Question Geekbench 6 released and calibrated against Core i7-12700

[igor_kavinski]

Geekbench

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.

 

[scineram]

Here's my 5900x with PPT lowered to ~120W (I don't remember exactly off the top of my head) and slight negative offsets with curve optimizer.

View attachment 76467

Gigabyte Technology Co., Ltd. X570 AORUS ELITE - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. X570 AORUS ELITE with an AMD Ryzen 9 5900X processor.

browser.geekbench.com

Pathetic ST you got there.🤪

 

[Sulaco]

13900k

 

[Exist50]

I think these threading changes make sense. Instead of trying to split the difference and being mediocre at both server and client workloads, which are increasingly divergent, they've decided to focus on making it a good test suite for what most users (client) actually do with their devices, which frankly matches most of what people reference Geekbench for.

 

[Carfax83]

@Sulaco, does your CPU boost to 5.8ghz during the single threaded portion of the run?

 

[Timmah!]

Here are the ST Speeds reported on that Run..!

Geekbench Browser

View attachment 76468

Clearly within Stock settings, perhaps very good cooling.

All the cores can do 5,7? I realize its not at the same time, but i thought the second CCD cores cant clock as high as the first one.

 

[JoeRambo]

One more datapoint: Fixed 5Ghz clock 7950x 16C16T with tuned 6200CL36 mem:

Gigabyte Technology Co., Ltd. B650M AORUS ELITE AX - Geekbench Browser





Mostly let down by immature BIOS and the need to run 2x32GB too.

I then proceeded to run 13900KS system @ fixed 5Ghz too and it scored very similar:

Micro-Star International Co., Ltd. MS-7D25 - Geekbench

Benchmark results for a Micro-Star International Co., Ltd. MS-7D25 with an Intel Core i9-13900KS processor.

browser.geekbench.com

So there we have it, clock for clock, Zen4 vs RPL*

*RPL run might have had some issues due to uncore clocks not boosting to 5Ghz as voltage for 5Ghz is peanuts compared to 6/5.6 operating points.

 

[Sulaco]

@Sulaco, does your CPU boost to 5.8ghz during the single threaded portion of the run?

Yep!

 

[Carfax83]

Yep!

What cooling are you using?

*Edit* Nevermind I just saw your sig!

No idea why mine is stuck at 5.5ghz. Maybe I need to reduce my undervolt

 

[Sulaco]

What cooling are you using?

*Edit* Nevermind I just saw your sig!

Just what's in my sig...NH-D15 and stock Lian Li case fans. Using a thermalright contact frame too for whatever thats worth

 

[SPBHM]

my poor 3GHz Sandy Bridge delivering 558 in single core

 

[Saylick]

my poor 3GHz Sandy Bridge delivering 558 in single core

😮 And I thought *my* phone was slow.

 

[JoeRambo]

😮 And I thought *my* phone was slow.

What is not fun, is that fruity vendor truly and utterly destroying other ARM SOC vendors in times, not percents.

 

[Carfax83]

Just what's in my sig...NH-D15 and stock Lian Li case fans. Using a thermalright contact frame too for whatever thats worth

Yeah I saw that. I checked the core frequencies again, but this time I used HWInfo and my CPU did indeed boost to 5.8ghz on two cores. I used CPU-Z the first time and that reported a max frequency of 5.5ghz.

I guess there is probably some runtime variance because my last score had 3099, just shy of 3100. Also, I suspect you may have slightly better temps than me because you have the Noctua NH-D15.

 

[moinmoin]

On one hand it makes GB6 useless for servers/workstations. But it does make it even better for normal consumers. Really these many core CPUs do only provide minimal gains for most people.

That's the reasoning behind userbenchmark. GB6 is by far not as bad, ST scores are fine, but the usefulness of the MT scores took a huge nosedive.

 

[Det0x]

Very light benchmark, hardly any benefit enabling SMT on a 7950x.
(full information in the pictures below)

SMT disable: (16 cores / 16 threads)



SMT enable: (16 cores / 32 threads)



BTW i'm pretty sure .gb6 at the end of the url is showing average cpu clockspeeds now, not highest like in the previous geekbenches..

Lastly i also did a openCL run with my plain old 3090. (still wating for 4090TI)

 

[Carfax83]

😮 And I thought *my* phone was slow.

View attachment 76492

Here is my Pixel 7 Pro:

 

[nicalandia]

Yeah. SMT is not favored on this GB6 Benchmark. Can anyone please disable HT on a 13900K and test it?

 

[Markfw]

OK< this is rediculous. Dual 7763 EPYC Milans CPU's, 128c/256t It gets beat in multicore by a 13600k ????

There is something seriously wrong with this software.

Supermicro Super Server - Geekbench

Benchmark results for a Supermicro Super Server with an AMD Eng Sample: 100-000000314-04_30/16_N processor.

browser.geekbench.com

1393
Single-Core Score
11427
Multi-Core Score

 

[Muadib]

Using Eco mode, with memory set to Expo.

 

[moinmoin]

MT score seems to highly depend on ST score somehow, for most scores so far of current chips with 8 cores or (even significantly) more the MT score is about 6-8 times the ST score. The only exception I saw so far is the 64c ARM server posted on last page with a ratio of ~13.

 

[Markfw]

MT score seems to highly depend on ST score somehow, for most scores so far of current chips with 8 cores or (even significantly) more the MT score is about 6-8 times the ST score. The only exception I saw so far is the 64c ARM server posted on last page with a ratio of ~13.

But seriously, a MT score ratio of 13 on a 64c server ? and a MT score ratio of 8 on a 128c/256t server ? How can this work for anything higher than a 13600k or a 7600x ?

 

[MrTeal]

But seriously, a MT score ratio of 13 on a 64c server ? and a MT score ratio of 8 on a 128c/256t server ? How can this work for anything higher than a 13600k or a 7600x ?

That's the best part. It doesn't.

 

[Abwx]

That's the reasoning behind userbenchmark. GB6 is by far not as bad, ST scores are fine, but the usefulness of the MT scores took a huge nosedive.

At the same time one has to wonder why GB did shrink the eventual buyer s population, that s not the best way to sell more of their stuff, quite the contrary, and so much that it s somewhat suspicious...

 

[Markfw]

At the same time one has to wonder why GB did shrink the eventual buyer s population, that s not the best way to sell more of their stuff, quite the contrary, and so much that it s somewhat suspicious...

Yes, not being able to use it to sell servers, HEDT, workstation, or even more than low to mid desktop, seems like they took their use population down to a very small subset.

 

[eek2121]

and my 2 256 thread systems ? I can't compare them. If what it does only works for middle of the line desktop workloads, what good is it ?

If my understanding of the new calculation is correct, you can compare them between GB6 runs.

They aren't 'ratios' or anything like that. The multicore workloads are based on real life performance, and scores are weighted as such. Zen 4 and RPL-S happen to score well because they have either several high frequency cores (Zen 4) or a mix of extremely high frequency cores and a bunch of slow ones (RPL-S). Either way they both bruteforce enough to get the job done.

Ignore benchmarks and specialty applications for a moment. Look at the workloads they run. Run them manually, you'll understand.

I thought their choice of calibration using a modern CPU was a bit odd, but given that GB5 was launched 5 years ago, I get it. 5 years from now, desktop CPUs will look very different.

Zen 4 Threadripper will very likely tell the complete story. A combination of high clocked Zen 4 cores should, in theory, beat out the 7950X significantly.

I do hope that someone here is able to play with it a bit more. If not, I hope to this weekend. I suspect that once you start to eliminate variables (frequency and/or small cores) you will see why things actually add up.

I am in no way defending them either. I thought GB5 was a great benchmark, but previous GB versions weren't. The new benchmark is unproven. It likely will be soon-ish.