Question Geekbench 6 released and calibrated against Core i7-12700

[igor_kavinski]

Geekbench 6 - Geekbench Blog

www.geekbench.com

Weird choice of baseline CPU and even weird is that the baseline score is 2500.

i7-12700 does hardly 2000 in GB5 with the fastest DDR5.

EDIT: Gideon's browser extension/add-on: https://forums.anandtech.com/thread...d-against-core-i7-12700.2610597/post-41452115

Helps you make sense of the results if you are left scratching your head about why a certain test is fast on a certain CPU.

Super useful!

Sample screenshot:

 

[igor_kavinski]

an even more absurd argument that it is "hard to write good AVX-512 code".

It was mentioned in one of the reviews on AT that only a handful of people on the planet know how to write good, performant AVX-512 code. If the code written by the GB dude isn't the best code to do the job, then it's not an accurate representation of the CPU's true potential.

Anyway, this discussion is kinda moot. No one uses GB to do actual work. It's a synthetic benchmark pretending to be about real workloads by approximating those workloads through isolated tests. M4 can stay the fastest FruitBench CPU for all I care. It won't make me dump my money on an Apple device (unless that money just dropped into my lap and I didn't have to toil day and night to earn it).

There are plenty of things that Apple is good at that I don't care about and plenty of things x86 is good at that I care about but Apple doesn't. It's just two different worlds and two different ways of thinking.

 

[igor_kavinski]

If it actually was true that writing SME code was much easier than writing AVX-512 code, wouldn't that be an indication of strength for ARM - that real world code would be more likely to realize those benefits?

Guess we'll see when QC introduces SME in their WoA CPUs.

 

[poke01]

M4 can stay the fastest FruitBench CPU for all I care. It won't make me dump my money on an Apple device (unless that money just dropped into my lap and I didn't have to toil day and night to earn it).

There are plenty of things that Apple is good at that I don't care about and plenty of things x86 is good at that I care about but Apple doesn't. It's just two different worlds and two different ways of thinking.

It’s also the the fastest CPU in SPEC and soon in Cinebench. Not just GB. If an x86 chip is faster than M4 it just means it had optimisations to that software like we see in the open benchmark suite. When both M4 and x86 CPU have optimisations in a piece of software, the M4 will come out ahead.

 

[Doug S]

It’s also the the fastest CPU in SPEC and soon in Cinebench. Not just GB. If an x86 chip is faster than M4 it just means it had optimisations to that software like we see in the open benchmark suite. When both M4 and x86 CPU have optimisations in a piece of software, the M4 will come out ahead.

He's been taking so much copium he's been building up resistance, it is starting to affect his sanity. If he wants to believe Geekbench is the only place where M4 leads and he can build up a reality where it is an irrelevant benchmark that may be what he needs to preserve his sanity.

 

[igor_kavinski]

If he wants to believe Geekbench is the only place where M4 leads

My beef is with GB, not really M4. It's probably (because it can't be perfect at everything) a good chip and I would love to see more benchmarks of it in Phoronix or other stuff. The M4 needs to exist because it shows other companies what is possible to achieve.

 

[H433x0n]

My beef is with GB, not really M4. It's probably (because it can't be perfect at everything) a good chip and I would love to see more benchmarks of it in Phoronix or other stuff. The M4 needs to exist because it shows other companies what is possible to achieve.

You really shouldn’t have a beef with Geekbench - it is the best benchmark available for end users that is free of charge.

If a CPU doesn’t perform well in it, that’s more on the CPU itself. Theres definitely a solid argument that the nT portion of benchmark isn’t perfect but nT performance in real world applications doesn’t scale like cinebench.

 

[igor_kavinski]

If a CPU doesn’t perform well in it, that’s more on the CPU itself.

And the developer is absolved of all responsibility? Who has audited his code to ensure that the best optimizations have been used for that particular CPU?

I'll shut up when Intel/AMD publicly say that GB6.3 is an accurate benchmark.

 

[okoroezenwa]

I'll shut up when Intel/AMD publicly say that GB6.3 is an accurate benchmark.

Why in the world do you need Intel or AMD to tell you whether a benchmark is accurate?? This is very bizarre.

 

[igor_kavinski]

Why in the world do you need Intel or AMD to tell you whether a benchmark is accurate??

Because it's being treated as THE benchmark by the fruit users, not A benchmark.

 

[Nothingness]

And the developer is absolved of all responsibility? Who has audited his code to ensure that the best optimizations have been used for that particular CPU?

Do you really think CPU makers don't discuss with PrimateLabs given how widely it's used?

I'll shut up when Intel/AMD publicly say that GB6.3 is an accurate benchmark.

They'll never say that because that'd be admitting they're behind Apple.
I will let you search what AMD and Intel officially say of other benchmarks.

You don't know what happens behind the doors between CPU companies and benchmark makers.

 

[igor_kavinski]

You don't know what happens behind the doors between CPU companies and benchmark makers.

If a benchmark maker goes, "I do not want any external meddling in my "objective" benchmark" and refuses to co-operate, the CPU company can't exactly say to them, "Stop allowing my CPU to be benchmarked in your benchmark!", can they?

Here's the thing, I wouldn't have suspicions about GB unless I had noted two things (which seem like pretty big red flags to me):

12700 was used as the baseline for GB6 when a faster CPU was available, the Apple M2. (You can see that in the blogpost that he posted a screenshot of M2 running the bench). He prefers Intel, whether that is personal preference or "financially motivated" preference, he didn't reveal in that post. I will choose to think that at that time, he was paid by Intel or at the very least, given a free Intel machine with 12700. If even the latter was the case, he should have declared that. If it's his personal preference, he should clearly say why he chose 12700.

Then with the latest version, he included SME support, just in time for the M4 reviews. Why would he be so concerned about Apple now than when he chose 12700 as the baseline instead of M2?

In order to vindicate himself in my eyes,

1) He has to explain his choices as a show of transparency.

2) GB7 should have the CPU with the fastest IPC as the baseline. Or, he can choose to NOT have a baseline and just some metric like operations completed per second etc.

EDIT: My current feeling is that he is pandering to whoever is paying him more which puts a big question mark on the objectivity of his benchmark results.

 

[Nothingness]

If a benchmark maker goes, "I do not want any external meddling in my "objective" benchmark" and refuses to co-operate,

Again you have no clue how CPU companies and PrimateLabs interact. Of course they might disagree with some decisions PrimateLabs make, but that's the case for all benchmarks including SPEC CPU. But CPU makers certainly discuss with PrimateLabs, they have no choice.

the CPU company can't exactly say to them, "Stop allowing my CPU to be benchmarked in your benchmark!", can they?

AMD does quote Geekbench in its announcements. Intel also officially uses Geekbench.

And yes CPU makers should have full control over what benchmark makers do. Sorry, that's a very silly argument.

If Intel and AMD have objections about some benchmark they should publicly make a statement. If they don't it means they have little background to do it. But since they publicly use Geekbench it means they implicitly endorse it.

 

[Nothingness]

I forgot to answer this part.

12700 was used as the baseline for GB6 when a faster CPU was available, the Apple M2. (You can see that in the blogpost that he posted a screenshot of M2 running the bench). He prefers Intel, whether that is personal preference or "financially motivated" preference, he didn't reveal in that post. I will choose to think that at that time, he was paid by Intel or at the very least, given a free Intel machine with 12700. If even the latter was the case, he should have declared that. If it's his personal preference, he should clearly say why he chose 12700.

You don't care about the reference point that much. He likely chose an Intel chip because that was more widespread than M2 and because that was a fast enough machine he had access to and that was dedicated to run things; you don't set a reference point on a machine that is under usage (he showing a screen with M2 running GB6 just tells us he likely is developing on Mac).

That's exactly how I work: I develop on a laptop and when I have to run benchmarks I run them on a dedicated headless workstation even if it's slower to reduce interactions with spurious processes.

EDIT: BTW Intel/AMD lovers have been claiming for years that GB is an Apple benchmark. Now you say they're being paid by Intel.

 

[igor_kavinski]

EDIT: BTW Intel/AMD lovers have been claiming for years that GB is an Apple benchmark. Now you say they're being paid by Intel.

Paid by Intel at the time when he chose 12700.

Paid by Apple to ensure his revised benchmark version released just in time for M4 reviews.

No proof. Just speculation.

 

[okoroezenwa]

EDIT: BTW Intel/AMD lovers have been claiming for years that GB is an Apple benchmark. Now you say they're being paid by Intel.

On this forum specifically it seems to be a thing to claim GB is Intel-favoured.

 

[okoroezenwa]

Paid by Intel at the time when he chose 12700.

Paid by Apple to ensure his revised benchmark version released just in time for M4 reviews.

No proof. Just speculation.

Why'd you leave out paid by AMD/Intel by adding AMX instructions in 6.1?

Also, releasing 6.3 on April 11th was more than a month before M4 reviews. That's not "just in time".

 

[igor_kavinski]

Why'd you leave out paid by AMD/Intel by adding AMX instructions in 6.1?

Also, releasing 6.3 on April 11th was more than a month before M4 reviews. That's not "just in time".

I don't know If AMD has AMX support in their CPUs.

He could've held off the SME changes until the end of the year when M4, Zen 5 and Lunar Lake will all be available.

If there were multiple CPUs with SME being released, I would get that and it wouldn't seem "suspicious".

 

[thunng8]

I don't know If AMD has AMX support in their CPUs.

He could've held off the SME changes until the end of the year when M4, Zen 5 and Lunar Lake will all be available.

If there were multiple CPUs with SME being released, I would get that and it wouldn't seem "suspicious".

Never seen such ridiculous arguments presented.

Geekbench tries to support the latest features in CPUs. Should it just ignore it and present a vanilla benchmark?

If that is the case, it should remove all extension for every CPU out there to be “fair”. But that total ignores what real application will optimise for.

So you will have a pure benchmark that does not represent applications. That would be a while is useless benchmark.

 

[igor_kavinski]

If that is the case, it should remove all extension for every CPU out there to be “fair”.

That's actually NOT a bad idea.

They can call it Geekbench Vanilla!

It will reflect the performance of code written by first semester CS students!

 

[igor_kavinski]

So you will have a pure benchmark that does not represent applications. That would be a while is useless benchmark.

It is not representing any application at the moment either, other than maybe Clang.

 

[thunng8]

It is not representing any application at the moment either, other than maybe Clang.

Haha .. with response like this without any proof whatsoever, I doubt any sane person can take you seriously.

 

[Nothingness]

It is not representing any application at the moment either, other than maybe Clang.

I think I've had enough.

 

[CouncilorIrissa]

Paid by Intel at the time when he chose 12700.

Paid by Apple to ensure his revised benchmark version released just in time for M4 reviews.

No proof. Just speculation.

Benchmarks are updated all the time when a new CPU is released.
Y-cruncher's creator is set to update his bench with Zen 5 optimisations later this month, for example. Is he paid by AMD? No.

Apple's cores performance is not exclusive to Geekbench suite, they perform well everywhere. The anger is misplaced and is best directed at AMD's/Intel's unwillingness/inability to compete in 1T with an iPad of all things.

 

[igor_kavinski]

The anger is misplaced and is best directed at AMD's/Intel's unwillingness/inability to compete in 1T with an iPad of all things.

It's more confusion and suspicion than anger.

AMD/Intel are not unwilling to compete. It's just that the GB dev was in a hurry to release his SME optimizations. I'm just wondering what prompted him to be in such a hurry? As for inability to compete in ST? Yeah, OK that's possible and I don't dispute that.

There's a reason why I'm calling it FruitBench rather than IntelBench. The developer was motivated to do this work before the launch of a new CPU. He timed the new version release accordingly. Is he going to release a new version before Zen 5 launch? Before Lunar Lake launch? These are not difficult questions.

 

[igor_kavinski]

Y-cruncher's creator is set to update his bench with Zen 5 optimisations later this month, for example. Is he paid by AMD? No.

y-cruncher is a program that can compute Pi and other constants to trillions of digits.

It has a USEFUL output. It's not just a useless benchmark. The developer is passionate about optimization. He cares about Pi digits and that's what his optimizations make possible in lesser and lesser time.