AMD Curve Optimizer any guides / experience

[Nordic]

Probably best to use PBO but manually set default power limits (142/95/140), should have the same effect as PBO off. But leave boost override at +200MHz. You'll still be able to push ST boost clocks to their max with default power limits.

Then if you want higher MT perf you can uncap the power limits later when you know that ST is for sure stable and all you have to test is all-core (which is trivially easy).

Sometimes high PBO power limits can tank ST performance. Still not sure why this is.

Are you still running only 1 minute iterations? It's not really long enough of a window to do much.

I haven't even been increasing the boost override because I have not even been hitting the limit.

I only ran 1 minute tests to get rough numbers. I have been doing 20 minute runs since to really fine tune the numbers.

 

[tabascosauz]

I haven't even been increasing the boost override because I have not even been hitting the limit.

I only ran 1 minute tests to get rough numbers. I have been doing 20 minute runs since to really fine tune the numbers.

The config file has estimates for the time to get through a full cycle of different FFT sizes:

 

[Nordic]

The config file has estimates for the time to get through a full cycle of different FFT sizes:

View attachment 212069

I am running 20 minutes because ((20 minutes * 16 cores) / 60 minutes)) *1.1 = about a 6 hour run time.

 

[Nordic]

I haven't tried OCCT. I am not even running an offset. It is only boosting upto about 4900mhz.When I get back home I am going to turn everything to stock and see if I still have errors. Some cores can boost upto 5050mhz at -10 and lower in curve optimizer.

I set my bios back to their defaults. Everything is at default. I then set up core cycler to run over night at 20 minutes a core. Core Cycler is still finding errors are Core 0 and Core 1. These are the same cores that needed positive offsets earlier.

It seems I either have some A) Inherent instability somewhere or B) Core Cycler is a faulty testing mechanism.

When I get a chance I am going to try OCCT and see if it finds errors too.

 

[freeagent]

Ok.. so.. this whole time that I have been saying I run 200 130 130 was a mistake. I had Asus performance enhancement enabled, and that brought tdc to 140 and edc to 180. Looking at my notes these are the same ppt tdc edc requirements to run the shit out of my 5600x. Soo.. I ran pi 32m a few times at 200 140 180 -30 +200 and it worked just fine, hit top boost better than I had it before, at least for single core stuff.. Not sure about multicore yet but it should be good I think..

 

[outpt]

200,140,180 on a 5600x??????????????!

 

[freeagent]

200,140,180 on a 5600x??????????????!

 

[jesdals]

Did an update to my current curve settings at 275/125/115

Negative 30 is apparently lowest value on my Auros Master board. Negative 24 is minimum on core 0 - less than that and there is issues with stability in longer periods of idle desktop - freezes

you can see my settings here https://www.techpowerup.com/forums/threads/aorus-x570-master.257392/page-20

 

[bubbleawsome]

Did an update to my current curve settings at 275/125/115
View attachment 213603
Negative 30 is apparently lowest value on my Auros Master board. Negative 24 is minimum on core 0 - less than that and there is issues with stability in longer periods of idle desktop - freezes

you can see my settings here https://www.techpowerup.com/forums/threads/aorus-x570-master.257392/page-20

That is some fantastic tuning or an awesome chip. Are those clocks seen during all-core loads, and is clock stretching not an issue on Zen3?

 

[jesdals]

That is some fantastic tuning or an awesome chip. Are those clocks seen during all-core loads, and is clock stretching not an issue on Zen3?

During all core benchmark like CPU-z you will se all core speeds up to 4600MHz and single up to 5150Mhz

 

[jesdals]

New AMD chipset driver may take the top of the boost clocks - went from 5250MHz on best core to 5189MHz and the rest locked at 5015MHz

 

[jesdals]

New AMD chipset driver may take the top of the boost clocks - went from 5250MHz on best core to 5189MHz and the rest locked at 5015MHz

After a coupple of days I am seeing the same boost behavior - do not know if there is some training in the newest chipset driver?
Have seen up to 5250MHz again on core 0

 

[jesdals]

Any one else using both AMD CPU and GPU haveing issues with the latest driver for GPU and CPU. I am getting hard resets - especially in idle situations

 

[Makaveli]

There seems to be a new chipset drivers that enables support for windows 11 have you tried it?

 

[droopyRO]

Any one else using both AMD CPU and GPU haveing issues with the latest driver for GPU and CPU. I am getting hard resets - especially in idle situations

I do. I even RMA'ed my 5600X yesterday. I was on my old 2700X for two days, i decided that the CPU was broken somehow. But then last night i walked away for about 10 minutes from my PC. It was at idle, with Firefox opened with a few tabs. I returned to see it just restarted, WHEA error in EventViewer just like my 5600X did. Bios is at stock, no settings were made, especialy no curve optimization since there is none for 2700X. I now suspect the AMD drivers or the motherboard. Only two things that are making sense.

@Makaveli i update two days ago. The WHEA error i got was with them installed.

 

[jesdals]

I do. I even RMA'ed my 5600X yesterday. I was on my old 2700X for two days, i decided that the CPU was broken somehow. But then last night i walked away for about 10 minutes from my PC. It was at idle, with Firefox opened with a few tabs. I returned to see it just restarted, WHEA error in EventViewer just like my 5600X did. Bios is at stock, no settings were made, especialy no curve optimization since there is none for 2700X. I now suspect the AMD drivers or the motherboard. Only two things that are making sense.

@Makaveli i update two days ago. The WHEA error i got was with them installed.

Well we may have to reinstall windows - hope for a better fix to the bug

There seems to be a new chipset drivers that enables support for windows 11 have you tried it?

Still on windows 10

 

[Makaveli]

Well we may have to reinstall windows - hope for a better fix to the bug


Still on windows 10

Its a windows 10 driver just has support for Win 11.

i'm still on the previous version

New version is

 

[jesdals]

Its a windows 10 driver just has support for Win 11.

Yes I just looked it up again and i does seem to support both - I do have a suspicion that it is a combination of gpu and chipset driver

 

[DemonicRyzen666]

I'm currently using a bus speed bump to get a little more out this 5600x
however most programs says 100.5-100.9 when I sent 101 in bios :/
I usually have spread spectrum enable just because I have a digital Tv tuner in this and with out it on a lose channels.

 

[Makaveli]

I would be careful with pushing that bus speed

And are you using PBO and CO?

 

[jesdals]

I'm currently using a bus speed bump to get a little more out this 5600x
however most programs says 100.5-100.9 when I sent 101 in bios :/
I usually have spread spectrum enable just because I have a digital Tv tuner in this and with out it on a lose channels.

I have had some performance in 100.01 to 100.1 settings but different bios versions is more or less tolerant - stability issues do often relate to these settings

 

[Deleted member 212040]

All I do is set a -15 all core and bump up my PBO limits a little to 165 PPT, 115 TDC and 150 EDC. 5900X does not scale past 170W. Went from 4.1 all core to ~4.5-4.6 GHz. Single core now sitting nicely at 5.05-5.1 GHz.

 

[Octopuss]

So you completely disregard the differences between cores and just blindly punch in some numbers. That's not how overclocking is done.

 

[Makaveli]

So you completely disregard the differences between cores and just blindly punch in some numbers. That's not how overclocking is done.

I would say doing all core -15 is a good starting point. However you will get the most benefit with per core tuning. This is just abit more time consuming but worth it.

 

[outpt]

So you completely disregard the differences between cores and just blindly punch in some numbers. That's not how overclocking is done.

Not bad for being blind. What happens with one or two eyes open?