Ryzen Master - PBO Curve Optimizer Control - Configure in Windows Now!

III_Slyflyer_III

[H]ard|Gawd
Joined
Sep 17, 2019
Messages
1,244
All i have to say is.... about freaking time! It was a huge PITA to tune my setup per core, stress testing, then going back to the bios to adjust and then start the process all over again.

It might be a little late for me I suppose as I have everything tuned, but if I had to do this all over again, having this in windows is super helpful. Of course, this is assuming you are not crashing you PC while stressing and just getting rounding errors.

I'm not sure if this works on all AGESA versions or not; I'd have to check. But at this time, I am on AEGSA 1.2.0.5 and it works great. I'd move up to 1.2.0.6c, but 1.2.0.5 fixed all my issues and the voltage "bug" is easy to overcome by setting EDC in windows over 140 instead of the Bios. I also do not get the stuttering people report when using fTPM, so quite frankly, I'm not sure I'll ever upgrade the Bios again unless they improve memory overclocking and stability for over 3800Mhz.

1650462188441.png
 
Correct. It's an offset, so lower values = higher clocks at lower voltages. So using negative numbers, you are under-volting and essentially allowing the CPU to boost higher over the range of allowable voltages (up to 1.5V). It is by far the best way to overclock Ryzen 5000 CPUs for ST and MT combined performance.

I run my best 4 cores on CCD0 at -20, and I run my best 2 cores on CCD1 at -20... all other cores on all CCDs on -30. The idea being that your best cores that run the single/low core loads need more voltage to remain stable as they run higher clocks (5.1Ghz+). However; when running MT loads, you can get away with less voltage on the other cores, allowing for lower temps and better boosting overall as clocks remain lower (4.9Ghz or lower). Stability testing this on Prime95 to ensure my single core loads were stable and all MT loads were stable was time consuming at best... lol. But fun to know I have things tweaked to the max.

My 5950X is a pretty damn good overlocker; I think most people top out around -20 or -25 for offsets, so I may have won the silicon lottery for once... lol.
 
I know how CO works.
All your cores pass CoryCycler?
I used Prime 95, single thread to test the best cores at -20 one by one. Once i cycled all 6 of the -20 cores, I increased the thread count in P95 by 1 each time and kept testing till I hit all threads (took forever to say the least). The -30's were only tested at MT loads and was a "gamble" of sorts due to the time required. However; in "theory", they should never be called upon for single thread loads. And to date (close to a year now), I have not had a single crash or reboot, so I'm pretty confident I am solid. I use my PC for a lot of gaming, but also for work, VMs, etc, so it gets some pretty rounded loads.
 
Does ryzen master read your current CO settings on install without modifying them? Or does it reset everything? I haven't used ryzen master in a long time
 
Does ryzen master read your current CO settings on install without modifying them? Or does it reset everything? I haven't used ryzen master in a long time
It reads whatever I have in the Bios. If I do not open it; the bios remains the default until I open Ryzen Master, at which point Ryzen Master takes priority of OC values. Ryzen Master (AFAIK) does not "write" to the bios beyond what's loaded within windows. It's how I work around the Vcore bug in AGESA 1.2.0.5. I have EDC set for 140A in the bios, but I set it to 160A in windows and it will use that value until I reboot with full range of voltage (vs. the 1.425V clamp in the new AGESA versions). Why that trick works is beyond me... lol.

I guess that type of functionality is not bad in terms of not allowing the user to input values that cause the PC not to boot up next time. But if the user wants their overclocked settings at boot, then it still has to be done in the bios (which is how I set mine up anyway).

So I guess one could consider Ryzen Master a type of "software" overclocking, allowing you to test and configure till your heart is content, but when you find something solid and stable, it's up to you to put that in the bios to make it permenent.
 
Did you ever run that utility that tells you what kind of chip you have, poor to average to exceptional or something like that? I don't remember what it was called, but mine showed as average. -30 is boss on your chip. I messed with the curve optimizer in BIOS a while ago and I think the lowest I got any of the best "working" cores was -10 or maybe even -50. I would get BSODs on boot at anything lower. My estimation at the time was the core was already so efficient, you couldn't wring any more out of it. I think I did that and then set the "bad" chiplet to -15 across the board. Recently though, I just put it all back to stock. I'm working on making the system quiet now and didn't think pulling 220 watts on the cpu instead of 120 watts was worth it. I may mess around with it again once I get all my new fans installed and re-tune it.

Also, you speak of overclocking RAM over 3600. I've had multiple 3900x chips, a 3950x, and a 5950x in this same Asus Hero VIII. I don't remember if I even bothered trying on the 5950x, but I could never get the machine to even boot at 3800. I got stability with 3733 on the 3900x chips, but I did it with about 1.42v on Samsung B Dies that were rated at DDR4000 at 1.35v CAS 19. The machine will not even boot at XMP settings with that RAM. I just conceded and went down to 3600 Hynix chips at 1.35 and CAS 16. I suppose I could have gone for 1.5v on the RAM, but decided it wasn't worth messing with if it was going to take that much. It used to bother me seeing so many other people running 3800 with normal voltage and somewhat normal SOC voltage. It's just never been in the cards for me.
 
How accurate is curve optimizer when you have it do the auto optimization? I did it on my 5950X and after 2 hours of it doing it’s thing it told me -30 on every single core. I set that in the BIOS and it runs without issue.

EDIT: Ok, curve optimizer is nutty. I have 8 cores that all go above 5Ghz. Of those 8, one tops out at 5.1GHz and two at 5.2+GHz. In Cinebench R20 my CPU pretty much sits at 68°C the whole time. Before CO I had two cores that topped out at 5050Mhz and my temps were always in the mid to upper 70°C range.
 
Last edited:
For my CPU it's hard to say whether the auto-optimization is good or not. I've had indications that my 5800x is going bad but haven't been able to generate repeatable bad results to do an RMA. That said, I ran the auto-optimization and for most cores it was very close to what I had manually tested when my CPU was fully stable (I don't run CO or anything anymore other than the stock boost algorithm.) However, there were a couple of cores a bit off. One core from the auto-optimization indicated my manual setting was too far and needed to be backed off a bit. Ryzen Master indicated another core could have been pushed much further into the negative offset than what I had manually tested.

Just for the hell of it I took the settings and set them in the BIOS. It was less than 12 hours before I shut CO back off. The CPU was running cooler while heavy load boosting quite a bit higher but stability was gone. Take that with a grain of salt because as I said, I think the CPU is bad and needs to be RMA'd as soon as I can come up with repeatable bad results. Also keep in mind that my 5800x is probably one of the biggest losers in the silicon lottery. It tends to run hot and doesn't boost worth a shit even on single core/thread loads. Any attempt to allow it to boost above the stock boost frequency is a recipe for instability and has been since I got the CPU over a year ago.

To answer your question a well as I can. I suspect the settings Ryzen Master spits out for CO use are a very good starting point. I wouldn't take them as 100% stable in all situations and definitely require more stability testing. However, it should shave off quite a few hours of testing if you were doing a proper job testing CO settings in the first place.
 
Dialing in per-core curve optimizer settings can be tricky. The nature of how curve optimizer works means that you are more likely to have instability when your system is idle than at full load (even when testing only one core). When a core is under load, it gives it extra voltage. At idle or moderate load the voltage is lower - sometimes too low with aggressive curve optimizer negative offsets. Core Cycler will get you 90% of the way there but it's not the be all, end all of curve optimizer testing as it's still fully loading the individual core that it tests.

Once I got my per-core offsets mostly dialed-in using Core Cycler, I simply used the system as normal. On occasion I would come back to my computer to find that it had rebooted. I'd check Event Log to see what core failed. Example:

WHEA.jpg

And then I would use CPU-Z to generate a report so that I could see which APIC ID matches up to which Core (These aren't always the same on something like a 5900X, because of the 4 disabled cores). Example:

Code:
Socket 0	
	-- Node 0	
		-- CCX 0		
			-- Core 0 (ID 0)	
				-- Thread 0	0
				-- Thread 1	1
			-- Core 1 (ID 1)	
				-- Thread 2	2
				-- Thread 3	3
			-- Core 2 (ID 2)	
				-- Thread 4	4
				-- Thread 5	5
			-- Core 3 (ID 3)	
				-- Thread 6	6
				-- Thread 7	7
			-- Core 4 (ID 4)	
				-- Thread 8	8
				-- Thread 9	9
			-- Core 5 (ID 5)	
				-- Thread 10	10
				-- Thread 11	11
		-- CCX 1		
			-- Core 6 (ID 8)	
				-- Thread 12	16
				-- Thread 13	17
			-- Core 7 (ID 9)	
				-- Thread 14	18
				-- Thread 15	19
			-- Core 8 (ID 10)	
				-- Thread 16	20
				-- Thread 17	21
			-- Core 9 (ID 11)	
				-- Thread 18	22
				-- Thread 19	23
			-- Core 10 (ID 12)	
				-- Thread 20	24
				-- Thread 21	25
			-- Core 11 (ID 13)	
				-- Thread 22	26
				-- Thread 23	27

Whenever that occurred I'd go into the BIOS and decrease the negative offset on that core. I quickly got to the point where it would only BSOD once per week. Then once per month. Now it only does it once every several months or less, so it's very stable at this point. This trial-and-error method isn't exactly elegant but it got the job done. There really isn't any stress-test program out there that fully replicates real-world usage. Being able to make these adjustments via Ryzen Master could be handy, but overall I've found Ryzen Master to be more trouble than it's worth.
 
I’ll have to see how the curve optimizer’s recommended -30 all core offset works with my 5950X. For now, browsing the web, running Cinebench, and playing various video games has been fine. The clock speeds significantly increased while my temps significantly decreased. All 16 cores are capable of hitting 4.9GHz+, 8 of them hit 5GHz+, 4 hit 5.1GHz+ and one core is hitting almost 5.3Ghz. Haven’t checked what the all core boost frequency is though.
 
I let the curve optimizer run tonight. I had just updated BIOS to the beta 4201 for my Asus Crosshair VIII and have everything stock except my RAM timings.

20220502_220035.jpg


Here's what it came out to, anywhere from -8 to -28 for the cores, with the majority being in that -20s range. I ran cinebench r20 a few times and then ran 3dmark times and it seemed stable enough. Got my best score in 3dmark of 17780. I may run it like this for a few days, then put it in BIOS, maybe toned down a little.
 
i got this, hit apply and reboot. that all there is to it? no idea what its saying either...
5600x
1651544544575.png
 
You got a good chip there just plop those values into bios and see how it goes. I had to increase all by +1 to keep it from restarting at idle but it's been super stable since then no issues at all.
 
Question for you guys that have successfully done this, does it matter that I ran this tool without PBO enabled? Mine was working with stock setting voltage wise. If I set this curve up on BIOS without PBO, do I have a shot at stability?
 
Automatic curve-optimizer seems useless to me. It picked a range of values from -22 to -29 for the cores on my 5600X which isn't remotely stable. I have one core that can't do more than -5.
 
You got a good chip there just plop those values into bios and see how it goes. I had to increase all by +1 to keep it from restarting at idle but it's been super stable since then no issues at all.
so i need to set it in bios? i thought it applied the oc via the software.
 
The amount of CPU's that will do -30 with 200 boost and be stable is going to be low. You are telling the CPU to boost higher with less voltage. I would try to set all cores to -15 with no extra boost, or even -20 then test and go from there.
 
i got this, hit apply and reboot. that all there is to it? no idea what its saying either...
5600x
View attachment 469790
Your power settings seem insane for a 5600X. My 5950X runs at much lower settings. You might want to look at your TDC, EDC and PPT settings and lower them manually. I found I got much better MT and ST scores and performance (plus MUCH lower temps) setting them myself and tweaking accordingly.

Picture of my settings (as of right now) below...

1651584955819.png
 
Mad kudos to you guys for getting RM to run like you want it. For the life of me I could never get it to work. I gave up and just used PBO & Curve Optimizer in bios, viola.
 
My computer got shut down due to a quick power outage today. I can't use my UPS because my computer draws more power than it can handle. Anyway, I verified that Ryzen Master had put the curve optimizer settings into BIOS, but today performance is way off. Much lower scores in CB20. So, I set the BIOS back to stock plus my RAM timings, but it's still off. Way off, CB20 multiscore in the 9000s. I cleared CMOS and used my stock profile. Same deal. Chip is running way cooler because it's not boosting. Optimized Defaults in BIOS got me the same results. Any ideas? I am on the beta 4201 BIOS, but it was fine yesterday.
 
Had to decrease the offset for 2 of my weaker cores to maintain stability. Otherwise 14 of 16 cores for my 5950X are stable at -30. Max boost on one core is 5.3GHz, two cores can hit 5.2GHz, and the rest can hit 5GHz. Pretty nutty!
 
Nice and easy tool to use. Gave me an extra 6.5% cpu score in 3dmark (I know, synthetic) with a constant boost of about 4.85ghz with my 5600X. We'll see if it's stable doing actual work.
It gave me these numbers: -22/-25/-27/-28/-28/-28

My wife just got a new 5600X as well. Going to try it out on hers too.
 
Can someone dumb down what the negative numbers do? I've noticed if I go too much negative, I get reboots.. is there a way to determine what core #'s are causing the issues?
 
Can someone dumb down what the negative numbers do? I've noticed if I go too much negative, I get reboots.. is there a way to determine what core #'s are causing the issues?
its the voltage offset, i think. you could try prime95 and watch for cores to fail the test(stop)
 
So I got my 5950X installed earlier today and was messing around with trying to cool this baby down and decided to give the auto-curve optimizer a go. It attempted to set everything to -30 offset which was so unstable that I had to reset my entire BIOS to get the system to boot again.

I have a lot to learn it seems.
 
My computer got shut down due to a quick power outage today. I can't use my UPS because my computer draws more power than it can handle. Anyway, I verified that Ryzen Master had put the curve optimizer settings into BIOS, but today performance is way off. Much lower scores in CB20. So, I set the BIOS back to stock plus my RAM timings, but it's still off. Way off, CB20 multiscore in the 9000s. I cleared CMOS and used my stock profile. Same deal. Chip is running way cooler because it's not boosting. Optimized Defaults in BIOS got me the same results. Any ideas? I am on the beta 4201 BIOS, but it was fine yesterday.
Had a very similar issue awhile back and the remedy was to clear cmos and reload my previously saved OC profile from bios. It saves a load of time for bios setup ram timings per CO etc.
 
So what I did was to choose some offsets and then I used this utility called core cycler https://github.com/sp00n/corecycler and it just goes one by one through each core using P95. If the core fails with the given offset, it throws and error and goes on to the next one. After the error, I went into bios and adjusted a little and tried again.

The config file is pretty easy to adjust. First I think everything to -30 and then used small FFTs and disabled AVX and ran each core for like 10min. If it popped and error, I took it down. Then re-ran the adjusted cores and re-ran just the ones I adjusted. Got everything to pass at non-avx for ten minutes. Then I upped the time, rinse and repeat. Then I went to full avx. Then I think I went a full P95 all core small FFTs and looked for anything. By then it was pretty much dialed in.
 
Has anyone tried the newer AGESA 1.2.0.7 yet with PBO? I've read mixed reports that performance seems to have been knocked down a little bit with an EDC/Voltage lock when using PBO and/or Curve Optimizer. I'm still on 1.2.0.5 and I get past the voltage lock by setting EDC to 140A in the bios but raising it with Ryzen Master. However, I've read that trick no longer works with 1.2.0.7, so I have not yet tried the new bios (although I have 0 issues right now on 1.2.0.5, so maybe no reason to change?).
 

So I'm trying to get the best pbo limits and curve op for the 5800x. I noticed temps were pretty high n fans always on full blast in gaming so I ran the cpu in eco mode 95w and this seems to have lowered temps but my cpu clocks are still boosting to 4800mhz. I've been reading you can lower temps without sacrificing as much performance as using the eco 95 w mode? But in confused as to how in losing performance when cpu still boosting to 4800? Also read this can be done in Ryzen master in windows but what app is used to stress test the settings does Ryzen master also stress test? Also in running the same bios that the mobo shipped with should I update this to the latest?​

 
Last edited:
1.2B Was just released this morning and well worth the price of admission
So, can you download any of the Hydras without being a patron? Does $3/month get you the current and $10/month get you the early access?
 
So, can you download any of the Hydras without being a patron? Does $3/month get you the current and $10/month get you the early access?
The $10 tier will get you access to the latest builds of Hydra. Yuri updates hydra 1-2 times monthly. It has been a great investment for me.
 
Back
Top