5950x optimisation advice

[TheSoupladel]

I have recently just upgraded to a 5950x, which is overkill in my current System of 3060 and 1080p panels but i bought it with the intention of eventually getting better panels and a better GPU when and if things normalise closer to MSRP.

So in my current situation i am not really looking to overclock as its not needed. however i would like to ensure i am getting the most out of the CPU at stock. To that end i have installed it to my motherboard (Crosshair VIII Hero) with a Noctua DH-15 and left everythng at default with the exception of enabling DOCP for the RAM.

The testing i have done so far show i am only hitting 5Ghz occasionally on two cores, a few other on 4.9 but most of the rest on much lower speeds. on Multicore laods such as Cinebench, all cores clock down to around 3.9Ghz which i understand is expeceted behaviour for a sustained mutlicore load.

Is there anything i can do to bring things up a little bit under single core performance such as gaming? i have read up a little bit on curve optimiser and is that about the only area i can get into to improve things?

 

[TheSoupladel]

I'd just leave it at stock or maybe adjust the curve optimiser a bit for now as its not like overclocking the CPU will make any difference on a card like a 3060 in gaming, even on high end cards there is very little to gain from overclocking the CPU as AMD pretty much maxes them out of the box.

Overclocking the GPU would be more worthwhile though as you may gain around 5-10% extra fps.

I am not looking to overclock specifically, just to make the most of the 5950x at stock or close to stock. the fact i am not hitting 4.9 on a substantial number of the cores despite adequate cooling suggests something is perhaps not quite right, either that or i really crapped out on the silicon lottery

 

[TheSoupladel]

Reasonable "plug and play" settings to give a go. Manual PBO settings, PPT 270, TDC 150, EDC 190 and then work on your curve. Try a minus -10 all core undervolt and then run this https://github.com/sp00n/corecycler/releases

The "problem" with doing an all core undervolt is one of your better cores might literally tap out at -5~-10, whereas the weaker cores might get up to -25~-30. If even one core is unstable it might lead to random reboots or issues weeks if not even a month down the line.

I've got one of my best cores at -5, the other 3 better cores at -15 and the rest at -25.

There is of course no such thing as "plug and play" settings with anyone's individual CPU, other than going almost full auto, so it's worthwhile giving this a watch too https://youtu.be/dU5qLJqTSAc

These chips are less about overclocking and more about finding out how to feed them with the minimum voltage/power needed so they run cooler and therefore might boost higher. That's why you can just stick with mostly auto if you can't be bothered tweaking, all you'll probably "lose" out on is a slightly better boost and you might have more heat to get rid of with your cooling system.

Will give this a bash a little later.

Something i didn't do when putting the 5950x in and was also something i didn't do with the 3900x previously as connecting the additonal 4pin CPU power connector, as i understand it that is only used when undertaking extreme overclocking/LN2. Am i right in saying that or is there any benefit to connecting that in normal use?

 

[TheSoupladel]

The 4.9 is only the single core boost for light workloads, when running stuff like cinebench all core then your more likely to be around 4.4

I am aware that the single core boost is for light laods, however in the research i have done so far, others are able to reach higgher figures on more cores with stock settings during those light loads such as 5.0 on perhaps 4 cores or more where as i am not. So i was hoping optimising the settings might improve this as i couldnt possibly be unlocky in the silicon lottery two CPUs in a row, lol

Cinebench all core drops to around 3.9Ghz for me currently

 

[TheSoupladel]

Aside from not having PBO enabled, are you giving the CPU enough power? Do you have both the 4 pin and 8 pin EPS connectors plugged in?

PBO is enabled, or at least is on Auto.

and as per one of my previous responses, i do not have the 4pin cable connected because i was given to believe that is only required for extreme overclocking/LN2

 

[TheSoupladel]

Reasonable "plug and play" settings to give a go. Manual PBO settings, PPT 270, TDC 150, EDC 190 and then work on your curve. Try a minus -10 all core undervolt and then run this https://github.com/sp00n/corecycler/releases

The "problem" with doing an all core undervolt is one of your better cores might literally tap out at -5~-10, whereas the weaker cores might get up to -25~-30. If even one core is unstable it might lead to random reboots or issues weeks if not even a month down the line.

I've got one of my best cores at -5, the other 3 better cores at -15 and the rest at -25.

There is of course no such thing as "plug and play" settings with anyone's individual CPU, other than going almost full auto, so it's worthwhile giving this a watch too https://youtu.be/dU5qLJqTSAc

These chips are less about overclocking and more about finding out how to feed them with the minimum voltage/power needed so they run cooler and therefore might boost higher. That's why you can just stick with mostly auto if you can't be bothered tweaking, all you'll probably "lose" out on is a slightly better boost and you might have more heat to get rid of with your cooling system.

So in addition to plugging in the 4pin cpu connector, i have run with these settings and -5 on the two best cores (as reported by ryzen master) -10 on the next two cores then everything else at -20.

the results are both encouraging and disappointing because the all core speed under load such as cinebench all cores has come up quite a bit to 4.4Ghz but the single core performance in Cinebench single core has dropped off to 4.95 on the two best cores on CCX0 then 4.75 and 4.725 on the better cores of CCX1

 

[TheSoupladel]

I am running that core cycler just now and have two errors on core 0 and core 2 with ryzen master reporting those as the best cores, it’s onto iteration 2 now.

I should add that since the original post I have made some changes with the 4 best cores on -15 and the rest on -25 my limits are at 270, 200 and 190, so it’s like cores 0 and 2 need to perhaps be on -10.

I presume I have done the right thing in not messing with the voltage direct and leaving that on auto?

 

[TheSoupladel]

Yeah I was already comfortable with the difference in core numbering between Ryzen master and BIOS etc… but that unfortunately hasn’t helped.

I thought I had managed to dial it in at -5 on the four best cores there was an additional two cores at -10 then the rest at -30. I wasn’t seeing any improvement in performance either single or multi core but I figured I would run core cycler at the out setting for 1 iteration to see how that turned out and it fell over at the very first core (core 0 as per bios) despite being perfectly ok in the previous runs.

I have returned to the curve being at auto but sticking with the limits at 270, 200 and 190 which maintains the 4.3 to 4.4 all core performance at sustained load in cinebench. I have not had any time to try out gaming performance to see how it operates in a real world test.

I am therefore open to suggestions as to what I can do to improve on things or have I just got a bit of a lemon of a 5950x, albeit a very fast lemon

 

[TheSoupladel]

The curve is a lot about heat because the way PBO works is it will boost to either frequency cap per power/current supplied or thermal limit, whatever comes first. If thermal limit is coming first that obviously hurts frequency potential.

You want as low voltage as possible being fed to the chip so that heat is as low as it can be. It's a balancing act between supplying voltage/power and handling the excess heat. These chips get very hot, this 5950x runs even hotter than my 3900xt did.

Did you try what I edited in above? Change the OC boost frequency to +50~100mhz on top of the curve modifying.

Also what did you upgrade from? Likely won't make much of a difference but AMD advises not to use Ryzen 3xxx power profiles with 5xxx, but stick to the Windows balanced power plan. If you came from an older Ryzen processor it will have installed the Ryzen power plans as part of the chipset drivers. The 5xxx chipset drivers don't even install them.

I upgraded from a 3900x but i was already using the balanced power plan anyway.

I did try changing the max boost overide frequency to 50 and 100Mhz but it doesnt seem to have done anything.

On the subject of thermals, i am using the Noctua DH15 so i know its reasonably good cooler and in some instances better than some AIOs, however i would be interested to know at what temps does the 5950x with PBO thermal throttle? i am assuming approaching 90? is Hwinfo accurate in confirming whether the CPU has throttled at any point or does PBO have a differing throttle point?

Also, are there other settings that should be changed for PBO in addition to the limits such as the scalar? presumably with the 5000 series there is no need to disbale SMT?

 

[TheSoupladel]

i dont know if this is expected behaviour or not, but i have noticed that if i run CBr23 single core, the clock speed of cores 0 & 2 seem ot fluctuate in an alternating pattern. The fluctuation doesn't seem to be tied to the onscreen completion of each tile either.

 

[TheSoupladel]

I feel like a bit of a muppet for not noticing this before, however it seems in my bios there are two entirely different areas for setting up the limits and curve optimiser one is under tweaker settings and the other is under AMD overclocking, seems stupid but is there a difference between these areas and their effect on speed etc…?

 

[TheSoupladel]

If its an ASUS, yes. Well, for the PPT, EDC and TDC limits. Under the Advanced AMD Overclocking section EDC and TDC are hard-limited at 190 and 160.

Have to use the AI tweaker section to set numbers higher for TDC and EDC.

Other than that I don't think so, just AMD bios jank where there is multiple sections that can do the same thing. The curve optimiser is under the advanced AMD section and that is the only place it exists. At least on ASUS boards.

Ryzen Master will tell you what your PPT, TDC and EDC values are.

I have the Asus Crosshair VIII hero and can confirm there are two complete seperate areas for setting curve optimiser and the limits, they look identical but they arent because i am getting different performance from setting this under "Advanced > AMD Overclocking > Precision Boost Overdrive" as opposed to "Extreme Tweaker > Precision Boost Overdrive".

Until this point i have been doing everything under the Extreme Tweaker menu, inlcuding the limits and curve optimiser and got improved multi core result after adjusting the limits. However after making the same adjustments (including Curve Optimiser) in the AMD Overclocking section the improvements are different again. For example, my CB multicore score had gone up by 1000 points from just over 28,000 to over 29,000. The core speed in the same test has gone from 4.325 after settling down to 4.45.

I also saw the maximum core frequency go up to 5Ghz on multiple cores when idling (info from HWinfo64) but thats where the good news end though, i am still not getting 5Ghz in any meaningful way in single core testing either in CB or in Games. I appreciate it wouldn't go to 5Ghz and stay there but its just not going there at all. I am currently redoing the dance with corecycler and my previous result (when set in AMD Overclocking) have caused a couple hard resets, therefore i am currently running with 0 on the best two cores -10 on the next 4 cores and -20 on everything else.

I still have fair bit to go but i am a little bit more hopeful i might actually get there now.

 

[TheSoupladel]

I am doing a second round of CoreCycler and Prime95 has thrown an error on Core's 0 and 2 which are untocuhed i.e there is no +/- entry against them. Might this indicate an issue with the CPU or just that occasional errors are thrown up that have nothing to do with the CPU?

Edit: Am i better running with Aida64 for Corecycler? have been using the default prime95 and wonder if that might be the issue?

 

[TheSoupladel]

When you say untouched you mean with a 0 on the curve? Best way to confirm is to turn the curve completely off by setting it back to auto and test CoreCyler with those two cores. They shouldn't be crashing if no undervolting is going on.

As for over 5ghz, you're unlikely to see it in games. My HWINFO can show my best cores touching 5.05ghz, and I've even once seen a 5.1ghz, but when games are running it's mostly up to 4.9ghz, and even that's fleeting boosts. When multiple cores are going it's more in the range of 4.7~4.8. Heavy benchmarking on all cores drops it to 4.6~4.65.

Maybe if you're playing a really low spec 2D sidescroller or something, you might get some higher boosts because it's running on one or two cores more often.

It wasnt a crash as in BSOD or a reboot, Corecycle threw up an error talking about a rounding error of some description, then restarted prime95 and kept on going.

Currently doing an iteration with Aida64 instead, i want to get to a position where i can get through a single iteration without error before i set it on a longer multiple iteration run.

 

[TheSoupladel]

I think I might have a lemon of a chip, I stripped the setting back close to default and ran corecycler just for core 0 and 2 and both cores threw an error, I am giving it one more shot, this time completely at default, not even DOCP enabled to just rule that out but it looks like it might be going back.

The only thing I can think of is to re-seat the chip, repaste etc… and perhaps do a complete reset of CMOS as well

any other suggestions?

 

[TheSoupladel]

right, i have re-seated the chip, cleared cmos and repasted aaaaand i am still getting errors on core 0 and 2 with 100% stock settings, back it goes...

 

[TheSoupladel]

Thank you, i will give this a go, but i am reasonably confidient it is faulty, specifically just poor quality silicon.

I have done some reading up on the error and there has been some suggestion it can actually be caused by other elements of the system such as RAM, however in such instances the error is intermittant and not always the same core. The instability can be overcome by putting a +value against the curve optimise for that core and to that end i have put +5 against 0 and +10 against 2 and so far i think it has solved the problem for those two cores.

However, i am not happy accepting a brand new £700 CPU that in effect requires overvolting in order to operate normally because so far it has added quite substantially to the thermals for that CCD and has negatively impacted on the performance as well, so i think RMA is the only option for me just now really.

 

[TheSoupladel]

Increase LLC and see if it helps.

As above, i ahve had to put +10 against core 2 and +5 aginast core 0 and am running a full run of CoreCylcer/Prime95 to test the current set up but i should have to be overvolting a new CPU to get it to run correctly, especially given the already high voltage these Ryzen chips run with.

 

[TheSoupladel]

Ok, before you do that, download OCCT (get the latest beta) and do this



Change threads to advanced and on the window that pops up deselect all the physical cores apart from 0 and 2. Then run the stability test.

CoreCyler is basically just a "fancy script" that sets up Prime95 in a way which tests things per core. But it still shouldn't be erroring on cores like that with default settings.

Still, to rule out there being some sort of issue with it, you should run another stability testing app which can do per-core testing.

I have OCCT running currently on all cores just for completness and as its the demo it will only last for an hour, but if it does come across an error i presume it will tell me there and then or does it wait until the test is complete to report any errors?

 

[TheSoupladel]

Will tell you if there is an error. 1 hour should be enough to hit either both cores at once, or 1 each for an hour at a time.

that

Well it ran for an hour and said nothing, i have already arranged an RMA so will se wgat comes of that, they are collecting it tomorrow and should hopefully have the replacement by the weekend, have gone back to my 3900x in the meantime