No good Ryzen 5000 series overclocking guides.

cyclone3d

[H]F Junkie
Joined
Aug 16, 2004
Messages
15,447
Looking for a guide that has any actual good information for overclocking Ryzen 5000 series is pretty much impossible.

Seeing some random people say to use PBO and set the wattage / amperage limits.

Seeing others say to set the voltage lower than stock while others say to set it higher than stock.

Then there are places where people are just saying what they have certain things set for and other places claiming what they achieved but give absolutely no settings that they used.

So is it just all conjecture and nobody has any real clue what to do so they just do random crap until something magically works?
 

cyclone3d

[H]F Junkie
Joined
Aug 16, 2004
Messages
15,447
Buy high rated ram
Use memory tuner
Enable PBO

Boom.
Hah, already done.

Simply enabling PBO on my setup kills single and all core boost speeds.

Going from 5.05Ghz for single core to 4.8 single core and 4.4 to 4.3 all core.

Makes no sense.

Manually setting the max power levels also makes no sense.

EDC set to 190 and single core says it is maxing it out. Set it to 250 and single core shows it using about 10%... What the actual what?!?!?!?!?!?!

Setting CPU voltage to offset negative and it slightly raises the CCD0 boost clocks and lowers the CCD1 boost clocks by about 400Mhz... What the crap is going on?
 

KazeoHin

[H]F Junkie
Joined
Sep 7, 2011
Messages
8,721
Hah, already done.

Simply enabling PBO on my setup kills single and all core boost speeds.

Going from 5.05Ghz for single core to 4.8 single core and 4.4 to 4.3 all core.

Makes no sense.

Manually setting the max power levels also makes no sense.

EDC set to 190 and single core says it is maxing it out. Set it to 250 and single core shows it using about 10%... What the actual what?!?!?!?!?!?!

Setting CPU voltage to offset negative and it slightly raises the CCD0 boost clocks and lowers the CCD1 boost clocks by about 400Mhz... What the crap is going on?
Possible Mainboard/BIOS related? I know that undervolting can help reduce heat which in-turn helps boost further.

Another thing with Ryzen processors is that the colder the better, so cooling plays more of a role in PBO than anything.
 

learners permit

[H]ard|Gawd
Joined
Jun 15, 2005
Messages
1,223
List your hardware specifics and you'll likely find more folks willing to help with debugging your setup and providing sound advice.
 

cyclone3d

[H]F Junkie
Joined
Aug 16, 2004
Messages
15,447
List your hardware specifics and you'll likely find more folks willing to help with debugging your setup and providing sound advice.
Pertinent specs.

ASUS Strix X570-e
Ryzen 9 5950x
G.Skill DDR4-3600 C16
Seasonic 1000w Platinum

CPU and VRMs are water-cooled via a custom loop. 420mm Thermaltake thick copper rad with push/pull 38mm thick fans.

Temps at idle are under 40c, and I have seen under 30c with certain BIOS settings that make 0 sense why the temps would be lower.

Max load temps are generally about 67c under single core load and less under all core load.
 

OFaceSIG

2[H]4U
Joined
Aug 31, 2009
Messages
3,561
I've always admired the "lets get more out of what we already have" sentiment of overclocking. But with these 5000 series Ryzens is it really worth it? Are you really going to see that big a difference?
 

cyclone3d

[H]F Junkie
Joined
Aug 16, 2004
Messages
15,447
I've always admired the "lets get more out of what we already have" sentiment of overclocking. But with these 5000 series Ryzens is it really worth it? Are you really going to see that big a difference?
With all-core boost, yes.

4.4 vs 4.7-4.8 can make a difference.

Single core boost can make a difference as well depending on how high you can get it.

The really sucky thing is that if you get all-core boost up on the automatic overclocking setting it seems like it lowers single core boost which makes absolutely no sense.

Oh yeah, some of the settings in the BIOS of this board seem to do the exact opposite of what they say... Such as the one setting that says if you turn it on, it may increase single core boost... Nope, it lowers single core boost to 4.175Ghz.
 

Domingo

Fully [H]
Joined
Jul 30, 2004
Messages
21,589
Out of curiosity, what kind of real-world performance increases are people seeing? When I first got my 5800X, it felt like people were reporting gains in super specific areas or pulling a hundred more 3DMark points, but it didn't seem like messing with. Especially since I'm gaming at 4K and my job doesn't involve video editing or 3D rendering. Are there compelling reasons to go for it beyond "why not?"
 

Whach

[H]ard|Gawd
Joined
Dec 22, 2011
Messages
1,204
Start PBO. Determine which cores are the best. Either decide to tweak individual cores with curve optimzer, or just set and forget PBO for all of them. Its a long process for granular tweaking, no doubt. I've set PBO to a higher negative value on all the identified best cores so PBO can do its thing on it, and a lower value on the less capable cores.

I'm also using the ASUS dynamic OC switcher feature so I also get the highest single core speeds on single thread workloads too.

Of course, what you get is dependent on your cpu quality.

I like this guide particularly.
 
Last edited:

learners permit

[H]ard|Gawd
Joined
Jun 15, 2005
Messages
1,223
The latest version of Ryzen master has a routine built in that generally gives a really good baseline for curve optimizer settings although it does require quite a bit of time for it to complete the test and may require a bit of fine tuning for optimum settings.
 

cyclone3d

[H]F Junkie
Joined
Aug 16, 2004
Messages
15,447
I did run Ryzen Master per core optimizer and it sets all but one of the cores at -30. The other one gets set at -17.

However, it was not giving me good single core boost or good multicore boost.

Going to start over from scratch even though I know I set everything I had changed back to default.
 

freeagentt

Limp Gawd
Joined
Dec 5, 2018
Messages
309
What I did with mine was find what the absolute max power this thing will draw, and entered the ppt tdc and edc values that I got from the testing. Then I adjusted my curve to that.
SuperPi32M each core.png
 

cyclone3d

[H]F Junkie
Joined
Aug 16, 2004
Messages
15,447
Thanks for the pic.

Default settings besides setting up DOCP and voltages for the RAM:
Cinebench R23.
Single core: 4,977Mhz (score of 1,626)
Multi core: 3,866Mhz (score of 25,094)

Stock wattage/amperage settings:
PPT - 142
TDC - 95
EDC - 140

TDC and EDC are of course what is holding back the multi-core boost.... well, PPT as well, but with the stock settings it is not maxing out like TDC and EDC are.

--------------------------------------------
PPT - 200
TDC - 120
EDC - 160

Single core: 4,890-4,915 (score of 1,604) - why did the max single core speed go down? It was not pulling any more power and temps were not higher?
Multi core: 4,215Mhz (score of 27,495)

TDC and EDC are of course still maxing out on multi core

---------------------------------------------
PPT - 220
TDC - 140
EDC - 160

Single core: 4,870-4,890 (score of 1,593)
Multi core: 4,390Mhz (score of 28,577)

---------------------------------------------
PPT - 240
TDC - 160
EDC - 200

Single core: 4,850-4,870 (score of 1,592) - fun part - this is getting slower and slower and temps are going slightly down each time.
Multi core 4,455 (score of 28,996) - all core load temps went up about 7c and voltage went up a lot. HMMMM.

---------------------------------------------
PPT - 250
TDC - 170 (Ryzen Master pegs this when running the tuning routing when it loads up all cores)
EDC - 280

After the per core tuning, rebooting, letting it enable, and then having to reboot again because it set it to "Auto Overclocking" instead of PBO, it is running the cores with these offsets:
-26,-24,-25,-26,-26,-15,-26,-26,-26,-26,-26,-26,-26,-26,-26,-26 (corrected as I was going column by column and not row by row)

Single core: 4,930-4,990 (score of 1,621) It was moving between core 2 (-26) and core 6 (-15) so it could have been a little higher if it had been smarter
Multi core: 4,525 (score of 29,472)

Finally feel like I am getting closer to getting this thing tuned. I wonder what it will do if I manually put those settings in the BIOS and then tweak from there. Guessing it will end up clocking a bit higher.
Oh yeah, I am just shy of 50c on single core and max 65c on all core loads.

I know the first time I ran the per-core tuning in Ryzen Master, all were -30 except for that one which was -17. No idea how in the world it came up with those numbers but I may just try those in the BIOS manually and see how it fares.

---------------------------------------------

Settings them all to the -30 except the one and the -15 one to -17 cause a BSOD during the single core test.... set them back and now trying lowering the EDC.

---------------------------------------------

225 EDC setting flips out and it shows as 0 on Ryzen master and the first CCD max clocks go up to 4,775 and the second CCD max clocks go down to 4,015... BIOS bug much?

Edit: I was wrong.. there is a setting called PBO Fmax Enhancer which says it may help increase single core boost.... It apparently sets EDC to infinite and doesn't help at all.

---------------------------------------------

PPT - 250
TDC - 170
EDC - 220

Single core: 4,930-5,015 (score of 1,629) - yay for core 2 and 6 again.
Multi core: ~4,575 (score of 29,785)

---------------------------------------------

PPT - 250
TDC - 170
EDC - 210

Single core: 4,935-5,025 (score of 1,632)
Multi core ~4,575 (score of 29,887)

---------------------------------------------

PPT - 250
TDC - 170
EDC - 200

Single core: 4,955-5,012 (score of ,1631) - the poor core max went up while the good core max went down.
Multi core: 4,583 (score of 29,852) - weird as the maintained boost was higher throughout the test.

---------------------------------------------

PPT - 250
TDC - 170
EDC - 190

Single core: 4,955-5,012 (score of ,1631)
Multi core: 4,583 (score of 29,644)

Looks like EDC likes a minimum of 210. Now to see what happens if I bring down TDC below what it is actually pulling.

Edit: Nope, that didn't help and neither did increasing TDC. Going to see if lowering the voltage by using a negative offset will help any.

---------------------------------------------

So setting the voltage offset doesn't do anything when using PBO.

I ended up rerunning the Per core optimizer after finding what looked like the best EDC setting and it did make the core offset settings a bit better.
-27,-27,-25,-27
-27,-16,-27,-27
-27,-27,-27,-27
-27,-27,-27,-27

Going to see what it does for the all-core instead of just the per-core tuning does.

---------------------------------------------

All core optimization set all to -19.. hrmmm.. so why would the per-core optimization set one to -16? I'll put the -16 one at -19 and see how it goes.

Edit: looks to be working fine with it set like that. So so far:

1. Set up RAM via DOCP.. had to adjust RAM and SOC voltage.
2. Enable PBO and run Ryzen Master in advanced view and the up PPT TDC and EDC so nothing maxes out for per-core PBO.
3. Do a full run of per-core optimization in Ryzen Master.
4. Save and apply the optimization settings.
5. Do runs of Cinebench R23 with Ryzen Master open on top so you can watch the clock speeds.
6. Take notes of the resulting Cinebench scores as well as clock speeds reported in Ryzen Master.
7. Lower EDC by 10 and run step 6 again.
8. Repeat steps 6 and 7 until you get to a point where performance starts dropping.
9. Set EDC to the highest Cinebench score even if clocks are not quite the highest they were showing.
10. Run Ryzen Master all-core optimization and take not of the resulting offsets and then save them to the per-core optimization profile IF they are better than the original per-core optimization offsets were originally detected as.

That should get you pretty close to max performance CPU-wise if you want auto-overclocking.

PPT and TDC really could be set to no limit as, from testing, limiting them below what they pull with EDC limited doesn't do anything but lower performance.

---------------------------------------------
 
Last edited:

freeagentt

Limp Gawd
Joined
Dec 5, 2018
Messages
309
Try 220/140/180 and see what happens

Edit:

With my current PBO/CO settings I score 23,6 in R23, with a static oc at 4750 I score about 24,4
 
Last edited:

TheSlySyl

2[H]4U
Joined
May 30, 2018
Messages
2,324
The latest version of Ryzen master has a routine built in that generally gives a really good baseline for curve optimizer settings although it does require quite a bit of time for it to complete the test and may require a bit of fine tuning for optimum settings.
Doing this, for me, created a setting that wouldn't even allow my computer to boot until I did a hard reset of the BIOS and changed every single setting. Haven't touched Ryzen Master since.

Due to RAM issues, i'm only running my RAM at 3000mhz at the moment, and I've set my PPT to 200 max because the difference between 200 and 220 were about 10c and only 100mhz or so at load.
I'm currently at 28668 with a 100mhz overclock at 200/140/175. This is with some background processes running because i'm too lazy to shut them down for benchmark ### and I really only care about stability when these background processes are gonna be open 100% of the time anyway. I have all my cores at a -15mv offset because they're stable that way and it didn't require more tinkering.

I'm running an Arctic Cooling Liquid Freezer II 420 and my max temp has been all over the damn place because I don't have AC and my ambient temp has been everywhere from 20c to 37c in the past few weeks. It's hell out here. Load temp has generally stayed below 80c though.
It's not as fine tuned as it could be, but i'm hitting over 5Ghz on a single core and my all core is around 4.3 and that's more than good enough for me considering I require stability over a few points on a benchmark. I could probably get another 50-100mhz and another 1000-2000 R23 points out of it if I tried, but... I haven't had a day that I cared enough to spend adjusting fine numbers. Especially because part of that is fine tuning my RAM back to 3200 or more and that's, frankly, a PITA that i don't care about right now.
 

cyclone3d

[H]F Junkie
Joined
Aug 16, 2004
Messages
15,447
Finished testing for now. Next is RAM timing optimization. See my above post for results.
 

learners permit

[H]ard|Gawd
Joined
Jun 15, 2005
Messages
1,223
I've always admired the "lets get more out of what we already have" sentiment of overclocking. But with these 5000 series Ryzens is it really worth it? Are you really going to see that big a difference?
Around 3 to 7% by rough estimation across a wide variety of CPUs I've tested(~30).
 

xer0

Limp Gawd
Joined
Apr 7, 2005
Messages
145
In my experience with my 5600x and 570 chipset MB, using PBO and undervolting (to maintain higher boost clocks on fewer cores) provides better increases in single core-centric tasks like most gaming and most day-to-day applications (internet browsing, office programs, etc). Overclocking/overvolting all the cores provides increases in multi-core benchmarks like cinebench at the con of high power usage and heat, but damn the multi-core benchmarks looks better. Since almost never require a lot of cores running concurrently, I go with the PBO/undervolting.
 

GotNoRice

[H]F Junkie
Joined
Jul 11, 2001
Messages
11,358
Yeah PBO is almost always the way to go unless you are doing some specific workload that actually loads down all the cores all the time. The biggest challenge with PBO is testing stability. When you under-volt, stability issues are more likely to occur during light-load rather than during full load, as it pumps extra voltage into the cores during full load. So traditional stability tests that load down the whole CPU, such as Prime95, are completely useless. When I was dialing in my PBO settings, most of my stability issues occurred in the middle of the night while I was sleeping - I'd wake up to find out that my computer had rebooted. Thankfully you can see right in the event log which core caused the WHEA error, so you can then make a slight adjustment to the offset on that core.
 

Starrbuck

2[H]4U
Joined
Jun 12, 2005
Messages
2,869
You are exactly right. Project Hydra will run for ~12 hours and figure all that out, supposedly. I haven't had any crashes, under high or low loads, since using it.
 

cyclone3d

[H]F Junkie
Joined
Aug 16, 2004
Messages
15,447
I was having reboots when computer was just sitting there. Ended up upping the RAM voltage by one tick, whatever that was.

Need to go back and look at what I had my PBO settings at because I messed with a couple of the cores, thinking it was one of them that was causing the reboots.
 

Vengance_01

Supreme [H]ardness
Joined
Dec 23, 2001
Messages
6,811
I like using OCCT to test individual cores. It tells which one errors out. From there you can use a global offset or just lower that 1 core down. Test and repeat till all cores are optimized.
 
Top