Question Geekbench 6 released and calibrated against Core i7-12700

Page 15 - 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
24,605
17,100
146
Last edited:
  • Like
Reactions: Gideon, Kaluan and Saylick
Sort by date Sort by votes
I

igor_kavinski

Lifer
Jul 27, 2020
24,605
17,100
146
Upvote 0 Downvote
N

Nothingness

Diamond Member
Jul 3, 2013
3,292
2,358
136
Upvote 0 Downvote
FlameTail

FlameTail

Diamond Member
Dec 15, 2021
4,384
2,757
106
That means perhaps the upcoming ARM Cortex X5 will support SME2.

Cortex X5 announcement is very near (next month?), and Android chips with it will be available by the end of the year.
 
Upvote 0 Downvote
D

Doug S

Diamond Member
Feb 8, 2020
3,168
5,430
136
igor_kavinski said:
github.com

GitHub - corsix/amx: Apple AMX Instruction Set

Apple AMX Instruction Set. Contribute to corsix/amx development by creating an account on GitHub.
github.com github.com



Now why would Apple refuse to admit they exist?
Click to expand...

Apple doesn't refuse to admit they exist, they want people to use library calls to use them. That way they can change the way the instructions work without breaking old software. Nothing stops someone from using AMX instructions, but if their code breaks with M4 they can't complain because Apple told them not to use the instructions directly.
 
  • Like
Reactions: scannall, Nothingness and Viknet
Upvote 0 Downvote
I

igor_kavinski

Lifer
Jul 27, 2020
24,605
17,100
146
Doug S said:
Nothing stops someone from using AMX instructions, but if their code breaks with M4 they can't complain because Apple told them not to use the instructions directly.
Click to expand...
But then aren't they preventing the use of these instructions in open source development? Or are their libraries using these instructions available for Linux?
 
Upvote -1 Downvote
D

Doug S

Diamond Member
Feb 8, 2020
3,168
5,430
136
igor_kavinski said:
But then aren't they preventing the use of these instructions in open source development? Or are their libraries using these instructions available for Linux?
Click to expand...

Apple doesn't care about Linux on ARM Mac. They haven't impeded the efforts of the Asahi Linux crew but they sure aren't making extra work for themselves by directly supporting them either. So no, they won't make those libraries available for Linux.

Do you think the (at the very most) one person out there interested in running Linux on an ARM Mac to do tons of matrix multiplication is going to be pissed they can't get the best performance out of it due to the inability to use AMX?
 
  • Like
Reactions: moinmoin, Tlh97 and scannall
Upvote 0 Downvote
G

gdansk

Diamond Member
Feb 8, 2011
4,128
6,835
136
So dismissive. There are people who would like to run a different operating system on a Mac and sometimes within it too.
I do miss Bootcamp but WoA works (virtualized) on M chips. It doesn't need amx but super secret features only known by reverse engineering are a step back to the Appendix H days.
 
Last edited:
  • Like
Reactions: Nothingness and igor_kavinski
Upvote 0 Downvote
D

Doug S

Diamond Member
Feb 8, 2020
3,168
5,430
136
gdansk said:
So dismissive. There are people who would like to run a different operating system on a Mac and sometimes within it too.
I do miss Bootcamp but WoA works (virtualized) on M chips. It doesn't need amx but super secret features only known by reverse engineering are a step back to the Appendix H days.
Click to expand...

If it benefits Apple's main customers - the ones who run iOS/iPadOS/macOS on their hardware - by allowing Apple to make non-compatible changes to the AMX instructions but disadvantages the tiny minority running Linux I think even the ones running Linux would concede that's what Apple should do.

Now whether Apple actually does make such non compatible changes is another matter, but they obviously feel it is important that they have the ability. They aren't doing it just to be jerks - there are some things they've done (for example the bootloader, read the Asahi blog for the details) that have really helped the Asahi effort. Not claiming they did those things with that goal in mind, but if they wanted to make it impossible or nearly so to run an unsupported OS on macOS hardware they easily could have been jerks and done so.
 
Upvote 0 Downvote
G

gdansk

Diamond Member
Feb 8, 2011
4,128
6,835
136
No, not documenting it doesn't benefit Apple's customers at all. It prevents some from using their hardware.

Having it documented wouldn't hurt anything, it is still unsupported. But instead you'll have to do HLE for a VM instead of allowing running software to issue AMX instructions directly. It should be good enough but it'll never be done.

And internally they already have such documentation but don't release it. Because it's Apple. The company that exploits BSD devs because they can and never gives back.
 
Last edited:
  • Like
Reactions: igor_kavinski
Upvote 0 Downvote
I

igor_kavinski

Lifer
Jul 27, 2020
24,605
17,100
146
Interesting example from my company of how a non-risky and wise decision by a "suit" can deprive a company from acquiring the proper computing resources:



That HP server that is losing so badly cost $8000+ WITHOUT a Windows Server license. My pleas to go with Epyc Milan or even Genoa fell on deaf ears coz the regular hardware vendor does not provide those.
 
  • Like
  • Love
Reactions: Tlh97, Captante and Markfw
Upvote 0 Downvote
I

igor_kavinski

Lifer
Jul 27, 2020
24,605
17,100
146
GB6 can be improved by running the accelerated tests in non-accelerated form as an extra data point and when doing comparisons, it can show both against CPUs that lack acceleration to make things seem more fair.
 
Upvote 0 Downvote
T

TwistedAndy

Member
May 23, 2024
159
150
76
Geekbench is a terrible benchmark to compare different platforms.

When you compare two platforms and want to be objective, you need to ensure that both platforms do not use additional optimizations. SME gives one platform a benefit over the other, and it is used in three tests.

AVX512-VNNI, as well as AMX, are supported only on some server Intel CPUs and are used in two tests. A more generic AVX-512 set is available on some server Intel processors and recent AMD ones but is not available on consumer Intel products (except Tiger Lake). That also does not make the comparison more objective.

All those cases make the GB6 results not comparable not only across different platforms (ARM and x86) but within each platform as well. You can't compare them objectively. And this is what the Geekbench team wrote in the release notes for 6.3.

Some may argue that some applications may benefit SMEs, AVX-512, etc. And they will be right. But in this case, it makes sense to measure the performance of those apps directly. It will be way closer to real life than synthetic tests.
 
Upvote 0 Downvote
I

igor_kavinski

Lifer
Jul 27, 2020
24,605
17,100
146
Opened a discussion at Primate Labs requesting new features: http://support.primatelabs.com/discussions/geekbench/85627-suggestion-for-geekbench-64

May I suggest running some of the accelerated tests using AVX-512/SME again in non-accelerated form so it serves as a better comparison point with CPUs that don't feature any acceleration? Also, it would be more informative if the test shows which SIMD extension was used (such as Blur Detection - AVX2) and the same one listed again without using AVX2. Even if this is not implemented for standard testing, it would be helpful to include it as an option that users may want to run and then you can see how popular it is with users who want to know how the subtest performance was achieved.

Related to comparisons in Geekbench Browser, may I suggest comparisons against more than one device, with UI similar to the one Intel ARK uses (checkbox for Add to Compare)? Furthermore, it would also be nice to have two separate categories shown for MT tests (one for tests that scale with the number of cores and another for the ones that don't) so there is no confusion and users can immediately jump to the comparison they are more interested in seeing.
Click to expand...
 
  • Like
Reactions: FlameTail
Upvote 0 Downvote
FlameTail

FlameTail

Diamond Member
Dec 15, 2021
4,384
2,757
106
Primate Labs should also add a 2nd type of MT test, the kind they had in GB5 which they kicked to the curb in GB6.

Then we won't have to rely on CB2024 MT anymore.
 
Upvote 0 Downvote
I

igor_kavinski

Lifer
Jul 27, 2020
24,605
17,100
146
FlameTail said:
Primate Labs should also add a 2nd type of MT test, the kind they had in GB5 which they kicked to the curb in GB6.
Click to expand...
They probably won't do that but they do suggest checking the embarrassingly parallel tests like Raytracing for GB5 type MT tests. I've suggested that they at least categorize such tests separately in the results browser so people know that these tests are scaling nicely with the number of their available CPU cores.
 
Upvote 0 Downvote
N

Nothingness

Diamond Member
Jul 3, 2013
3,292
2,358
136
TwistedAndy said:
Geekbench is a terrible benchmark to compare different platforms.

When you compare two platforms and want to be objective, you need to ensure that both platforms do not use additional optimizations. SME gives one platform a benefit over the other, and it is used in three tests.

AVX512-VNNI, as well as AMX, are supported only on some server Intel CPUs and are used in two tests. A more generic AVX-512 set is available on some server Intel processors and recent AMD ones but is not available on consumer Intel products (except Tiger Lake). That also does not make the comparison more objective.

All those cases make the GB6 results not comparable not only across different platforms (ARM and x86) but within each platform as well. You can't compare them objectively. And this is what the Geekbench team wrote in the release notes for 6.3.
Click to expand...
All what you say is applicable to all benchmarks, not only GB. Do you think it makes more sense to compare SPEC FP results of a CPU without AVX against a CPU with AVX?

The main issue is that to really exploit any result from any benchmark, you need knowledge of the benchmark and of the platform you run it on. This goes well beyond what 99% of benchmark users know or even understand. While when you know the benchmark you can surely make objective comparisons across ISA and platforms, including with GB.

But that doesn't mean GB or other benchmarks are useless or terrible. All CPU design companies I know are using GB because they know how to exploit its results (and because their customers want to know). Most of them are also involved with PrimateLabs in the development of Geekbench. In itself that doesn't prove much, but GB is definitely a useful benchmark. Of course it's just one among many others that are used to improve micro architectures.

TwistedAndy said:
Some may argue that some applications may benefit SMEs, AVX-512, etc. And they will be right. But in this case, it makes sense to measure the performance of those apps directly. It will be way closer to real life than synthetic tests.
Click to expand...
Yes. The typical end-user only needs to know how fast her software will run no matter whether the SW/CPU support any or all extensions. But even that might be misleading or not future proof (see Cinebench results on Arm which changed dramatically when Maxon started making proper support for Arm ISA).
 
  • Like
  • Wow
Reactions: Hitman928 and igor_kavinski
Upvote -1 Downvote
poke01

poke01

Diamond Member
Mar 8, 2022
3,473
4,772
106
TwistedAndy said:
Geekbench is a terrible benchmark to compare different platforms.
Click to expand...
Depends which version.
TwistedAndy said:
When you compare two platforms and want to be objective, you need to ensure that both platforms do not use additional optimizations. SME gives one platform a benefit over the other, and it is used in three tests.

AVX512-VNNI, as well as AMX, are supported only on some server Intel CPUs and are used in two tests. A more generic AVX-512 set is available on some server Intel processors and recent AMD ones but is not available on consumer Intel products (except Tiger Lake). That also does not make the comparison more objective.
Click to expand...
Regarding Client:
Intel and AMD benefit from geekbench optimizations more than Apple.

Intel and AMD both support AES-NI, VAES, AVX, AVX2 and AVX_VNNI (upcoming support in Zen 5).
AMD benefits from AES-NI, VAES, SHA-NI, AVX2, AVX, AVX-512, AVX-VNNI(In Zen5) and AVX512-VNNI(In Zen 5).

Notice the higher SIMD sets on x86, this is actually unfair to Apple

1717578087964.png

Apple benefits from ARMv8 AES, ARMv8 SHA1, NEON, DOTPROD, I8MM and SME.
1717578069282.png
TwistedAndy said:
All those cases make the GB6 results not comparable not only across different platforms (ARM and x86) but within each platform as well. You can't compare them objectively. And this is what the Geekbench team wrote in the release notes for 6.3.
Click to expand...
We can compare Apple M4 or later GB 6.3 scores with each other by removing object detection across different plaforms. Since SME provides a huge uplift in ML/AI. But 6.3 scores cannot be compared with 6.1 and 6.2 if the system does not have SME.

TwistedAndy said:
Some may argue that some applications may benefit SMEs, AVX-512, etc. And they will be right. But in this case, it makes sense to measure the performance of those apps directly. It will be way closer to real life than synthetic tests.
Click to expand...
You claim you want to be objective and compare fairly but if an Zen 5 x86 system uses AVX-512 and a ARM system doesn't in Handbrake, its not fair is it?

If you want objectiveness in core architecture IPC uplift you use SPEC.





BTW: The SME unit is really good for AI/ML. The fact it gives such an uplift in Object Detection shows why Apple is fine when it comes to total platform TOPS. Beats out Lunar, X Elite in CPU TOPS. Apple can provide up to 16 TOPS using P cores SME unit which pales in comparison to 5 TOPS provided by AVX-VNNI on Lunar.

1717579686036.png
 
Last edited:
Upvote 0 Downvote
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom