How to limit vcore/clock Ryzen 5900X

[Tuvoc]

Hi,

My 5900X tries to maximise performance by significantly increasing vcore to attain ever high clocks. Sometimes that is welcome, but other times not. It generates a lot more heat and subsequently noise from the cooling solution. We know the law of diminishing returns, and for at least some of the time I want to achieve a better compromise.

So mostly I'm running 12 threads, and the clock sits at 4.5GHz with vcore at about 1.3v. I'd like to limit the maximum clock to 4GHz and with that would come less vcore and less heat. How much less vcore that would be I don't know. Ideally I'd like to use Ryzen Master for this so that I can toggle it back to full performance when needed. But I can't see how to do this.

I did try reducing the TDC and that did bring the clock speed down, and the vcore to 1.1v (and 20C less heat) on the 12 threads. Great I thought, but Prime95 immediately failed on a couple of cores so it seems the motherboard/CPU were not intelligent enough to work out the right balance.

Motherboard is MSI B550M Mag Mortar Wifi. If the best way to do this is through the BIOS then that is what I'll do.

Any suggestions for me much appreciated. Happy new year all.

 

[Twinz]

You could try curve optimizer, on a per-core basis. (BIOS) Very time consuming to nail down, but could probably get you the results you want. (After a lot of trial and error)

Also in the BIOS, is the temp limit. That may be a more direct way since agressive single core boost can generate a temp spike on that individual core and the temp limit should pull voltage and clocks to stay within your set limit.

 

[Tuvoc]

Thanks - perhaps just changing the TDC value by itself was not the right approach. I'll further experiment with all three values, maybe they need to be reduced proportionately. Or else maybe PPT by itself might achieve it.

 

[Tuvoc]

Looks like simply lowering PPT is the answer.

 

[Cyber-Mav]

Ppt is the way to go. However do try setting pbo to manual and leave scalar at 1x or auto. Can also try curve optimization of -5 across all cores.
If pbo is on auto it tries to auto oc and uses silly high voltages.

My 5600x is on stock clocks but ruining 105w ppt which raises its all core boost clocks a fair bit. Overnight video encoding going on and its mostly at all core 4.52ghz and 83c temps. In prime95 clocks are lower around 4.3ghz. however in prime95 using 8 threads it will boost to 4.6ghz and hold. Temps at 82c.

Rather than lock the clocks to 4ghz make use of zen3s amazing boost algorithms and allow it to clock up on its own to stock boost clocks.

Ppt is actually higher on these chips but you can manually set it to 105w and then run some benches and observe clock behaviour. Then if your cooling is up for it keep raising the ppt and test with prime95 small fft to see if you can handle the heat. If heat is under control then keep going up on ppt and benching. More ppt will raise your all core clocks at expense of heat. Chip should run fairly cool at a manual locked 105w and be very power efficient.

No point limiting edc or tdc since those are current values and you just choke the CPU of current.
You cant over supply current since its a demand based value. You can however over supply voltage thats where some trial comes in with negative curve optimiser values.

Id like to go higher on my 5600x on the ppt but all depends on how comfortable i am with the temps. In video encoding and prime its up to 83c so still far from the 95c throttle point but im outside my comfort zone with temps over 85c.
Maybe better thermal paste can get my temps down from 83c to 80c in which case id then use 110w ppt to end up back at those 83ish c temps.

Lot of trial and error.
Just remember auto oc in pbo does get a bit excited with the voltages hence why i dont overclock and aim for higher than stock single thread speed. More ppt gives me faster video encoding performance.

 

[Joxeon]

@Tuvoc you can make a ryzen master profile for 4ghz all core and 1.1v or slightly more if it's unstable then you can flick between that and the normal settings depending on what workloads your running.

@Cyber-Mav have you tried an all core OC for the video editing as I'm sure you could run 4.5 at 1.25 np which you give you lower temps than your PBO settings and you could probably push 4.65-4.7 @ around 1.3v which would likely give compatible temps to what your seeing right now.

 

[Cyber-Mav]

doing an all core oc for video encoding doesnt run any cooler. here is a screenshot pc has been running overnight doing encoding and is still encoding as i type, lets look at the results:

it seems to be doing 4.53ghz on all cores, around 1.2v and pulling 103.5w of the 105w limit with peak ppt pull of 104.947w at times. temps hit a peak of 82.3c. curve optimizer is being used i think i have it on -10 or -15 on all cores. need to double check in bios. -20 was a no go resulting in crashes.

now i will get some more screenshots up after this encoding session is done, but from what iv seen in some loads like like winrar multithread bench the cpu will sit at 4.65ghz on all cores and pull around 67w of the ppt. so using a locked clock is not a good idea. dynamic works best with zen3.

also as the handbrake screenshot shows the encoding is going around 4.53ghz but in prime95 the clocks drop to 4.3ghz at same 105w ppt. running fixed clock forcing it to 4.5ghz wont work for heavy loads and thats where stability issues arise.

those with a 5900 or 5950x should have more power headroom since im putting 105w through a single ccx its gonna get hot. (air cooler arctic freezer 34, fan is going full whack 2000rpm its loud). having dual ccx means power is spread over more surface area so it should not hot spot so easily on multithread load.

 

[Joxeon]

Maybe it's changed since Zen 2 then as my old 3600 would run way cooler with a locked 4.2 all core over the default all core boost of 3.95-4.0

 

[Tuvoc]

Ppt is the way to go. However do try setting pbo to manual and leave scalar at 1x or auto. Can also try curve optimization of -5 across all cores.
If pbo is on auto it tries to auto oc and uses silly high voltages.

My 5600x is on stock clocks but ruining 105w ppt which raises its all core boost clocks a fair bit. Overnight video encoding going on and its mostly at all core 4.52ghz and 83c temps. In prime95 clocks are lower around 4.3ghz. however in prime95 using 8 threads it will boost to 4.6ghz and hold. Temps at 82c.

Thanks, coincidentally the last few hours I did set PPT to 105w and that seems t be doing the job very nicely. It is running around 4.1GHz and 1.18V and I'm sure that is rock-solid. Good reduction in temps and noise compared to letting it run up to 4.5GHz at 1.3v.

@Joxeon, Yes I've named a profile for this reduced setting, and if I want maximum performance again there is a reset option to click which puts things back to stock

Thanks for the help guys.

 

[Cyber-Mav]

Does it still boost up to higher clocks on lighter loads?

 

[Cyber-Mav]

Maybe it's changed since Zen 2 then as my old 3600 would run way cooler with a locked 4.2 all core over the default all core boost of 3.95-4.0

Yes from what i read with zen2 all core clocks is the way to go.
Zen3 just keep raising ppt till the heat is too much. But on the 5600x iv noticed going over 95w ppt gives diminishing returns as in heat goes up a lot for not much rise in clocks.
Keeping zen3 in its efficiency zone works well. Hoping to pick up a 5900 or 5950x once people sell up and move to 3d vcache

 

[Tuvoc]

Does it still boost up to higher clocks on lighter loads?

Me ? Yes, it still boosts up to 4.9 under 1 or 2 cores, at silly high voltages, because it can do that and still stay within the 105w TDP. It is not very often that the machine is in that situation.

 

[Cyber-Mav]

I dont like the silly high volts dont sit right with me. Maybe you can set a negative frequency offset of -200mhz so it boosts to 4.7ghz insted and it may max voltage at around 1.35v to get there?
Dunno if thats possible would need to experiment some.

 

[Cyber-Mav]

ok the pc has finished its encoding session so ran this quick test for you to show in winrar how the clocks boost and its power usage:

it goes to full boost on all cores at 4.65ghz and uses around 63w of ppt. if i had locked in a all core clock of 4.5ghz or less then in situations like these (and there are plenty of them) id loose out on the higher boost bins. there is many low loads on many apps that can trigger and hold high boost since the load is not high enough to hit any power limits. in situations like these i could tune the system to add a +200mhz frequency offset so it boosts to 4.85ghz however cant get too greedy here since it end up hitting 1.5v to get those clocks which im scared of. also temps would rocket up a lot with high volts.
got to find that sweet spot. as said before with my testing 95w ppt is a good balance for the 5600x.

as seen above with zen3 fixed clocks is only good if you can fix them at the highest single thread boost clock otherwise your loosing out on low load boosts. and to run all core at high clocks its gonna need high ppt limit and probably lot of volts.

 

[Tuvoc]

I dont like the silly high volts dont sit right with me. Maybe you can set a negative frequency offset of -200mhz so it boosts to 4.7ghz insted and it may max voltage at around 1.35v to get there?
Dunno if thats possible would need to experiment some.

Yes I don't like it either, although it is by design. its no different from my Intel 10900 actually, which boosts to 5.2GHz on 1 core at around 1.43v (although typically 5.0GHz with less vcore). It seems more tolerable on that though, the cooling system doesn't react to it very much at all, maybe it is reading a different sensor perhaps. The overall socket temp surely can't be increasing much. On the Ryzen though when 1 core boosts like that the fans ramp up as if the entire CPU was fully loaded.

 

[Cyber-Mav]

Yep noticed that too when 1 or 2 cores are used at full load they both get to 80c fast and fans ramp up making noise.
Guess its just too much heat in too small physical space.
Probably get worse as it drops to 5nm

 

[DarkBahamut]

as seen above with zen3 fixed clocks is only good if you can fix them at the highest single thread boost clock otherwise your loosing out on low load boosts.

I'm going to slightly disagree with that. The problem is how Ryzen boosting behaves. We can see it in the screenshots you posted above to use an example. The two datapoints to pay attention to are:

Core clock: 4,656.6 MHz
Effective core clock: 4.131.0 MHz

What this tells you is in any instantaneous moment your CPU was doing 4.65 GHz, you can sort of consider this to be the requested clock speed. However, on average during the polling period your CPU was actually running at 4.13 GHz. There will be a few reasons why the true average clock speed is down, but some of it is how the SMU handles the boost mechanics, and that the maximum frequency is only available for very short periods of time.

Another example using a fixed(ish) frequency OC on a 5950X I have here using Cinebench R15 single core to test.

Stock PBO: 258 points - Max clock - 4.9 GHz, Max volts 1.45 V, Average CPU temp 66 C
Fixed OC: 255 points - Max clock - 4.725 GHz, Max volts 1.25 V, Average CPU temp 54 C

PBO has nearly 200 MHz extra, but where is the performance? It's certainly not realised in the running application, and that’s because the true achieved average frequency isn't really any better, despite HWinfo/AMDs SMU reporting much more pleasing frequency numbers. You can see the delivered Vcore is much higher though given the large difference in thermals during both runs. I'd personally take a profile switched fixed OC any day of the week on Ryzen to avoid the huge boost voltages, especially considering in most instances the performance is no different, or actually better in heavier workloads (where the saved power/thermals gives you more frequency headroom).

Many people just see 4.9/5.0/5.1 GHz in HWinfo and assume they're getting that fixed level of performance, and therefore fixed OCs can't compete. Usually that isn't true unless you have some fantastic sample (the ones that do -30 CO with +200 MHz boost offsets)

 

[Cyber-Mav]

thats interesting to see. as i raise the ppt wattage score does go up in cinebench.

id like to see if i can do an all core 4.6ghz oc but after seeing what happens in prime95 which pulls 105w ppt and clocks are much lower 4.3ghz ish i cant see it being stable. id like to try this though and see how all core clock works out.

got some info here:
https://www.hwinfo.com/forum/threads/effective-clock-vs-instant-discrete-clock.5958/

one thing is for sure, i cant run 4.6ghz all core in prime95 without raising ppt to probably 140w+ and that means temps will be out of control 100c+ etc. this is also why those that dont have a higher power limit for the cpu set dont see any performance increase in benches and they usually loose some performance.

lot of trial and error. but for more performance it comes at the price of power usage. and since i prefer to use prime95 for max load testing tuning the oc for lighter workloads wont give the stability if something harder comes along.

 

[Cyber-Mav]

Stock PBO: 258 points - Max clock - 4.9 GHz, Max volts 1.45 V, Average CPU temp 66 C
Fixed OC: 255 points - Max clock - 4.725 GHz, Max volts 1.25 V, Average CPU temp 54 C

this is very interesting, i will give this a test. i will try a fixed of of something like 4.4ghz and 1.25v and see how it affects cinebench scores.

 

[Cyber-Mav]

just did the test at 4.4ghz and 1.25v fixed clock. this is not gonna work for me.

temps went out of control as soon as prime was fired up. and power pull is much higher.

my cinebench r20 score dropped to 4400 with the fixed clock. before it was at 4593.