Question Geekbench 6 released and calibrated against Core i7-12700

Page 37 - 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
27,305
18,762
146
Last edited:
  • Like
Reactions: Gideon, Kaluan and Saylick
L

LightningZ71

Platinum Member
Mar 10, 2017
2,452
3,125
136
Not really surprised. I9 class product with 8 big cores, slightly higher boost clock, more L3 and AVX-512 support vs. the lowest end Arrow Lake i3 class processor... Yeah, it SHOULD be better in MT at this point.

In MT, 10 core comet lake was better in MT than the lowest I3 class Alder/Rocket lake parts. The 9900KS demolishes dual core Rocket Lake and Tiger Lake parts in MT.
 
T

Tigerick

Senior member
Apr 1, 2022
819
783
106
Supporting SME in WoA will push up the scores. By how much, we shall know.

Meanwhile, PTL by 18A might not even support AVX512, hmm. See how bad the process will push Intel down to the bottom of performance crown...
 
Jan Olšan

Jan Olšan

Senior member
Jan 12, 2017
571
1,127
136
To no one's surprise, more use of SME.

I'd expect that the tasks that were optimised (manually even?) for SME should show quite a speedup from 512bit AVX-512 units too, if the benchmark was equally optimised for x86...

(edit: sorry for the tyops)
 
Last edited:
  • Haha
  • Like
Reactions: Nothingness and igor_kavinski
5

511

Diamond Member
Jul 12, 2024
3,986
3,689
106
M

mikegg

Golden Member
Jan 30, 2010
1,943
547
136
Jan Olšan said:
To no one's surprise, more use of SME.

I'd expect that the tasks that were optimised (manually even?) for SME should show quite a speedup from 512bit AVX-512 units too, if the benchmark was equally optimised for x86...

(edit: sorry for the tyops)
Click to expand...
You mean like when Zen4 got 242.1% higher uplift in Object Detection over Zen3 via AVX upgrade?

That's higher than the 227.9% that M4 Max got over M3 Max.

I didn't see you complain then.

Zen3 vs Zen4: https://browser.geekbench.com/v6/cpu/compare/13703871?baseline=13703302

M3 Max vs M4 Max: https://browser.geekbench.com/v6/cpu/compare/13690938?baseline=13701855

1757309737945.png
 
Last edited:
  • Like
Reactions: Nothingness
poke01

poke01

Diamond Member
Mar 8, 2022
4,045
5,371
106
Jan Olšan said:
I'd expect that the tasks that were optimised (manually even?) for SME should show quite a speedup from 512bit AVX-512 units too, if the benchmark was equally optimised for x86...
Click to expand...
Geekbench IS optimised for x86. as @mikegg pointed out. AVX-VINNI and its 512-bit version do speed up Object detection.

There was no discussion around AI/ML extensions when Zen4 came out. Its only brought up when Apple introduced SME2 last year and the drama that followed.

1757310845106.png

Thats a massive difference, should we exclude that test because its an "outlier".
 
5

511

Diamond Member
Jul 12, 2024
3,986
3,689
106
There is a difference between SME/AVX-512 SME is not part of an individual core it is shared by the entire CPU while AVX-512/VNNI is per core and when comparing single core performance how can you use a shared processing unit. It's like one core is using the ALU of another core
 
O

okoroezenwa

Member
Dec 22, 2020
157
154
116
511 said:
There is a difference between SME/AVX-512 SME is not part of an individual core it is shared by the entire CPU while AVX-512/VNNI is per core and when comparing single core performance how can you use a shared processing unit. It's like one core is using the ALU of another core
Click to expand...
That has nothing to do with SME vs AVX-VNNI and everything to do with how Apple implemented SME. Nothing says that ARM/QC/whoever else will do what Apple did.

Also this continues to be a funny deflection from the fact that GB added support for VNNI and SVE before SME. And no one had a problem with that.
 
  • Like
Reactions: poke01
D

Doug S

Diamond Member
Feb 8, 2020
3,461
6,142
136
mikegg said:
Zen3 vs Zen4: https://browser.geekbench.com/v6/cpu/compare/13703871?baseline=13703302

M3 Max vs M4 Max: https://browser.geekbench.com/v6/cpu/compare/13690938?baseline=13701855
Click to expand...


I was looking at those comparisons and just to get off the boring object detection dispute, what's going on with structure from motion? Zen4 gained only 4.7% in that benchmark despite higher clock rate, larger L2, and upgrade to DDR5 versus gaining 26% in Clang (something neither AVX512 nor SME2 will help)

Meanwhile M4 gained 17.7% in structure from motion versus 14.5% in Clang. Since the Mac comparison was GB6.5 vs GB6.5 it doesn't seem like this could have had anything to do with new instructions M4 added - according to the Geekbench PDF structure from motion uses AVX2 and NEON, so neither AVX512 nor SME2 will help.

So why such a minuscule gain in structure from motion on Zen4 despite all the advantages, and clearly being a rather big upgrade based on Clang, when M4 saw a gain in structure from motion very similar to the gain from Clang (and its overall gain, if you excluded SME2) What is happening with that benchmark that seems to have confounded Zen4? Looking at Zen5 numbers which were over 30% higher for it, whatever that was AMD fixed it.
 
poke01

poke01

Diamond Member
Mar 8, 2022
4,045
5,371
106
okoroezenwa said:
That has nothing to do with SME vs AVX-VNNI and everything to do with how Apple implemented SME. Nothing says that ARM/QC/whoever else will do what Apple did.
Click to expand...
I would think ARM's SME would be inside the core since they sell core IP.
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom