Question Benchmarks.... PassMark vs. Geekbench

[kirkdickinson]

I have always checked PassMark for benchmarks, but fell into GeekBench last night. The top performing CPU's in the lists look a lot different. Wondering why that is? Maybe I am on the wrong pages of those sites.

PassMark Intel vs AMD CPU Benchmarks - High End

PassMark Software - CPU Benchmarks - Over 1 million CPUs and 1,000 models benchmarked and compared in graph form, updated daily!

www.cpubenchmark.net

Processor Benchmarks - Geekbench

Geekbench has the 9900KF at the top above all the Ryzens, but PassMark has the 9900KF down a ways.

Wondering what is going on here?

 

[Markfw]

The geekbench scores look ridiculous. A 22 core OLD Xeon (E5-2696v4) at 2.2 ghz beats a 2990wx 32 core 3.0 ghz chip ?. There are other examples, but that one is enough.

 

[kirkdickinson]

The geekbench scores look ridiculous. A 22 core OLD Xeon (E5-2696v4) at 2.2 ghz beats a 2990wx 32 core 3.0 ghz chip ?. There are other examples, but that one is enough.

Maybe their test is skewed toward the Intel chipsets??? But still!

 

[Hitman928]

Geekbench is very sensitive to memory bandwidth and probably bugs out with the 2990wx's multiple numa node configuration.

 

[keysplayr2]

Been out of the loop for a LOOOOONG time. Am I to understand that AMD has taken back the performance crown with a vengeance? And Wassup MarkFW.

 

[itsmydamnation]

Been out of the loop for a LOOOOONG time. Am I to understand that AMD has taken back the performance crown with a vengeance? And Wassup MarkFW.

neck and neck, but with AMD's much faster execution and far smaller number of dies by this time next year AMD will likely be in the lead with Zen3.

 

[keysplayr2]

neck and neck, but with AMD's much faster execution and far smaller number of dies by this time next year AMD will likely be in the lead with Zen3.

This is good! Intel needed a kick in the rear. Stagnating for almost a decade.

 

[Markfw]

Been out of the loop for a LOOOOONG time. Am I to understand that AMD has taken back the performance crown with a vengeance? And Wassup MarkFW.

In multi-thread and especially server, they are kicking Intel in the arse. And for value ? They are killing it.

Good to see you back, why did you now use your regular keysplayer account ? If you can't remember the password, start a thread in moderator discussions and ask that it be reset,. Then we can delete the new account and get back all your old goodness.

 

[YBS1]

Been out of the loop for a LOOOOONG time. Am I to understand that AMD has taken back the performance crown with a vengeance? And Wassup MarkFW.

Well that begs the question, what kept you away for so long from not only the forum but just general tech news that you had no clue AMD was back in the race?

 

[keysplayr2]

Well that begs the question, what kept you away for so long from not only the forum but just general tech news that you had no clue AMD was back in the race?

Life changes. And walking away was probably a good thing at the time. I had enough of the fights and vitriol. Thought I'd check out AT and see old friends. Last I upgraded my rig was a few years ago. 7600K. I gotta say it wasn't that noticeably faster than the 2500K that it replaced. I was considering another upgrade and when I saw benchmarks for new AMD and EPYC??? Wow!! Impressed.

 

[keysplayr2]

In multi-thread and especially server, they are kicking Intel in the arse. And for value ? They are killing it.

Good to see you back, why did you now use your regular keysplayer account ? If you can't remember the password, start a thread in moderator discussions and ask that it be reset,. Then we can delete the new account and get back all your old goodness.

I don't even think my old account exists or is active any longer. I tried it but says account doesn't exist.

 

[DrMrLordX]

Geekbench is very sensitive to memory bandwidth and probably bugs out with the 2990wx's multiple numa node configuration.

GB4 was worse in that regard. GB5 doesn't even modify the score based on memory subsystem scores (anymore). So it should be less problematic.

edit: I will say that maybe mem is still playing a big part in the background though. Here's my 3900x (not overclocked) with DDR4-3733 CAS/CL14:

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

Benchmark results for a Gigabyte Technology Co., Ltd. X570 AORUS MASTER with an AMD Ryzen 9 3900X processor.

browser.geekbench.com

I have also modified LLC settings downward and added a negative voltage offset to improve boost performance, so that may be an additional factor. The scores that Primate Labs lists for the 3900x seem kinda low.

 

[Ajay]

The geekbench scores look ridiculous. A 22 core OLD Xeon (E5-2696v4) at 2.2 ghz beats a 2990wx 32 core 3.0 ghz chip ?. There are other examples, but that one is enough.

Looks like the benchmark can only handle 1/2 the available cores.

 

[SAAA]

Or maybe there's just an issue with Passmark scores for Zen 2? Just look at previous gen Zen:

AMD Ryzen 7 PRO 3700 Benchmark

AMD Ryzen 7 2700X Benchmark

I don't think Ryzen 3700 is 40% faster than Ryzen 2700 in single anything save some specific loop code.