Samsung chip?
 
L

Lodix

Senior member
Jun 24, 2016
340
116
116
igor_kavinski said:
GB 6.2

Apple 6S Plus
599 ST
900 something MT

Motorola Edge 30
1068 ST
3013 MT

Both on battery power.

Motorola Edge 30 (with charger)
1071 ST
3022 MT
Click to expand...
Phones don't unlock more performance for being on the charger LOL. It is the other way around, they can throttle more because of the heat generated by charging.
 
I

igor_kavinski

Lifer
Jul 27, 2020
28,173
19,202
146
Doug S said:
That almost seems too good to believe...was that 599 throttling? Is yours the TSMC or Samsung chip?
Click to expand...
It didn't seem hot so I'm sure it was fine. No easy way to check the chip type. There used to be some battery monitor app that revealed the CPU codename but that's not available on the app store anymore.
 
I

igor_kavinski

Lifer
Jul 27, 2020
28,173
19,202
146
Lodix said:
Phones don't unlock more performance for being on the charger LOL. It is the other way around, they can throttle more because of the heat generated by charging.
Click to expand...
Seems you are right, with respect to the iPhone 6S Plus only. On charger, it delivered 521 ST and 485 MT. The Moto Edge 30 seems to not get hot enough with the charging so there is no throttling and it even improves the score.

I turned on the AC, took the iPhone out of the case, stood it up with support at a place where the cold air from the AC was flowing above and ran the test again. This time, it managed 635 ST and 964 MT. Did the same thing for the Moto Edge 30 and this time, it underperformed with 1063 ST and 2998 MT. Maybe it doesn't like being cold :)
 
A

Abwx

Lifer
Apr 2, 2011
11,888
4,874
136
Doug S said:
It is kind of amazing to me that even after the A9's massive 70% performance jump in a single year, since then Apple has managed to go a full 5x faster in only 8 years in ST performance. That's despite their rather lackluster gains the last few years.
Click to expand...
1.85GHz for the A9 and 4.05 for the M3, that s 2.19x the frequency, guess that a similarly clocked A9 would be still quite competitive currently even if not high end.
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom