• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

AMD Curve Optimizer any guides / experience

Anyone here has any experience with overclocking a 5950x and testing stability on Linux? I tried PBO+CO settings and tested stability on a Windows install, and everything seemed fine, till I went to Linux and started seeing crashes immediately. So, all my CO values ended up being thrown out of the window on Linux. FYI, I have a Gigabyte X570 AORUS PRO WIFI.

Hi @RandomJerk, I've only ever had linux for the last decade in all my PCs, Workstations, servers, and IoT SoCs.

I haven't got a 5950x but I do have a 5800x watercooled all core overclock in my main workstation server; it is coupled with a Powercolor Liquid Devil Ultimate 6900xt.

For CPU stability testing I use prime95 called "mprime" in linux: https://www.mersenne.org/download/
I typically run it for about 2 hours on all threads on the torture setting option 2 small FFTs maximum power test.
Then I run the blend test for 12 hours.
I'm stable on all core OC at 4.65 under load. I do all core OC as my workstation does 24/7 multithreaded compute so my results are better compared to PBO2.

For GPU stability testing I use Furmark or the Piano Test from Geeks3D. https://www.geeks3d.com/20140304/gp...w-fp64-opengl-4-test-and-online-gpu-database/
It is an older version but does the job natively. If you want the newer version of Furmark then you can run it in Wine I suppose, but I don't bother and the 0.7.0 version does the job.
I modify the various .sh scripts to fit my screen resolution. Just open it up in Geany or the Gnome Text Editor and change the config. For stability testing it is the Windowed scripts that you want to modify since it will run indefinitely until you terminate it.
 
Just because i've got the info scattered in half a dozen threads now, but its relevant to this topic:

Temps are key to the clocks when using PBO.

I'm at a mere 98W PPT on my 5800x, but because its under/at 60C i'm seeing 4.85GHz all core while gaming.
Performance is absurd.
 
Just because i've got the info scattered in half a dozen threads now, but its relevant to this topic:

Temps are key to the clocks when using PBO.

I'm at a mere 98W PPT on my 5800x, but because its under/at 60C i'm seeing 4.85GHz all core while gaming.
Performance is absurd.
98PPT damn, hope my new cooler improves it as well, but not as absurd as yours though, now im curious how your system handles the likes of heavy avx2 RPCS3 (and avx512 instructions, without needing avx512 itself)
 
98PPT damn, hope my new cooler improves it as well, but not as absurd as yours though, now im curious how your system handles the likes of heavy avx2 RPCS3 (and avx512 instructions, without needing avx512 itself)
I actually am not sure if i'm on 108 or 98 right now, as i was testing a few different options tbh

AVX stuff runs lower clocks than gaming, cause its more power hungry i assume

Used OCCT and its AVX2 setting - 4.65Ghz 108W, 62C

1636696969195.png
 
I actually am not sure if i'm on 108 or 98 right now, as i was testing a few different options tbh

AVX stuff runs lower clocks than gaming, cause its more power hungry i assume

Used OCCT and its AVX2 setting - 4.65Ghz 108W, 62C

View attachment 224887
Yeah, AVX Workloads tends to downclock the frequency, even more so for AVX2 running AVX512 instructions in which RCPS3 is doing
one of the areas where the 11th gen intel can overtake zen3 because of AVX512
1636704530053.png

1636704695961.png
 
Yeah, AVX Workloads tends to downclock the frequency, even more so for AVX2 running AVX512 instructions in which RCPS3 is doing
one of the areas where the 11th gen intel can overtake zen3 because of AVX512
View attachment 224892
View attachment 224893
You can enable AVX-512 on Alder lake in Linux. But the catch is you will need to disable the E-cores...
Here are the benchmarks and how to do it... https://www.phoronix.com/scan.php?page=article&item=alder-lake-avx512&num=1

There is a story within a story here... Intel always had AVX-512 earmarked for Alder-lake but the 10nm design and the sheer density of stuff all working within the 10nm design lead to too much power usage and not enough heat dissipation. It is still a problem and it will be interesting in future architecture on how they will get around that problem and solve it.

The easy path is to do less in the volumetric area they are playing with to have less heat, which means less featured CPUs; however with AMD, Nvidia and Apple going even smaller on TSMC's and Samsungs node-tech; the only way Intel can compete is to do more in the volumetric area they are playing with, which for now is 10nm. They are basically trailing the other two so have to compensate by stuffing more in, in the volumetric area that they have. (Volumetric as architectures are now 3D and no longer planar 2D).
 
Last edited:
Hey all, just found this thread as I was looking for help with my 5950x PBO OC settings. Here is where I'm at...

I have tried to follow guidance on this thread and have researched what each setting does separately to try better understand what I am doing. Unfortunately I am a little stuck because certain settings seem stable until they are not, for no apparent reason. Let me explain, first off here is the hardware:

5950x
MEG X570S ACE (latest bios)
3600 CL16 (Samsung Bdie) running XMP
beQuiet Titanium 1000W PSU (connected to an Eaton UPS so power delivery is as clean and stable as it can be)
Cooled by Arctic 360 AIO with Corsair ML Pro fans in a Meshify 2 with ML Pro Fans all around.
Win 11 Pro

Right then, so the bios settings I have are all stock apart from:

XMP on
PBO Advanced
PTT 220 TDC 135 EDC 180 (tried lower EDC but performance was worse)
Scalar Auto
+50mhz boost clock
temp target 90

Curve started out at -10 all core, which then passed CPU Z bench, 10 minutes of Cinebench R23 and 30 mins of Linpack OCCT + OCCT bench
Temps were solid, I was happy with the performance so I moved on to -12 CO, all good, -15 CO all good, -18 CO still fine, this time I even let it stress for longer and again no WHEA errors and temps were solid (max 73C in an ambient 24.5C).

Best scores at this point were:
C23 - 29000
CPUZ - 688 single 13320 multi

Then I decided to try out -20 CO and the machine did boot into windows and ran CPU Z bench, it also ran OCCT linpack however it failed on OCCT bench (pc just shut down). I thought, ok I will just back down to 18. Yeah well I did that and Windows wont boot (BSOD straight after bios loading screen). Went back, tried -15 and this time it boot but once on desktop it crashed again with WHEA error. Now I am at -10 again and it seems to work.

I need help before I go mad trying to figure out how something is stable for so long and then once it crashes, I have to go back several steps just to get a working system?

Also how do I figure out my best cores so I can start to optimise CO on a per core basis?

Thanks in advance, this is a nightmare but I hope to see myself through it.
 
Ryzen Master shows your best cores in each CCD. No idea how accurate it is though.
 
Hey all, just found this thread as I was looking for help with my 5950x PBO OC settings. Here is where I'm at...

I have tried to follow guidance on this thread and have researched what each setting does separately to try better understand what I am doing. Unfortunately I am a little stuck because certain settings seem stable until they are not, for no apparent reason. Let me explain, first off here is the hardware:

5950x
MEG X570S ACE (latest bios)
3600 CL16 (Samsung Bdie) running XMP
beQuiet Titanium 1000W PSU (connected to an Eaton UPS so power delivery is as clean and stable as it can be)
Cooled by Arctic 360 AIO with Corsair ML Pro fans in a Meshify 2 with ML Pro Fans all around.
Win 11 Pro

Right then, so the bios settings I have are all stock apart from:

XMP on
PBO Advanced
PTT 220 TDC 135 EDC 180 (tried lower EDC but performance was worse)
Scalar Auto
+50mhz boost clock
temp target 90

Curve started out at -10 all core, which then passed CPU Z bench, 10 minutes of Cinebench R23 and 30 mins of Linpack OCCT + OCCT bench
Temps were solid, I was happy with the performance so I moved on to -12 CO, all good, -15 CO all good, -18 CO still fine, this time I even let it stress for longer and again no WHEA errors and temps were solid (max 73C in an ambient 24.5C).

Best scores at this point were:
C23 - 29000
CPUZ - 688 single 13320 multi

Then I decided to try out -20 CO and the machine did boot into windows and ran CPU Z bench, it also ran OCCT linpack however it failed on OCCT bench (pc just shut down). I thought, ok I will just back down to 18. Yeah well I did that and Windows wont boot (BSOD straight after bios loading screen). Went back, tried -15 and this time it boot but once on desktop it crashed again with WHEA error. Now I am at -10 again and it seems to work.

I need help before I go mad trying to figure out how something is stable for so long and then once it crashes, I have to go back several steps just to get a working system?

Also how do I figure out my best cores so I can start to optimise CO on a per core basis?

Thanks in advance, this is a nightmare but I hope to see myself through it.
Welcome - when you got a all core setting stable, then you can progress by changeing one core at the time.
 
Ryzen Master shows your best cores in each CCD. No idea how accurate it is though.
Ok I ll give those a go.


Welcome - when you got a all core setting stable, then you can progress by changeing one core at the time.

Thank you for all your useful posts earlier in the thread, they definitely helped determine a starting point for me!

And yes that is the aim, however as per my opening post I had (or at least assumed) an all core stable CO which turned out to not be so stable.

On another note I just discovered CoreCycler so have been running that for an hour which has really helped determine which of my cores couldn't deal with the offset, that explains the odd behaviour before. I will see how far I can get with CO, but other than that is there any other settings I should tweak within BIOS to help with a more stable OC? Voltage, LLC, SOC ect ?

Thanks!
 
Definitely use the single threaded tester (i forgot its name, ugh) - it caught something i'd missed where my top performing core needed a tad higher setting than the rest

Go for all core that boots and is roughly stable, then run that tester through allowing it to stop on cores that error. Then you can raise those back up, and go from there.
Early guess: higher perormance cores will want more, slowest cores want less.
 
So after 6 hours of testing on CoreCycler I thought I actually had a stable system! Sadly, I was wrong once again... 2 hours of Battlefield 2042 and blue screen with WHEA errors... ffs

So now I have dialed things down even more and at this point, unless I am clearly missing a setting here (for stability), I give up and will just get a 12900k later in the year when DDR5 is readily available.

Pretty sure I have a dud for a CPU when it comes to CO:

Core 5, 6 and 7 seem to only do -5, the rest can do -15 (although were previously stable at -20 but now it wont even boot with that).

Shame.
 
this is only a 5800x it might be of some use for you.
ppt:120w,tdc:85A,edc:108A max boost +50mhz
Cores set to C0-6,
C1-6, C2-16 ,C3-16, C4-16, C5-16 ,C6-5 ,C7-5. GSS-Disabled
not a engineer but it seems that core 2,3,4,5 are "grouped" together and i was abled to more negative offset. still working on this idea. Cores 0 and 1 are by there selves as are core 6and 7.
A-XMP is enabled for 3600mhz memory
in testing i do not get whea errors but event id:41which cause bsod or it simple restarts if this happens 2 times in a row windows will go into a repair state and ask if i want boot back into windows or use a usb device to repair. boot back into windows and use Tweaking.com-windows repair and follow the steps for the program this works like a charm and i am able to put the last working curve offsets as stated above. core 6and 7 will not tolerate no more than -5 this known. the other cores i am still working on as i am in no hurry. my 5900x dose not have these problems so far still working on this R9 cpu. i think that most of my problems revolve setting up CO absolutely correct. now i can get both CO and pbo to work together.
i have looked thought the forums here and have not the core clusters like i have.
power plan set to best performance and idle voltage does not drop.
 
Last edited:
So I raised EDC to 200 and TDC to 145, I then re-did the CO with the strongest 3 cores getting -6 and everything else -16. This is the best scores I can get after an hour or so of messing around with stress tests, at XMP profile ram (3600 CL16)

results.png


Going to run a few more hours of various stress tests to make sure its 100% rock solid, before I start tuning the memory, I am going to go for a 3800 CL16 Fast preset using DRAM calculator and see how I get on. Should help push me over the 30k in cinebench and if it's stable, that is all the performance I need for now. I'd like a bit better single-core but it is what it is.
 
for me gaming is the acid test. CC dose not equal stability or any other stress test
 
for me gaming is the acid test. CC dose not equal stability or any other stress test

Yeah I agree, once stable in benches and stress programs, I will run various games for a few hours as I would normally and confirm stability.
 
For me it’s test mem 5 lol..

it runs those cores so fast, I usually get a thread exception error. I am just running an all core clock right now at 4500. It’s pretty boring.
 
I registered here specifically in order to present my problem with my processor and I'm not good with OC I just try for the first time to OC my CPU

First of all I just need the best temps and best low voltage to my CPU keeping the performance at the best as possible I don't need crazy numbers.

The first Issue when I set PBO to manual and add PPT and TDC and EDC and enable Curve and making It -16 and enable XMP My PC It work but Restart randomly.

The Second Issue i found the bench in CPU Z give me the same result with XMP or without XMP so by default my ram run at 2666 and with XMP 3600 no change In CPU Z score but I can see the bench give me little high score with 2666 !!!

Also my PC Restart randomly more with XMP Enabled. It restarts in both but more with XMP.

I have latest Bios Update already Installed.

Any advise ?
 
I registered here specifically in order to present my problem with my processor and I'm not good with OC I just try for the first time to OC my CPU

First of all I just need the best temps and best low voltage to my CPU keeping the performance at the best as possible I don't need crazy numbers.

The first Issue when I set PBO to manual and add PPT and TDC and EDC and enable Curve and making It -16 and enable XMP My PC It work but Restart randomly.

The Second Issue i found the bench in CPU Z give me the same result with XMP or without XMP so by default my ram run at 2666 and with XMP 3600 no change In CPU Z score but I can see the bench give me little high score with 2666 !!!

Also my PC Restart randomly more with XMP Enabled. It restarts in both but more with XMP.

I have latest Bios Update already Installed.

Any advise ?

Maybe -16 was to much in some cores of your processor. You need to see in event viewer if you got whea error, after that open the whea error and see what apic ID it was, like APIC ID 0 OR 7 OR 3. See it and told us if you have any whea error. It's not normal with xmp restarts so I assume it was to aggressive co in some core
 
Maybe -16 was to much in some cores of your processor. You need to see in event viewer if you got whea error, after that open the whea error and see what apic ID it was, like APIC ID 0 OR 7 OR 3. See it and told us if you have any whea error. It's not normal with xmp restarts so I assume it was to aggressive co in some core

Thank you for your reply

I attached the pic from Event Viewer

I will try to lower the Curve and will try to make It per core
 

Attachments

  • kp.png
    kp.png
    23.5 KB · Views: 152
  • whea.png
    whea.png
    37.5 KB · Views: 151
Last edited:
Thank you for your reply

I attached the pic from Event Viewer

I will try to lower the Curve and will try to make It per core
Its not that you need to find. You need to find something like my attach pic to see what core fail.
Pic it's not mine its from Google but it's what you need to find in event viewer.
Try per core its the best way but it consumes time. You can run now with your settings corecycler and see what core cause problems.
 

Attachments

  • 90a4ff47-3c8d-4256-9ee0-c1120c6e81a2.png
    90a4ff47-3c8d-4256-9ee0-c1120c6e81a2.png
    176.5 KB · Views: 156
Its not that you need to find. You need to find something like my attach pic to see what core fail.
Pic it's not mine its from Google but it's what you need to find in event viewer.
Try per core its the best way but it consumes time. You can run now with your settings corecycler and see what core cause problems.

I attached the pic from my PC like what you said, I hope this can make us fix the issue
 

Attachments

  • wweha.png
    wweha.png
    45.5 KB · Views: 131
I attached the pic from my PC like what you said, I hope this can make us fix the issue
But this is whea 46 you need to find some whea 19 error, related to apic ID. Or your problem that's not a curve optimizer but other thing
 
I only use Radeon software to OC my CPU.
Screenshot 2022-02-04 144211.png
 
But this is whea 46 you need to find some whea 19 error, related to apic ID. Or your problem that's not a curve optimizer but other thing

You are right the Issue from the fast cores and slow one each one need to be optimized alone.

I found the faster two cores from Ryzen master now The problem has been solved.

Thank you
 
Back
Top