[SOLVED] Ryzen 5000 higher frequency, but lower performance?

Irisena

Commendable
Oct 1, 2019
94
10
1,565
Hi, so the other day, I undervolted my 5600X's curve by -10, and put PBO at +200. Now I get 4.85 ghz single core and about 4.5-ish all core. But, the most puzzling thing is my cinebench score fell by 80 points compared to stock 4.4 all core and 4.6 single core. Stability is fine though. At stock, I can get 1880 at cinebench R15. After the -10 curve and +200 on PBO, No matter how hard I try, I can only muster about 1800-ish. even though it boost higher. temp is under check with a 280mm AIO hovering around 60-65c on both settings.

So, can anyone explain what the heck is going on? How can a higher frequency translate to worse performance? Is this an undervolting bug just like the ryzen 3000 series?
 
Solution
Yeah, it's odd.

The only thing I could think of right now is to quit all applications irrelevant to the benchmark, I've known them to have a variable impact on full-processor benchmarks myself - This is in the range of 1%-5%, even if they aren't doing anything apparent.

Try again, maybe with another benchmark. I've found Prime95 and y-cruncher to be particularly sensitive.

JWNoctis

Respectable
Jun 9, 2021
443
108
2,090
Did you change any other settings? Are things like memory frequencies and timings the same? What happens when you restore them to what you had? Are there anything else - browsers, gaming platform clients, software and OS updates, etc. - happening to be running in the background, in either case?
 
  • Like
Reactions: Irisena

Irisena

Commendable
Oct 1, 2019
94
10
1,565
Xmp turned on for both test, i double checked each time. All the test done right after startup too, so no opening other software invloved, except what i had at startup, which is the same between both test. No windows update going on at the time, and i'm using the same bios version too. The difference is literally only the pbo settings and curve offset.
 

JWNoctis

Respectable
Jun 9, 2021
443
108
2,090
Yeah, it's odd.

The only thing I could think of right now is to quit all applications irrelevant to the benchmark, I've known them to have a variable impact on full-processor benchmarks myself - This is in the range of 1%-5%, even if they aren't doing anything apparent.

Try again, maybe with another benchmark. I've found Prime95 and y-cruncher to be particularly sensitive.
 
  • Like
Reactions: Irisena
Solution

Irisena

Commendable
Oct 1, 2019
94
10
1,565
Well, that might be a good idea to try. I still have 0 idea what happened as well. I ran the test several times over several power cycle, and the result is more or less consistant, give or take 10 points-ish.

It does scream like another bug in ryzen chips though. I'm more interested whether this is a widespread issue or is it just my particular system