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:

 

[Geddagod]

To be fair ARM got a LOT better between the time they announced the Nuvia acquisition and time the first fruit of that was released a year ago.

Good point

That "crowding" between Apple, Qualcomm and ARM might be because they've all mostly exploited the same ideas.

Something interesting, Apple and ARM seem to be going high IPC and lower clocks vs Qcomm going higher clocks and relatively lower IPC... but then Apple and Qualcomm use the same general caching setup (and interestingly enough, Tenstorrent too) while ARM uses the more traditional one.
Apple seems to have the best of both worlds lol.

 

[gdansk]

phones

This is the laptop part, yeah? We don't know that the clock rate is accurate. But it being roughly correct is a very simple explanation to the low results in e.g. ... all the non-SME subtests.

What's even less reliable than GB is wherever the 4.7GHz comes from. Even if that rumor originated with Qualcomm they missed last time..

 

[MS_AT]

This is the laptop part, yeah?

The part in question is meant for smartphones. I mean Snapdragon 8 Elite 2 that the leak is about.

 

[coercitiv]

Lowkey if I was Qcomm and my expensive cpu team acquisition can't differentiate themselves well from the stock ARM stuff, since that didn't really happen this gen with the x925, I would start getting mad lol.

The expensive team failing to deliver is the easy answer, but not a complete certainty. The new team still had to integrate in a complex organization, and that cost (in time, stress, politics etc) may reflect on the Geekbench sheet more than we'd like.

I don't understand why they're pushing so high with clocks, though I guess that makes more sense if they plan to use the arch for PC devices as well. Still a recipe for disaster though, my guessometer would argue high speed designs fail more often than they deliver.

 

[gdansk]

The part in question is meant for smartphones. I mean Snapdragon 8 Elite 2 that the leak is about.

Oh, right. No X.
I forgot they 1337'd their phones.

 

[CouncilorIrissa]

This is the laptop part, yeah? We don't know that the clock rate is accurate. But it being roughly correct is a very simple explanation to the low results in e.g. ... all the non-SME subtests.

What's even less reliable than GB is wherever the 4.7GHz comes from. Even if that rumor originated with Qualcomm they missed last time..

4.74 is the figure reported by CPU itself.

We still don't know what that "4050" clock rate means, and whether that's accurate since we've all seen GB6 produce wildly wrong frequency numbers on unreleased parts. If it can run at 4.7 GHz why wasn't it tested at that speed? Or is it that it can "run" at that speed, but only in one of those ridiculous phones that has a fan in it? I don't buy the "its an engineering sample" argument. Its not far from release and it is on N3P which is effectively a mature process from day one - TSMC never has yield issues with their iterations. So if it can only run at 4 GHz they can't blame the process on that, it would indicate some shortcomings in design.

Very true, GB6 clock readings are wildly inconsistent. Given that the predecesessor already pulls considerable amount of power @ 4.37, I wouldn't be surprised if 4.7 is only achievable in actively cooled devices (or in a fridge lmao)

 

[igor_kavinski]

Here's a laptop I ignored for a year because I blamed LPDDR4 for its high latency: https://forums.anandtech.com/threads/post-your-maxxmem²-score.2608515/post-41492029

Here's the Geekbench 6.2 comparison before and after fixing the laptop's performance by nuking power saving settings in the BIOS: https://browser.geekbench.com/v6/cpu/compare/13297047?baseline=5554294

I feel sorry for the suckers living with the OOB lame performance of this laptop. It may have even contributed to the sales failure of this model. If only the engineers had provided a turbo button on the laptop to provide boosted performance on AC power instead of forcing the user to figure out how to unlock the dormant performance...

Notice the tests that loved the latency being halved:

HDR
File Compression
Navigation
PDF Renderer

 

[poke01]

Here's a laptop I ignored for a year because I blamed LPDDR4 for its high latency: https://forums.anandtech.com/threads/post-your-maxxmem²-score.2608515/post-41492029

Here's the Geekbench 6.2 comparison before and after fixing the laptop's performance by nuking power saving settings in the BIOS: https://browser.geekbench.com/v6/cpu/compare/13297047?baseline=5554294

I feel sorry for the suckers living with the OOB lame performance of this laptop. It may have even contributed to the sales failure of this model. If only the engineers had provided a turbo button on the laptop to provide boosted performance on AC power instead of forcing the user to figure out how to unlock the dormant performance...

Notice the tests that loved the latency being halved:

HDR
File Compression
Navigation
PDF Renderer

Here’s a good benchmark for memory latency etc, been playing with it. Will share my results soon.

GitHub - mbntr/PYPrime-2.x: PYPrime is an open source python based CPU and RAM benchmark

PYPrime is an open source python based CPU and RAM benchmark - mbntr/PYPrime-2.x

github.com

If you want to run on Linux do not use the complied version. Compile it your self cause the complied version is using python 3.7 and messes up and gives a lib shared error.

macOS is easy too just follow the steps listed in the readme file.

For the windows version, I recommend running it from Benchmate.

 

[igor_kavinski]

For the windows version, I recommend running it from Benchmate.

I'll try compiling it in Windows with aggressive optimizations if possible (python libs and include folder may pose a challenge ).

Would be fun to see if the cheating leads to any good gains