Discussion 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:

 

[LightningZ71]

Note the L3 cache difference...

 

[igor_kavinski]

Det0x's Zen 5 ES vs. M5 Max: https://browser.geekbench.com/v6/cpu/compare/11007418?baseline=16884909

An above average 9950X3D: https://browser.geekbench.com/v6/cpu/compare/16874620?baseline=16884909

Above average 285K: https://browser.geekbench.com/v6/cpu/compare/16706397?baseline=16884909

Clang MT



Raytracer MT



Structure in motion MT



Det0x RAM overclock



9950X3D RAM overclock



285K RAM speed info not available but since it's the Z890 OCF mobo, it's probably also above 9000 MT/s.

 

[511]

Det0x's Zen 5 ES vs. M5 Max: https://browser.geekbench.com/v6/cpu/compare/11007418?baseline=16884909

An above average 9950X3D: https://browser.geekbench.com/v6/cpu/compare/16874620?baseline=16884909

Above average 285K: https://browser.geekbench.com/v6/cpu/compare/16706397?baseline=16884909

Clang MT

View attachment 139357

Raytracer MT

View attachment 139358

Structure in motion MT

View attachment 139359

Det0x RAM overclock

View attachment 139360

9950X3D RAM overclock

View attachment 139361

285K RAM speed info not available but since it's the Z890 OCF mobo, it's probably also above 9000 MT/s.

wish it was linux but RAM Speed is in similar ballpark to M5 which is 9600MT/s also SME will influence the score a bit but still impressive that M5 is achieving that with stock config

 

[MS_AT]

RAM Speed is in similar ballpark to M5 which is 9600MT/s

Is it? Or it is just the MT/s but the width of the interface is different. Otherwise people wouldn't buy macs in droves for local LLMs 😉.

 

[johnsonwax]

Is it? Or it is just the MT/s but the width of the interface is different. Otherwise people wouldn't buy macs in droves for local LLMs 😉.

Not remotely. M5 Max is 8 channel, Pro is quad, Ultra is 16 channel. Are any of those comparisons higher than dual?

 

[igor_kavinski]

wish it was linux

9950X3D: https://browser.geekbench.com/v6/cpu/compare/16435722?baseline=16884909

285K: https://browser.geekbench.com/v6/cpu/compare/15600969?baseline=16884909

Both of these were not easy to find. The usual MT scores for both CPUs are much lower, especially under Linux.

 

[511]

Is it? Or it is just the MT/s but the width of the interface is different. Otherwise people wouldn't buy macs in droves for local LLMs 😉.

M5 achives nearly the same single core score it's MT where this will make a difference

 

[johnsonwax]

M5 achives nearly the same single core score it's MT where this will make a difference

Geekbench doesn't care about memory bandwidth as evidenced from the M5 Pro vs Max benchmarks - they're nearly identical despite the Max having 2x the memory bandwidth so you can't infer anything here.

 

[511]

Geekbench doesn't care about memory bandwidth as evidenced from the M5 Pro vs Max benchmarks - they're nearly identical despite the Max having 2x the memory bandwidth so you can't infer anything here.

I was mostly inferring the ST part that all three Base/Pro/Max achieve nearly the same score for ST. Geekbench 6 MT is just bad we have a 256C GNR System scoring lower than LNL

Intel Corporation AvenueCity - Geekbench

Benchmark results for an Intel Corporation AvenueCity with an Intel Xeon 6980P processor.

browser.geekbench.com

 

[igor_kavinski]

MacOS showing better multicore performance in Clang, Raytracer and Structure in Motion: https://browser.geekbench.com/v6/cpu/compare/16923081?baseline=16856653

 

[johnsonwax]

MacOS showing better multicore performance in Clang, Raytracer and Structure in Motion: https://browser.geekbench.com/v6/cpu/compare/16923081?baseline=16856653

View attachment 139465

And Intel MacOS was/is considered kind of trash compared to Apple Silicon MacOS.

 

[poke01]

I was mostly inferring the ST part that all three Base/Pro/Max achieve nearly the same score for ST. Geekbench 6 MT is just bad we have a 256C GNR System scoring lower than LNL

Intel Corporation AvenueCity - Geekbench

Benchmark results for an Intel Corporation AvenueCity with an Intel Xeon 6980P processor.

browser.geekbench.com

Nobody should be running Geekbench on server CPUs.

MacOS showing better multicore performance in Clang, Raytracer and Structure in Motion: https://browser.geekbench.com/v6/cpu/compare/16923081?baseline=16856653

View attachment 139465

isnt that known? You have to compare with Linux.

Also the Arrow Lake scheduler is optimised for the latest macOS.

 

[igor_kavinski]

isnt that known? You have to compare with Linux.

Still coming out ahead: https://browser.geekbench.com/v6/cpu/compare/16883024?baseline=16870701

 

[igor_kavinski]

Even CachyOS looks helpless: https://browser.geekbench.com/v6/cpu/compare/16883024?baseline=16823473

 

[511]

Even CachyOS looks helpless: https://browser.geekbench.com/v6/cpu/compare/16883024?baseline=16823473

View attachment 139469

I am don't think a 2 year old Core will outscore Apple

 

[poke01]

I am don't think a 2 year old Core will outscore Apple

Igor's comparing a 290K Plus on macOS and Linux

 

[poke01]

@igor_kavinski

Its just geekbench being geekbench. If we take a look at CB2024 on macOS and Windows running on 285K its the same.

Cinebench results Intel Core Ultra 9 285K on macOS Sequoia

Cinebench R23 and 2024 results on macOS Sequoia with Core Ultra 285k. -Processor Intel Core Ultra 9 285K -Mobo MSI PRO Z890-P WIFI

www.insanelymac.com

 

[511]

Igor's comparing a 290K Plus on macOS and Linux

290K Plus what when did it launch

 

[igor_kavinski]

MacOS in a VM trouncing Windows 11 in MT subtests: https://browser.geekbench.com/v6/cpu/compare/16996240?baseline=16997181

 

[igor_kavinski]

Someone's super chilled 285K @ 6.3 GHz and 8000 MT/s vs. 14900KS @ 6.1 GHz and 8800 MT/s: https://browser.geekbench.com/v6/cpu/compare/17014177?baseline=16241169

Following MT subtests probably suffering due to lower RAM speed on the 285K:

 

[igor_kavinski]

Medusa ES @ 2070 MHz: https://browser.geekbench.com/v6/cpu/17130734
Ryzen 365 @ 4720 MHz: https://browser.geekbench.com/v6/cpu/17177556
Core Ultra 338H @ 4650 MHz: https://browser.geekbench.com/v6/cpu/16982967






Unless there is a frequency reporting error, Medusa ES is showing almost double single core points per 100 MHz compared to Zen 5 and Panther Lake.

Look at the Medusa multicore scaling in Clang, Raytracer and Structure from motion. Either Medusa has much higher memory bandwidth or SMT efficiency has been improved considerably.

 

[adroc_thurston]

Medusa ES @ 2070 MHz:

Those are dense clks.
The thing for bigs is fused at 3.4G or therein. Low 3s basically.

 

[gdansk]

Unless there is a frequency reporting error

The frequency is not an error but it's not the core running the test.
It's basically a useless leak since we can't determine the actual clock rates.

 

[igor_kavinski]

Those are dense clks.
The thing for bigs is fused at 3.4G or therein. Low 3s basically.

Ah. Let's take 3 GHz for the P-core then.



And with 3.4 GHz:



Hmmm....does look saner with 3.4 GHz.

 

[Abwx]

Ah. Let's take 3 GHz for the P-core then.

View attachment 140460

And with 3.4 GHz:

View attachment 140461

Hmmm....does look saner with 3.4 GHz.

Methink that it s 3.1GHz, because there s a sample at GB running at this frequency according to the OPN.