Question Ryzen 7000 overclocking/undervolting advice needed

Paul Anderegg

Distinguished
Aug 30, 2012
81
6
18,535
I've spent the last 5 minutes going through the stickies and the posts in the AMD OC specific threads, but not coming up with anything.

I upgraded my PC from a 9900K to a Ryzen 5 7600X, as a temp CPU until the 7800X3D comes out.

For my 9900K air cooled, I simply synced all cores to 50, set voltage high enough that it didn't crash P95, and made sure the line level load thingy was set to account for droop when the power hits the CPU. The PC ran at 5.0GHz at around 1.3v whatever, and the temps went up and down with power depending on what stress tester I used, anywhere from 120-230watts.

For my new AMD, I am completely lost as to what to do, and the online videos for PBO, or curve optimizing etc, are either all Zen3 related (seeing that Zen4 OC is completely different animal), or are doing weird super power cut stuff to the Ryzen 9 stuff to deal with multithreaded superloads. I just need to throw some simple settings at my 7600X for gaming, specifically Microsoft Flight Simulator, which wants single main thread as fast GHz as possible, and is happiest with SMT disabled. I did some 88000 75000 15000 PBO thing, I also did some negative 3 thing, and then tried the Ryzen Master Eco and OC curve GFX -30 things, but I have no clue what I am doing, other than I am seeing CPU speeds of anywhere from 5150-5450 (5.4 is stock, meh) depending on which of the above I plug in, with watts between 65-130 and temps between 45-90. The only time I ever saw 95c was stock out of the box MB settings with SMT.

I would love suggestions as to which overclocking or undervolting method I should pursue for 6 core SMT off gaming, and any links to up-to-date tutorials for those would be awesome. Looking for something simple I can apply with minimal stress testing at this time. The PBO -3, the PBO curve -30, and the Ryzen Master curve optimizers are the only three methods I am aware of currently, owning my first AMD CPU as of yesterday. I don't do any heavy multi threaded workloads other than stress testing programs. which from what I am reading don't account for undervolted Ryzen 7000 crashing under low load. :)

Paul Anderegg

(was not sure if I should have put this into the AMD OC sticky, but it seems to have died in 2020)
 

Karadjgne

Titan
Ambassador
The 7000 series is days old. You aren't going to find Anything reliable on anything to do with OC or undervolting yet, because it hasn't happened yet.

It's a new architecture on a new platform that doesn't behave exactly like any older architecture.

Same thing happened with the change from FX to Ryzen, lga775 to lga1156 etc. It's going to take time for ppl to figure out just what's possible and what's not.
 

Paul Anderegg

Distinguished
Aug 30, 2012
81
6
18,535
Doing nothing but entering 88000 75000 15000 into curve optimizer resulted in a Cinebench 23 single core score of 1926, 5450MHz 40 watts and 50c average 1.37 VID. I didn't wait for a multicore score because SMT off, but hitting all 6 cores 5300MHz 90 watts and 65c average 1.30 VID. This is really all I have done to the thing, just that manual curve optimizer.

Without knowing these CPUs and Ryzen 7000 specifically, I have no way to tell where the thing can be brought up more.

Paul

*found an article showing stock the 7600X should do 1954 C23 single core :(
*additional test with stock curve CB23 single same, multi 5375MHz 105 watts and 75c average 1.375 VID
 
Last edited:

Karadjgne

Titan
Ambassador
With what ram? Ram speeds, Secondary and Tertiary timings adjustments, fclock settings, there's multiple ways to improve Ryzen stock cpu settings and not really do anything.

Can't rely strictly on videos or articles for exactness of what's possible because that pc as tested is different to yours. Same ram, same cpu, different mobo will end up with different results that may or may not be in your favor.

Cinebench cannot give exact results the same every time with a Ryzen. It's impossible. A fraction of a degree difference in 1 core under load can and will change the amount of time that core spends at any given frequency, since the cores are dynamic not static.

So that score of 1954 may have been on a semi-cool cpu whereas yours as tested had been warmed up. Or vice versa. You cannot tell. 1954 and 1926 are close enough to be within margin of error the same thing.
 

Paul Anderegg

Distinguished
Aug 30, 2012
81
6
18,535
I have cheap 5200 CL40 32GB kit for now while my Z5 Neo 6000 CL30 kit ships from China or wherever.

Not as much interested in the CB23 score as much as the clocks juts seem to be running all stock and I don't know how to make them go faster. Should I look into Zen3 overclocking guides and just ignore my temperatures, since temps are not an issue for my 6 cores or not so multicore workload?

I was expecting the damn thing to boot at 95c and just stay there all the time, so not sure what all that Ryzen 7000 temps panic is all about...maybe for stress testers on a 7950X.

Paul
 

Karadjgne

Titan
Ambassador
Ryzens are funny. They don't behave like Intels at all. Ryzens are very temp sensitive in that IPC is related to temps.

For instance, my 3700x gets higher CB20 scores at a 4.28GHz all core at 62°C than it does at a static 4.40GHz OC at 80°C. Wish I could find that article, but basically what it boils down to is a Ryzen 'thinks' quicker in the 50-70° range than outside those temps.

I guess it boils down to data and bandwidth connectivity when traces and circuits are warmer or colder than a certain comfort zone between the ram, i/o and cores on the infinity fabric.

So right now you have one of the best performance cpus there is, ties or beats a 12900k in almost everything, at a fraction of the power draw and cooling.

You literally do not need to OC it. At all. You have plenty of time to wait and see what the pros come up with, upsides and drawbacks, tinker with ram settings for a little tweak in performance and still be in the top5 best and fastest cpus
 

Paul Anderegg

Distinguished
Aug 30, 2012
81
6
18,535
I think I am technically undervolted, as I entered those lower 88/75/150 curve figures into the BIOS and although the speed didn't increase, it is running cooler.

I mean, it is still kicking my old 90c 5.0 GHz 9900K overclock, which also was stuck with 16MB of L3...eww.

Paul
 

Karadjgne

Titan
Ambassador
Cpus are greedy bastages. Seriously spoiled greedy brats. Give them Auto vcore and they'll demand the VRM's supply them with 1.45v. Even though they only use 1.1v.

That Optimizer is basically limiting the VID. It's the Nun with a ruler rapping the knuckles of the cpu when it gets out of line and makes stupid demands. So when the cpu demands 1.45v, Nun steps in, whack the ruler, cpu gets 1.2v, and only uses 1.1v.

If the cpu objects and says it really does need at least 1.3v for stability, Nun agrees and gives it 1.35v.

The trade off is lower temps, relaxed VRM's and a smoother power delivery because the VRM's and other voltage regulatory circuitry isn't forced to be constantly supplying voltage the cpu isn't going to use. Less stress, less waste, lower temps = happier Ryzen.

That can all be seen in HWInfo64, the supply, the demand, actual use. There are other voltage considerations, such as that used by memory controller, or other parts of the cpu not just the cores themselves, so part of that demand and supply is for those components, so even if vcore is only 1.1v, the extra goes to them for a total package use, which will be higher, so supply will need to be somewhat more than vcore use.
 
Last edited:

Paul Anderegg

Distinguished
Aug 30, 2012
81
6
18,535
AMD is like the Linux of overclocking. :p

Paul

Cpus are greedy bastages. Seriously spoiled greedy brats. Give them Auto vcore and they'll demand the VRM's supply them with 1.45v. Even though they only use 1.1v.

That Optimizer is basically limiting the VID. It's the Nun with a ruler rapping the knuckles of the cpu when it gets out of line and makes stupid demands. So when the cpu demands 1.45v, Nun steps in, whack the ruler, cpu gets 1.2v, and only uses 1.1v.

If the cpu objects and says it really does need at least 1.3v for stability, Nun agrees and gives it 1.35v.

The trade off is lower temps, relaxed VRM's and a smoother power delivery because the VRM's and other voltage regulatory circuitry isn't forced to be constantly supplying voltage the cpu isn't going to use. Less stress, less waste, lower temps = happier Ryzen.

That can all be seen in HWInfo64, the supply, the demand, actual use. There are other voltage considerations, such as that used by memory controller, or other parts of the cpu not just the cores themselves, so part of that demand and supply is for those components, so even if vcore is only 1.1v, the extra goes to them for a total package use, which will be higher, so supply will need to be somewhat more than vcore use.
 
  • Like
Reactions: greenreaper

Paul Anderegg

Distinguished
Aug 30, 2012
81
6
18,535
I have an LF2 280mm A-RGB on top of my O11 Air Mini. I didn't know how to do duplicate runs in MSFS, and the main concern going forward was the L3 cache increases of Ryzen (32MB vs 16MB on my 9900K) and the upcoming X3D helping with stuttering and VR. The average fps doesn't do much for me as I like to run locked v-sync as I don't have a VRR TV, so anythign above or under 30/50/60 vsync looks stuttery and bad for immersive experience in 2D, and chopy ucky in VR.

Paul
 
lol how times have changed that right now people talk about 5450MHZ like if it was 1 GHz.

Game perfomance comes in many ways with newer CPUz, is not always tie to Max Frecuency. Of course one would like to get the highest stable frecuency avialable.

If I were in your place, after the many reviews, investigations, etc. that I have saw and read, I would set my BIOS back to default, go to MSFS and try to measure and log the temp, max frecuency and avg and .1 low FPS while landing a plane at your favorite airport.

Then I would go back to BIOS look for the PBO2 Core Curve option, and try for example a "10", go back to MSFS and do it all again.

Then compare the numbers, and repeat the process till you find the best performance. Keep in mind that depending on your platform (RAM, GPU, cooling device, etc) maybe stock conditions are the best of all.

And yes I know MSFS is not the easiest game to bench but I don't think there much to do other than try. Unless you wana wait till you can find a guide for optimizing MSFS with a Ryzen 5 7600X :)

I hope someone else can give you another idea/option/suggestion.
 

Paul Anderegg

Distinguished
Aug 30, 2012
81
6
18,535
By the time people figure out the secrets to getting more boostyjuice from the little 7600X, the 7800X3D will be in my system :)

I am still trying to figure out how they do the side by side "identical" benchmarks for MSFS, not the "similar" but the "identical" ones back to back, which is the only way I would feel comfortable comparing any OC mods.

Paul

lol how times have changed that right now people talk about 5450MHZ like if it was 1 GHz.

Game perfomance comes in many ways with newer CPUz, is not always tie to Max Frecuency. Of course one would like to get the highest stable frecuency avialable.

If I were in your place, after the many reviews, investigations, etc. that I have saw and read, I would set my BIOS back to default, go to MSFS and try to measure and log the temp, max frecuency and avg and .1 low FPS while landing a plane at your favorite airport.

Then I would go back to BIOS look for the PBO2 Core Curve option, and try for example a "10", go back to MSFS and do it all again.

Then compare the numbers, and repeat the process till you find the best performance. Keep in mind that depending on your platform (RAM, GPU, cooling device, etc) maybe stock conditions are the best of all.

And yes I know MSFS is not the easiest game to bench but I don't think there much to do other than try. Unless you wana wait till you can find a guide for optimizing MSFS with a Ryzen 5 7600X :)

I hope someone else can give you another idea/option/suggestion.
 

Paul Anderegg

Distinguished
Aug 30, 2012
81
6
18,535
View: https://youtu.be/QkoS-bHyTZM


I think the manual undervolting looks like the best for Ryzen 7000, 1.1v or so here, and my ECOm config is in the 1.4v's for the same -100MHz speeds and performance. Everything I am seeing is showing these CPUs simply do what they do stock and you can't get anything more out of them using currently known methods. #staytuned

Paul
 

Paul Anderegg

Distinguished
Aug 30, 2012
81
6
18,535
Using Skatterbenches suggestions, I enabled PBO +200 all core, negative curve 15. I have an ASUS board, so also enabled the Medium Load Boostit. AGESA 1002. SMT off so only 6 cores for gaming results here.

Vs my PBO limits undervolting, single core speed went from 5450MHz to 5650MHz, 45w to 60w and 53c to 65c. The all core speed went from 5250MHz to 5475Mhz, 90w 65c to 103w 73c. P95 small AVX enabled went from 4925MHz o 5300MHz for those that do a lot of video rendering using P95 :p

Time to install the AGESA 1003 firmware from ASUS and see what happens. I would be happy if these little bottom end 7600X's could do flat 5.7GHz single and all core.


Paul
 
Last edited:
  • Like
Reactions: RodroX
Using Skatterbenches suggestions, I enabled PBO +200 all core, negative curve 15. I have an ASUS board, so also enabled the Medium Load Boostit. AGESA 1002. SMT off so only 6 cores for gaming results here.

Vs my PBO limits undervolting, single core speed went from 5450MHz to 5650MHz, 45w to 60w and 53c to 65c. The all core speed went from 5250MHz to 5475Mhz, 90w 65c to 103w 73c. P95 small AVX enabled went from 4925MHz o 5300MHz for those that do a lot of video rendering using P95 :p

Time to install the AGESA 1003 firmware from ASUS and see what happens. I would be happy if these little bottom end 7600X's could do flat 5.7GHz single and all core.


Paul

Everyones free to test their own way, but one thing I do everytime I update BIOS is to load and save optimized (default) bios settings, before and after I upgrade my BIOS.
 

Paul Anderegg

Distinguished
Aug 30, 2012
81
6
18,535
I did do the F5 reset to default, started the update,a nd now 20 minutes later looking at a black screen and solid red LED inside the case. I think 20 minutes of brick black death means nothing good has taken place :(

Paul
 

Paul Anderegg

Distinguished
Aug 30, 2012
81
6
18,535
Correct BIOS, but I have always had issues with ASUS bricking BIOS.

Regardless, BIOS flashback successful, AGESA 1003 update shows identical performance and clocks as the 1002 BIOS. P95 did gain 23MHz for some reason all cores...odd.

Paul
 
  • Like
Reactions: RodroX