Offsets, Curves, And Overclocking, Oh My!

I have to be a man and eat crow. My apologies to GotNoRice. You were correct and it was my bull-headedness I was overlooking the obvious. Before I removed the board, chip and cpu, I decided to re-try and overclock using the suggestion you posted:

"Just set the power limits according to your motherboard (usually a single setting where you just select "motherboard"), and bring the curve optimizer negative offset lower and lower until you become unstable, then raise it a few points"

I was able to get my 5800X to boost to a max of 5.0Ghz and an average boost between 4.75-4.9Ghz. Ran Vray and CB23 rock steady stable and max temps tapped at 80c My negative offset is -25.

Yes, my butt still hurts.
 

Attachments

  • vray_2.JPG
    vray_2.JPG
    140.2 KB · Views: 0
  • Cinebench23_1.JPG
    Cinebench23_1.JPG
    61.6 KB · Views: 0
  • Cinebench23_2.JPG
    Cinebench23_2.JPG
    61.9 KB · Views: 0
The method for maxxing out these new cpus is just soooo different than previous generations. We all went through the same learning curve so don't be too hard on yourself.
 
That's damn impressive. The thing I would want to know, and I know that it isn't something that matters to everyone, but do those settings hold up under Prime 95?
 
JSHamlet234 said:
That's damn impressive. The thing I would want to know, and I know that it isn't something that matters to everyone, but do those settings hold up under Prime 95?
Click to expand...
For me it did. I tried an offset of -30 and had 2 cores fail. Once I dropped the offset to -25, it was rock steady stable. I might have to reseat the block as my temps hovered around 80-ish under load.

*As for the "sucker" I sold it to, he understood I wanted to keep it...for now.
 
Nebulous said:
For me it did. I tried an offset of -30 and had 2 cores fail. Once I dropped the offset to -25, it was rock steady stable. I might have to reseat the block as my temps hovered around 80-ish under load.

*As for the "sucker" I sold it to, he understood I wanted to keep it...for now.
Click to expand...
Using PBO and Curve Optimizer is not difficult, just different. I came from a Q6600 and a then a Ryzen 5 2600x before my 5800x. I was familiar with PBO but not curve optimizer and the way it was setup in the BIOS was different than PBO for my 2600x so at the beginning it all looked like a bunch of voodoo to me. Once I understood what I was doing it was very easy but time consuming.

It took a lot of time for me to get curve optimizer dialed in properly but I have good reason to believe I happen to have a rather bad sample of the 5800x. I dialed in the individual cores which was very time consuming but the cores range from -5 to -20. Most cores are at -10 to -15 but a couple will do -20 and one really shitty core which will do nothing more than -5 or it will take a shit eventually.

To dial in the cores I used Prime95 blend test. Running the different types of maxed out CPU tests which would stress memory and cache as well as the CPU seemed the best option. Also, since Prime95 starts an instance for each thread automatically it simplified the testing. Also, the threads stay locked to specific cores and in order so when a Prime95 thread crapped out it was easy to tell what core was affected. If it was one of the first two threads it was core 0. One of the second two threads was core 1 and so on.

My CPU will single or lightly threaded load boost to 4.8-4.9 stable with all core being around 4.6 depending on heat but will be lower in something like Prime95 SmallFFT. My CPU also will not boost +200mhz stable but max out with +150mhz. It would probably clock higher, especially during all core loads if I had better cooling. Cooling is an Enermax LIQMAX III 240 AIO and actually does well considering it only cost me $36.
 
SmokeRngs said:
Using PBO and Curve Optimizer is not difficult, just different. I came from a Q6600 and a then a Ryzen 5 2600x before my 5800x. I was familiar with PBO but not curve optimizer and the way it was setup in the BIOS was different than PBO for my 2600x so at the beginning it all looked like a bunch of voodoo to me. Once I understood what I was doing it was very easy but time consuming.

It took a lot of time for me to get curve optimizer dialed in properly but I have good reason to believe I happen to have a rather bad sample of the 5800x. I dialed in the individual cores which was very time consuming but the cores range from -5 to -20. Most cores are at -10 to -15 but a couple will do -20 and one really shitty core which will do nothing more than -5 or it will take a shit eventually.

To dial in the cores I used Prime95 blend test. Running the different types of maxed out CPU tests which would stress memory and cache as well as the CPU seemed the best option. Also, since Prime95 starts an instance for each thread automatically it simplified the testing. Also, the threads stay locked to specific cores and in order so when a Prime95 thread crapped out it was easy to tell what core was affected. If it was one of the first two threads it was core 0. One of the second two threads was core 1 and so on.

My CPU will single or lightly threaded load boost to 4.8-4.9 stable with all core being around 4.6 depending on heat but will be lower in something like Prime95 SmallFFT. My CPU also will not boost +200mhz stable but max out with +150mhz. It would probably clock higher, especially during all core loads if I had better cooling. Cooling is an Enermax LIQMAX III 240 AIO and actually does well considering it only cost me $36.
Click to expand...
Yeah I'm still learning about the Curve Optimizer. It was quite confusing at first, thus is why I overclocked the old fashion method ( Multi+Voltage). It was a pain in the arse trying to figure why it kept giving me issues for the simplest of overclocks. Helping my friend with his rig was also challenging while trying to figure out mine. Once GotNoRice simplified the PBO & Curve method, it was a breeze. I have to admit I should have kept an open mind and learn the new way of overclocking these new chips instead of fighting against it.

Now I know, well more or less lol, what I'm doing. I'm very impressed on how the 5800X reacts and responds to the proper way of overclocking it. Last night I was gaming with my bud for several hours and it was boosting to 5.0Ghz which was quite surprising and temps never passed 55c :D
 

Attachments

  • IMG_20210801_165833.jpg
    IMG_20210801_165833.jpg
    395.1 KB · Views: 0
I did what no one does. I didn't want to build a frustration box. I cross checked QVL's from ram, nvme, motherboard. If the manufacturers said its ok, I put it in my cart as long as it was a two way street from both sides. I have never, ever done that before. And I felt like such a pussy. But who wants to drop a ton of cash on a frustration box? Not this guy.

But what I got in return was a sweet fucking rig that is just as, if not more stable than any of my Intel rigs have been. I am typing this on my Z77 system which is still pretty good, and of course its overclocked and has been for years. Like a rock.

I didn't watch any videos, I just read a little and played with it. It took some time but I got my 5600X and 5900X to rip shit up. Setting your PPT, TDC, and EDC is absolutely critical to getting PBO and CO to work right. If those are off it just wont do what you want.

Glad you got it going man, because fuck frustration boxes.
 
Well I call that common sense maybe not as common these days. I try for a 5 yr hardware cycle but it doesn't always work. Hardware does crap out but 5 yrs I wont complain.
 
DaeviousMax said:
My 5900x is a golden chip sample. -12 offset on the 4 best cores and -30 offset on the 8 remaining cores. Boosts to 5.15ghz on light loads and 4.7-4.9 ghz all core in games and non-avx loads, and 4.5ghz on avx loads, and 4.3ghz in prime 95 on small ffts
Click to expand...
Yeah, I think my 5950x is a golden chip for Zen 3 as well. I have PBO all core at -20; manually set my power limits to PPT: 200W | TDC: 180A | EDC: 160A and added FMax of 250Mhz. Single Core boosts to 5.25Ghz+ and all core gaming is about 4.8~4.9Ghz. Prime95 Blend test is 4.7Ghz and small FFTs brings it down to about 4.5~4.6Ghz.

Best of all; this thing never tops 72C no matter how much I throw at it load wise... but when I increase power limits; I gain some MT performance at the cost of ST performance. I think I found the perfect balance honestly.
 
III_Slyflyer_III said:
Yeah, I think my 5950x is a golden chip for Zen 3 as well. I have PBO all core at -20; manually set my power limits to PPT: 200W | TDC: 180A | EDC: 160A and added FMax of 250Mhz. Single Core boosts to 5.25Ghz+ and all core gaming is about 4.8~4.9Ghz. Prime95 Blend test is 4.7Ghz and small FFTs brings it down to about 4.5~4.6Ghz.

Best of all; this thing never tops 72C no matter how much I throw at it load wise... but when I increase power limits; I gain some MT performance at the cost of ST performance. I think I found the perfect balance honestly.
Click to expand...
Did you test each core individually on small ffts? Cause my chip would do -30 offset on all cores with all core small ffts since all core uses lower clock frequency. The proper way to do this is turn smt off so that you can target single cores without smt taking a prime 95 worker and lowering clocks on that particular core. I basically prime small fft tested at 1 worker, then 2 workers, then 3 and so on so that I could test stability at each thread counts highest pbo frequency. You might be surprised that your 4 best cores are not stable using my method. If they are though then congrats!
 
DaeviousMax said:
Did you test each core individually on small ffts? Cause my chip would do -30 offset on all cores with all core small ffts since all core uses lower clock frequency. The proper way to do this is turn smt off so that you can target single cores without smt taking a prime 95 worker and lowering clocks on that particular core. I basically prime small fft tested at 1 worker, then 2 workers, then 3 and so on so that I could test stability at each thread counts highest pbo frequency. You might be surprised that your 4 best cores are not stable using my method. If they are though then congrats!
Click to expand...
That's why I was using the P95 blend test when testing out Curve Optimizer settings but running the full number of threads. The blend test runs through a bunch of different tests stressing different parts of the CPU/RAM and will have higher and lower clock speeds, different temps and power usage. Turning off SMT is a bad idea unless you're going to keep it off. Cores will be run harder with higher temps when SMT is on and you want to make testing the worst case scenario so you don't have to worry about stability issues and crashes when running normally.

Testing only one core at a time and without SMT will not test stability with all core or heavy core loads.
 
SmokeRngs said:
That's why I was using the P95 blend test when testing out Curve Optimizer settings but running the full number of threads. The blend test runs through a bunch of different tests stressing different parts of the CPU/RAM and will have higher and lower clock speeds, different temps and power usage. Turning off SMT is a bad idea unless you're going to keep it off. Cores will be run harder with higher temps when SMT is on and you want to make testing the worst case scenario so you don't have to worry about stability issues and crashes when running normally.

Testing only one core at a time and without SMT will not test stability with all core or heavy core loads.
Click to expand...
Yes and no, first do me a favor and re-read my post and wait until you fully understand what I was saying and the context of what I am saying in relation to the post I am responding to before attempting to nullify my advice.

I turn smt on again after doing the smt off testing and repeat the n+1 worker small fft validation. And I never said I didn’t test all core, I said that if that is all you do then you aren’t stable validated for less than all core boost frequencies which are considerably higher clocks than full all core clocks. Like I said earlier, all core frequency is 4.3-4.5 ghz which you might very well pass if you CO -30 offset all cores but when you start testing single, dual, triple, etc etc all the way to all core then you may find your best 4 cores are not stable at -30 offset due to the higher low thread count frequency ie up to 4.9-5 ghz on low thread usage vs 4.3-4.5 ghz all core.

I’m a validation expert in real life and come up with validation protocols all the time. In my line of work thoroughness is next to godliness. I highly doubt your ryzen cpu is truly stable unless you take the time to truly put the cpu through its paces in a number of different parameters.
 
DaeviousMax said:
Did you test each core individually on small ffts? Cause my chip would do -30 offset on all cores with all core small ffts since all core uses lower clock frequency. The proper way to do this is turn smt off so that you can target single cores without smt taking a prime 95 worker and lowering clocks on that particular core. I basically prime small fft tested at 1 worker, then 2 workers, then 3 and so on so that I could test stability at each thread counts highest pbo frequency. You might be surprised that your 4 best cores are not stable using my method. If they are though then congrats!
Click to expand...
I did this, but ONLY for my 4 best cores (according to Ryzen Master), after running 1,2,3 & 4 threads, I ran all threads as well, no rounding errors after several hours. There was no way I was going to do this for all 16 cores. At one point I was doing per core offsets and my 4 best were still at -20, the others at various -25 or -30 offsets. The gain to be had by doing that (in my testing) was very minimal and mainly gave me a small % boot in MT benching, nothing else or anything practical. I've had absolutely 0 issues running the -20 all core lazy route... lol.
 
III_Slyflyer_III said:
I did this, but ONLY for my 4 best cores (according to Ryzen Master), after running 1,2,3 & 4 threads, I ran all threads as well, no rounding errors after several hours. There was no way I was going to do this for all 16 cores. At one point I was doing per core offsets and my 4 best were still at -20, the others at various -25 or -30 offsets. The gain to be had by doing that (in my testing) was very minimal and mainly gave me a small % boot in MT benching, nothing else or anything practical. I've had absolutely 0 issues running the -20 all core lazy route... lol.
Click to expand...
Like I said if you did the n+1 test on your 4 best cores and it’s still stable at -20 then more power to you! You definitely have a golden sample!
 
Nebulous said:
Welp I'm happy for ya. :LOL: Anyhows, I swapped back to the 7700K and retired the 5800X. I might find a sucker to sell it off to. Maybe they would know better than I.

*Edit* Found a sucker. Just sold board, chip and waterblock. :D
Click to expand...
Hmmm.... that tone. Sounding like someone who has an agenda or at worst trying to sound out as......? :ROFLMAO::ROFLMAO::ROFLMAO:
 
You must log in or register to reply here.
Back
Top