[SOLVED] Ryzen 3600 not reaching 4.2 GHz in single thread applications

[sowa705]

Hello, im having a problem with my Ryzen 3600, i cant get it too boost to advertised clocks even in non AVX single threaded applications. For a while i ran a static 4.2GHz all core OC at 1.325 V but i abandoned it to avoid long term degradation, now im trying to get PBO+Auto OC to get something close to my manual OC but with proper auto voltage control. Im using 395W PPT, 255A TDC and 200A EDC with 10X PBO scalar and a -0.1V offset to get to stock voltages but the highest clock i get is just 4050 MHz on a single thread and 3950 on all threads (stock is about 100MHz worse) and im running between 60 C (single thread) to 75 C (prime 95 small FFTs) so temperatures aren't a problem. What settings should i adjust to get higher frequency without degrading my CPU?

Specs:
CPU: AMD Ryzen 3600
MOBO: Asus Prime B450-Plus
RAM: 2x8GB G.Skill Ripjaws V 3200 MHz CL16
GPU: XFX RX580 8 GB
CPU Cooler: Silentium PC Grandis 3 (extremely overkill for a 65W CPU)

 

[hotaru.hino]

Reset the CPU settings to defaults to make sure the processor functions more or less normally.

 

[drea.drechsler]

Hello, im having a problem with my Ryzen 3600, i cant get it too boost to advertised clocks even in non AVX single threaded applications. For a while i ran a static 4.2GHz all core OC at 1.325 V but i abandoned it to avoid long term degradation, now im trying to get PBO+Auto OC to get something close to my manual OC but with proper auto voltage control. Im using 395W PPT, 255A TDC and 200A EDC with 10X PBO scalar and a -0.1V offset to get to stock voltages but the highest clock i get is just 4050 MHz on a single thread and 3950 on all threads (stock is about 100MHz worse) and im running between 60 C (single thread) to 75 C (prime 95 small FFTs) so temperatures aren't a problem. What settings should i adjust to get higher frequency without degrading my CPU?

Specs:
CPU: AMD Ryzen 3600
MOBO: Asus Prime B450-Plus
RAM: 2x8GB G.Skill Ripjaws V 3200 MHz CL16
GPU: XFX RX580 8 GB
CPU Cooler: Silentium PC Grandis 3 (extremely overkill for a 65W CPU)

How long did you run it with a fixed OC and VCore? You may already have degraded it if for long and doing heavy processing work, like video rendering or transcoding.

But...for some exotic PBO settings to try, especially if yours is an early silicon 3600 (one built around and just after the Zen 2 launch):

Set PPT 330, TDC 120, and EDC of 10.

Yes....EDC=10. It's a bug....or maybe loophole...in the boosting algorithm that lets the processor hold higher boosting in heavy work. But the downside is you can lose light threaded performance (single thread) if you don't also disable Advanced C States. It may not work but it's something to try and you may have to experiment a bit in the EDC 5 to 15 range. When it works, it works pretty well, as with my early silicon 3700X.

Some other pointers, but these also apply even if not exploiting EDC=10: first, make sure CPPC is enabled in BIOS and you're running a 'later' BIOS. It doesn't have to be one for Ryzen 5000, in fact it may be better if not. Make sure your chipset drivers are up-to-date, and run one of the Ryzen power plans; I use a 1Usmus plan.

A slight negative VCore off-set can help, but too much will definitely hurt boosting. Running a higher scalar is much worse than leaving VCore higher, that's because the high scalar delays the voltage pull-back to hold high boosts longer. That means voltage isn't allowed to drop away so quickly at high-temp conditions, the time when greatest degradation occurs. As long as VCore isn't spiking above 1.5V in light threaded boosting it's perfectly safe, and with a lower scalar setting it will pull it back quickly when temperature gets out of bounds for it's FIT values.

I'm not sure what your 'AutoOC is', but running an additional core boost of 200-225Mhz is often helpful. It will not do +200Mh, but it does hold about 25Mhz higher clocks after 10 min's of Prime 95 on my system.

And last: cooling, cooling, cooling. Ryzen won't toast itself even on stock cooling but better cooling means it will hit high clocks more eagerly and hold them longer. In fact, you probably won't see any performance gain at all from PBO if you don't have better than average cooling simply because it DOES raise heat output considerable. If the temperatures are not controlled the algorithm is still there to lower clocks and prevent degradation.

 

[sowa705]

I used fixed OC for a month or two, so hopefully degradation isnt that bad (i kept temps close to 70C max), EDC bug kills the ST performance which unacceptable IMO, unfortunately disabling C states is not an option for me as i like working power management. im running the latest bios (June 2021) but i havent noticed any differences compared to older versions, as for the VCore offset im already using -0.1V, would increasing VCore while reducing the PBO scalar improve clocks? Basically i'd like to force PBO to clock my CPU higher while keeping voltages the same.

 

[sowa705]

For reference here are my ryzen master settings, reported voltage is incorrect - subtract 0.1V from it

 

[drea.drechsler]

I used fixed OC for a month or two, so hopefully degradation isnt that bad (i kept temps close to 70C max), EDC bug kills the ST performance which unacceptable IMO, unfortunately disabling C states is not an option for me as i like working power management. im running the latest bios (June 2021) but i havent noticed any differences compared to older versions, as for the VCore offset im already using -0.1V, would increasing VCore while reducing the PBO scalar improve clocks? Basically i'd like to force PBO to clock my CPU higher while keeping voltages the same.

Try the setup...see if EDC=10 is netting you an actual performance improvement. It's not even worth going further if you can't realize a performance improvement. With my system, it's very real. I measure it with CB20 and CB23 so it's measured in a real-world work too, not synthetics.

Disabling Advanced C States hasn't proven the problem I thought at first. With my EDC=10 PBO tweak and advance c states disabled, at IDLE core power is running around 4-8W minimum, it of course runs a lot higher when it boosts. With C States enabled core power runs the same, I can't really tell a difference. I think that's acceptable and makes sense since the other C-states are still functioning, in particular C-1 which stops the CPU from clocking. So power management is still functioning as it only disables C-6 sleep state. And of course, C-States has no bearing when CPU cores are working hard.

Raising voltage/lowering scalar may have no influence on 'clocks', or it may have considerable. Too much depends on your motherboard's presets by the manufacturer. You have to experiment.

But as far as experimenting: don't rely on looking at clocks to assess performance. Ryzen's way to dynamic with how much it dithers clocks and voltage. You have to look at actual performance in a real-world task, that's why CB20 and CB23 are so useful.

 

[sowa705]

Try the setup...see if EDC=10 is netting you an actual performance improvement. It's not even worth going further if you can't realize a performance improvement. With my system, it's very real. I measure it with CB20 and CB23 so it's measured in a real-world work too, not synthetics.

Disabling Advanced C States hasn't proven the problem I thought at first. With my EDC=10 PBO tweak and advance c states disabled, at IDLE core power is running around 4-8W minimum, it of course runs a lot higher when it boosts. With C States enabled core power runs the same, I can't really tell a difference. I think that's acceptable and makes sense since the other C-states are still functioning, in particular C-1 which stops the CPU from clocking. So power management is still functioning as it only disables C-6 sleep state. And of course, C-States has no bearing when CPU cores are working hard.

Raising voltage/lowering scalar may have no influence on 'clocks', or it may have considerable. Too much depends on your motherboard's presets by the manufacturer. You have to experiment.

But as far as experimenting: don't rely on looking at clocks to assess performance. Ryzen's way to dynamic with how much it dithers clocks and voltage. You have to look at actual performance in a real-world task, that's why CB20 and CB23 are so useful.

I just applied EDC = 10A on my system and i get a nice frequency boost to 4.1GHz ST and 4.0 MT, unfortunately MT performance dropped for some reason while ST performance improved as expected. raising voltage and reducing scalar had an opposite effect to what you described, i got lower clocks and higher voltages. Ill try disabling advanced C states and report in a few minutes

 

[drea.drechsler]

.... raising voltage and reducing scalar had an opposite effect to what you described, i got lower clocks and higher voltages. ...

Try lowered voltage with a slight offset...just one 'notch' to the negative.

But also, don't focus on clocks but actual performance first. That means benchmarks with Cinebench 20. The reason is Zen2's max boost clock isn't really all that important, it's only very briefly and one core at a time after all. What really matters for it's performance is the mid-range boosting when it's working.

You will never get all cores hitting the max boost clocks on a 3600, in a heavy work load. So that should not be an expectation. It really only hits max clocks in light threaded, light work loads. One way I see it is to run a Defender quick scan.

 

[sowa705]

You will never get all cores hitting the max boost clocks on a 3600, in a heavy work load. So that should not be an expectation.

I know that, still i'd like to get 4.2 on a single thread (and stock settings dont even reach 4.0)

But also, don't focus on clocks but actual performance first

Im measuring performance all the time, thats how i know EDC bug actually decreased performance even with high clocks, im worried because my cpu never reached the expected performance, its falling behind a 1700X in cinebench which is ridiculous
Anyway EDC bug seems to decrease multi threaded performance even with C-States disabled so i guess im back to high scalar PBO with a negative voltage offset

 

[drea.drechsler]

I know that, still i'd like to get 4.2 on a single thread (and stock settings dont even reach 4.0)

Im measuring performance all the time, thats how i know EDC bug actually decreased performance even with high clocks, im worried because my cpu never reached the expected performance, its falling behind a 1700X in cinebench which is ridiculous
Anyway EDC bug seems to decrease multi threaded performance even with C-States disabled so i guess im back to high scalar PBO with a negative voltage offset

It's also important to look at cooling...both case and case fans as well as the CPU.

But also...are you using Ryzenmaster as your only overclocking tool? If so, that can be part of the problem. Some have said it can be a performance drag while running. Which seems reasonable since it was intended as an overclocking tool for extreme overclockers to use during demonstrations and competitions, not really a performance enhancing tool. I'd only do this in BIOS and uninstall RM.

 

[sowa705]

It's also important to look at cooling...both on the CPU as well as case and case fans.

Im running under 70 C in prime95, as i stated before i have an pointlessly overkill cpu cooler rated for 250W TDP

But also...are you using Ryzenmaster as your only overclocking? If so, that can be part of the problem. Some have said it can be a performance drag while running.

Im using ryzen master for monitoring only, all changes are done in BIOS, ill try to do some benchmarking without it

 

[drea.drechsler]

Im running under 70 C in prime95, as i stated before i have an pointlessly overkill cpu cooler rated for 250W TDP

Im using ryzen master for monitoring only, all changes are done in BIOS, ill try to do some benchmarking without it

Just pay attention to case ventilation too...any air cooler is rendered ineffective if left to recirculate it's own hot air.

RM's not very good even as a monitoring tool. You might get HWInfo64 for that, much better and much, much less intrusive. Also, when running a CB20 BM always do it after a re-start and with nothing else running. It makes for best and most repeatable results. Some of the review site methodologies even leave the system untouched for ~20min's before starting the BM run.

 

[sowa705]

Just pay attention to case ventilation too...any air cooler is rendered ineffective if left to recirculate it's own hot air.

I have pretty good ventilation to the point i have my own custom front panel thats about 60% holes by area, airflow is pretty good
As for testing this is the best i get on a clean restart, still far from what it should be but i dont see any options left