Overclock.net banner
1 - 20 of 51 Posts

·
Registered
Joined
·
8 Posts
Discussion Starter · #1 · (Edited)
Hello!

I'd like to share with you my overclocking guide that I recently shared on a post in Reddit, so I thought it might help others if I make a separate post, so here it is!
Please keep in mind that settings will vary from motherboard to motherboard and CPUs. I'm currently using 5900X, but this guide will be suitable for the other 5000 series cpus nevertheless.
This is your PBO settings priority - what you're aiming at first:
PBO Curve > PBO Limits > PBO Boost Override > RAM OC
Intro:

AMD 5000 series work best on Auto voltage, keep that in mind. Setting a vcore offset for instance, combined with increased LLC settings will provide lower temperatures and lower highest voltage, but will be unstable, at least from my experience. You are free to experiment, it would be cool to read about a combination that works, and it's different than others. Can't say anything about static voltages in the long term for the 5000 series, whether there will be any degradation or not, so you do all at your own risk.
Strategy and first settings:
/- Load BIOS default settings, set your XMP profile only and restart Windows.
/- Open HWInfo software as soon as you enter Windows without any benchmarking and stresstesting. Open some random programs, your browser also, doesn't really matter, usual stuff you do when starting the operating system, then check HWInfo sensor tab. You will notice that 1-2 (most cases they are 2) cores' frequencies are boosting / performing best among the other cores. Of course, there will be other cores that are performing better than others, etc.This is an important step, because we will now set your curve, depending on your cores' performance, and find the maximum that you can go, before implementing PBO BO (PBO Boost Override).
Step 1:
-> Enter BIOS, set your XMP, set PBO Limit to "Motherboard" (I've personally found this option to work best, because manual option does not always apply the limit, or the board will not always pull the max, of course every board is different, but that's the situation with my board at least: Gigabyte Aorus Elite X570. If your board still apply the default PPT / EDC / TDC values when the "Motherboard" option is used, go for "Manual" mode and enter higher PPT / EDC / TDC values than your default motherboard ones. You can see the values in HWInfo sensors tab.)
-> Set PBO Override to 0 Mhz or disable it if that option is available.
-> Enter the PBO Curve and set the option "Per Core". Now, set Negative for all cores, then set -30 for your less performing cores and - 28 for your top performing cores. Save and exit the Bios.
//** We start from -30, because it's a lot less time consuming IMO than start from 0 to 30. Also, I've found that 2 digits difference (28 to 30) among the cores works best in terms of end results and overall stability. Again, every chip is different so you might get similar or even better results with a 1 digit difference between weaker and stronger cores. You can use 1 digit difference in order to squeeze the last bit of performance for your stable overclock. Otherwise, use 2 digit difference, in order to speed up the process. Feel free to experiment, but the above method is a great starting point, based on you can come up with easier decision making.
Step 2:
-> Stress test for about 20-30mins max.
//**You need to have a good cooling solution, otherwise you might run into problems with stability, thermal trottling, etc. Temperatures in the range of 80-90 are ok, but stress-testing in the range of 88-90 degrees or more for a long time in not recommended.
I recommend you test for about a day or two with regular usage and gaming also before continuing further, since stresstesting even for 24/7 is not always reliable / trust-worthy in real - life usage scenarios. This will save you time mostly, and lower the chances of confusion if something goes wrong. It will help you troubleshoot easier also.
-> If you run into crashes, boot problems, bosd, random restarts, choose 1 core from the Curve with - 30 and set it to - 28. Test again and repeat with other cores until you're stable.
//**If you come to the point when your CPU is not stable with 28/30 digits, (you will basically reach the point when all cores are set to -28 in that case), set the worse performing cores to -28 and top performing cores to -26. Repeat the technique above, until you're stable, then continue to step 3 below, when ready.
!!! You must not set PBO BO until you find the max stable curve. Also, keep in mind that the more cores you have on higher multiplier, the higher the results you will reach, so it's a matter of testing and finding the maximum your CPU can handle.
"Per Core" option is used, because "All Core" will result into instability - not all cores would be able to handle the negative offset you set, unless it's a way lower all core offset your cpu can do, compared to a per core mix multipliers, so we choose "Per Core' here in order to balance the offset among the cores, thus reaching higher scores and stable behaviour.
Step 3:
- > If you don't run into any problems, computer restarts or game crashes with Step 2, you can start increasing the PBO BO by 25mhz steps. Test, then increase by another 25mhz until you reach the limit which is usually in the range of 50-125mhz, depending on the CPU silicon quality.
At this point you will start hitting 5GHZ or a bit more max frequency on some cores (2-6 cores approximately) in HwInfo.
//** You do not need 200mhz PBO BO, because it does pretty much nothing when your curve is maxed out. You will only get instability issues and 0 performance / score increase, or even worse scores as end result.
Step 4:
This is the final step - Memory OC.
*//**You want to do this as a final step, because otherwise you might run into the confusion whether the instability is coming from the memory oc or the cpu oc. Windows Event Viewer will not show whether the crash was a WHEA error most of the time, keep that in mind. It's a matter of testing, since all memory sticks are different. I won't go into much detail here, but here's a general guide:
-> Disable XMP.
-> Set 3800mhz frequency (MCLK). Increase the voltage to 1.4V or set to Auto, since your RAM may require more if your board overvolts that segment. If the board does not, simply enter a bit higher voltage manually.
-> Set FCLK to 1900mhz or Auto (auto will provide 1900 fclk). Set the timings as the same timings as your XMP profile for the first test. This is very optimistic, but you may get away with that. If not stable, increase the timings, etc. Try with 3800 MCLK, if you're stable, you can even try 4000mhz and 2000mhz FCLK if you feel lucky. Keep in mind that 2000mhz FCLK and 4000mhz MCLK will not be reachable by all CPUs or memory modules, so don't stress about it. Performance difference is negligible. :)

Well, that's pretty much it. Full OC guide as short as possible :) Good luck, it will be interesting to read about your results and whether this guide helped you.
Cheers!
 

·
Registered
Joined
·
1,084 Posts
I think this is a pretty good guide, but I have some input.

1) My board won't boost passed stock max freq of 4.65GHz unless input something for CPU Boost Clock Override, which I set to +200MHz. I think this is typical behavior for the 5600X series. I haven't seen many 5600X boosting to 5.0GHz on this forum.

2) I think your process of manually tuning each core's offset by 2mV is pretty intense. That would take forever haha. Thorough....yes....time consuming, very!
 

·
Registered
Joined
·
3,158 Posts
Nice guide, I shared something similar on another site.

Don't forget to stress test, Y-Cruncher test 15/16 is great for finding weak cores and the adjusting curve from there.
 

·
Registered
Joined
·
91 Posts
nice quide bro !!
i try your advise and now i am oc my 5600x to 4800 i find after many test/bench/games all core -26 is stable
max vcore 1.31 and my mem patriot sit to 3933 cas 16 stable and free from WHEA
 

·
Registered
Joined
·
8 Posts
Discussion Starter · #5 · (Edited)
I think this is a pretty good guide, but I have some input.

1) My board won't boost passed stock max freq of 4.65GHz unless input something for CPU Boost Clock Override, which I set to +200MHz. I think this is typical behavior for the 5600X series. I haven't seen many 5600X boosting to 5.0GHz on this forum.

2) I think your process of manually tuning each core's offset by 2mV is pretty intense. That would take forever haha. Thorough....yes....time consuming, very!
Yes, it is time consuming indeed, however I tried to make it as time saving as possible. I went into a lot of detail with the curve, because this technique will save a lot of time also. It may sound it's something you need 1-2 months for it, but in reallity it takes few days only. Good news is that you do this once, then use it for the next several years :)
 

·
Registered
Joined
·
8 Posts
Discussion Starter · #6 ·
nice quide bro !!
i try your advise and now i am oc my 5600x to 4800 i find after many test/bench/games all core -26 is stable
max vcore 1.31 and my mem patriot sit to 3933 cas 16 stable and free from WHEA
Great! :) Now leave -26 to the strongest cores, and put -28 on the wekest ones. Test, if you run into instability issues, try backing 1 core from -28 to -26 until you reach stability. That way you will reach higher scores. -26 all cores mean that you might put -28 on weaker cores and still be stable, that's why I'm saying this.
 

·
Registered
Joined
·
2,291 Posts
snip................Don't forget to stress test, Y-Cruncher test 15/16 is great for finding weak cores and the adjusting curve from there.
This is a very important step!

Most peeps running huge negatives offset who think they are "stable" are in for a rude awakening when they run test 15/16!

Though using test 15/16 is a great way to speed up finding instabilities in the cores.

I do the opposite to what you do, i.e. I leave CO as the very last thing, but of course the method varies for what you are tuning for i.e. different methodology for higher boosting single cores than for higher boosting all core loads

For me the emphasis is on single core boosts, however we have to bear in mind each CPU acts differently!
1/ Set LLC values, for me its either 5 or 6 for CPU LLC which for my motherboard means quite alot of droop
2/ Set PBO limits
3/ Set Scaler 10x, BO to the highest value that you can see the cores still boost. i.e. if you apply a value and the core does not boost to it then drop back until your boost equates to what the CPU can do.
4/ Only now do I start optiminsing the CO usng Y-Cruncher 15/16 to quickly weed out instabilities in cores.
5/ Memory overclock, this usually I do first to get an idea of whats stable or not, but for the final stabily testing I leave it for last as it may be effected by the boosting of certain cores, i.e. TM5 can stop working, i.e. it looks like its running but its not actually doing anything, the cause is a core/thread crashing because of either boosting too high of vDDP, vDDG needs tweaking etc.

Just to go back on my earlier mention of CPUs acting differently, the current 5600x I am using does does the following, if I apply a negative CO on the best cores than for some reason all cores boost frequencies push 100-150 mhz higher than if I apply a positive offset on the good cores.

But this cause instabilities in all core work loads.

So I use the positive offset on the best cores to "tame" the weaker cores in all core workloads, so have found a balance where best cores can boost to best frequencies where all core loads boost well but stay stable.

But all in all its a good starter guide
 
  • Rep+
Reactions: Joeking78

·
Registered
Joined
·
3,158 Posts
This is a very important step!

Most peeps running huge negatives offset who think they are "stable" are in for a rude awakening when they run test 15/16!

Though using test 15/16 is a great way to speed up finding instabilities in the cores.

I do the opposite to what you do, i.e. I leave CO as the very last thing, but of course the method varies for what you are tuning for i.e. different methodology for higher boosting single cores than for higher boosting all core loads

For me its emphasis on single core boosts, however have to bear in mind each CPU acts differently!
1/ Set LLC values, for me its either 5 or 6 for CPU LLC which for my motherboard means quite alot of droop
2/ Set PBO limits
3/ Set Scaler 10x, BO to the highest value that you can see the cores still boose. i.e. if you apply a value and the core does not boost to it then drop back until your boost equates to what the CPU can do.
4/ Only now do I start optiminsing the CO usng Y-Cruncher 15/16 to quickly weed out instabilities in cores.
5/ Memory overclock, this usually I do first to get an idea of whats stable or not, but for the final stabily testing I leave it to last as it may be effected by the boosting of certain cores, i.e. TM5 can stop working, i.e. it looks like its running but its not actually doing anything, cause if a core/thread crashing etc.

Just to go back on my earlier mention of CPUsacting differently, the current 5600x I am using does does the following, if I apply a negative CO on the best cores than for some reason all the other cores all core boost frequencies push 100-150 mhz higher than if I apply a positive offset on the good cores.

But this cause instabilities in all core work loads.

So I use the positive offset of the best cores to "tame" the weaker cores in all core workloads, so have found a balance where best cores can boost to best frequencies where all core loads boost well but stay stable.

But all in all its a good starter guide
I got the tip from you and passed it on many times.

I thought I was stable at -30 until I ran Y-Cruncher for an entire day and eventually ended up at -14 on best cores and between -15 and -17 on the others.
 

·
Registered
Joined
·
2,291 Posts
I got the tip from you and passed it on many times.

I thought I was stable at -30 until I ran Y-Cruncher for an entire day and eventually ended up at -14 on best cores and between -15 and -17 on the others.
Glad its helped you

:)

Someone asked me why I placed emphasis on these tests and its exactly how you saw it to, I thought I was stable, but was runnng all Y-Crunchers tests, when it came to test 15/16 it would sometime pass sometimes crash, so I decided to use those test to speed up the troubleshooting process.

It seems to be a good test to weed out idle crashes as I noticed it would freeze/blue screen the PC when the frequency was alternating between the two tests.

So for me its the first go to test, after its a full run of Y-Cruncher for sevceral hours, then to TM5, followed by Realbench, then some games. Since using Y-Cruncher my use of prime95 has reduced as its seems to find the same instabilites but faster!

:)
 
  • Rep+
Reactions: Joeking78

·
Registered
Joined
·
15 Posts
@mongoled Just to be clear on the methodology, the idea is to change CO on a single core and then run y-cruncher only on that specific core?
I've never used y-cruncher, i would like to give it a go but i just wanna make sure i'm following a 'correct' approach first.
 

·
Registered
Joined
·
189 Posts
For me at least, the last bit of instability is exposed with single core tests.
Y-cruncher 15/16 all core are a very good way to find errors due to FLCK being unstable (aka WHEA errors).
So far the quickest way I've found to error a single core due to an unstable curve offset is with p95.
I'd be interested in feedback or results with this script: Single core Prime95 test script for Zen 3 curve offset...
Testmem5 (TM5) with a custom config to run only 2 threads also works very well. See this post for more detail on how to set that up: AMD Curve Optimizer any guides / experience
I'll give y-cruncher tests 15/16 single-core a try as well...
 

·
Registered
Joined
·
15 Posts
The 'issue' i'm still debating, and i hope someone here have had a similar impression, is how changing CO on a core seem to influence others aswell and i'm not only talking about the boost possibilities but rather the stability aswell. In many occasions, after having exclusively tuned my strong cores first and having successfully tested them for stability, i found them to suddenly not be as stable anymore while tuning other cores and i had to readjust them, which seems really weird but maybe from an engineering pov makes sense...
 

·
Registered
Joined
·
189 Posts
The 'issue' i'm still debating, and i hope someone here have had a similar impression, is how changing CO on a core seem to influence others aswell and i'm not only talking about the boost possibilities but rather the stability aswell. In many occasions, after having exclusively tuned my strong cores first and having successfully tested them for stability, i found them to suddenly not be as stable anymore while tuning other cores and i had to readjust them, which seems really weird but maybe from an engineering pov makes sense...
Were you running single, or multicore stress tests?

For multicore, then I think what happens is that all cores will run at the same voltage and frequency. In that case, the cores that you had not set a CO for would have forced a higher voltage, masking the unstable CO.
 

·
Registered
Joined
·
2,291 Posts
The 'issue' i'm still debating, and i hope someone here have had a similar impression, is how changing CO on a core seem to influence others aswell and i'm not only talking about the boost possibilities but rather the stability aswell. In many occasions, after having exclusively tuned my strong cores first and having successfully tested them for stability, i found them to suddenly not be as stable anymore while tuning other cores and i had to readjust them, which seems really weird but maybe from an engineering pov makes sense...
For me this occurs when I change the CO for the best cores. But I am sure the same effect happens with other cores but just to a lesser effect.

I am guessing the algorithm takes into account some global maximum power envelope, so any CO you adjust effects the boost/voltage curve of the other cores
 

·
Registered
Joined
·
15 Posts
@blu3dragon usually single core, btw i'm playing with your powershell script and it's already been very useful in finding instabilities i wasn't really aware of (one of my poopy cores really doesn't like too much CO, even though it will never boost that high anyway, while the poopiest of them all doesn't give a **** and -given the chance- it would probably go -50 CO and not even flinch)

@mongoled exactly, it's mostly when tweaking the good core and in my specific case, when tweaking the core between the good ones (5-7 are my good ones, tweaking 6 usually makes 7 unstable for some damn reason)

I'm also in a precarious situation in which my least preferred core happens to be 0 and windows very often arbitrarily does background stuff on that one, ignoring cppc and making it constatly boost at max frequency.
 

·
Registered
Joined
·
804 Posts
For multicore, then I think what happens is that all cores will run at the same voltage and frequency
Actually, it might not be the case at all. Let alone my observations, here's what 1usmus said in his CTR 2.0 article:
During the development of CTR, it was discovered that during the boost to all cores, each core gets its own individual portion of voltage depending on the individual silicon characteristics (FIT). Information about this nice architectural feature started to flicker at the time of the Cezanne codename processors announcement, although it existed already in the Renoir codename processors

it will be interesting to read about your results and whether this guide helped you.
I tend to agree with mongoled here, and although both approaches may be valid, setting Boost Override as base line thought more systematic to me, so you can improve both single and multi-performance, unless you're going to sacrifice former for the sake of later.
 

·
Registered
Joined
·
53 Posts
Anyone else negatively impacted by PBO2 in real world applications? With benchmarks its awesome in multi and single core however real world applications see a massive drop in frames for example when software calls for additional worker counts. Its super annoying. It's hitches in BF, WZ is a mess. It feels like the drops may be a result of the shifting boost on the various cores (speculating). When switching from an all core OC to PBO2 one title in particular literally suggest a hardware change has happened, would I like to optimize. Judging by the core count being set in the .ini file not having changed its odd not to mention visually seeing and feeling the performance hit. I can see what the cores are doing so it's a bit hard to understand for me. By far best performance is an all core overclock so I've been trying to track down realistic (safe) all core OC voltages, temps, and speed based on a 360 AIO Cooler. Anyone else experience this?
 

·
Registered
Joined
·
804 Posts
@TheBrandon
The behavior you describe reminds me what I've been fighting for some time ago on my previous system (3600X/X370) and that was caused by high DPC rate, which in itself was due to buggy mouse / usb port issue. All gone after switching to logitech g305.
Nothing since then, on new system as well. Using PBO2 pushed close to its limits (Boost Override +450, dram [email protected] tight, Limits - MB, Scalar 5x, agressive CO)
 

·
Registered
Joined
·
53 Posts
@TheBrandon
The behavior you describe reminds me what I've been fighting for some time ago on my previous system (3600X/X370) and that was caused by high DPC rate, which in itself was due to buggy mouse / usb port issue. All gone after switching to logitech g305.
Nothing since then, on new system as well. Using PBO2 pushed close to its limits (Boost Override +450, dram [email protected] tight, Limits - MB, Scalar 5x, agressive CO)
That is super interesting especially considering the heat AMD has on them right now with USB 2.0 devices. I am running all 3.0+ ports on a GPro SuperLight. I'll check with LatencyMon today. I'll have to try removing all USB drivers, I can try another port and have a dozen mice I can test but that certainly wouldn't have been something I thought of lol. I am switching boards soon and will do a rebuild of the OS. An all core OC resolves this entirely. So you simply switched mice? Didn't hit the device manager or anything?
 
1 - 20 of 51 Posts
Top