5800X Boost set to 5050Mhz all cores with AMD Curve Optimizer

[BigBANGtheory]

So far I'm at -15 all core on curve optimiser, +100mhz. Single cores only hit 1.46v so might try the scalar and more mhz to see if I can squeeze some more out of it.
Cinebench results seem okay ish, 6046 multi and 646 single so far on day to day settings.

I can run Cinebench fine with CO settings and +100, +150 or even +200 but what I see in practice is no real improvements (except temps) and random crashes even at idle not under load. Maybe my chip is close to its limits already your millage may vary of course

 

[KungFuSpaghetti]

I can run Cinebench fine with CO settings and +100, +150 or even +200 but what I see in practice is no real improvements (except temps) and random crashes even at idle not under load. Maybe my chip is close to its limits already your millage may vary of course

Fortunately no crashes, idle or otherwise in 16 hours use yesterday (mix of work and gaming) and temps don't exceed 72 degrees under stress. Bit hotter than I'm used to but reading up these things run hot! The idle crashes you saw are probably LLC related as Ryzens don't have a real idle, they constantly have bursts and small jumps in temps and clocks so most likely that. I think that will be bios implementation and likely fixed in later release.

I'm actually blown away by the quality of Asrock's bios implementation, which is something I've been pretty critical of them in the past for. This is the best bios I've used on this board, previous ones either had poor memory overclock but great CPU boost or vice versa... this one seems the finished article. I think a lot of people's success is down to bios and not the chips from what I've read. Built a 5600x based system for a friend and was very impressed with MSI bios, so much more user friendly and easier to use and navigate compared to mine, was considering changing if the bios implementation was garbage again... thankfully asrock surpassed expectations!

 

[JohnMatrix]

This might be obvious



is that -25 with a 0mhz offset?

Yes

PBO advanced
PBO limits disabled
PBO scaler Auto

 

[MegaPinches]

So did you guys come to conclusion that lowering PPT limit scales down voltage as well? otherwise I cant see how PPT limits effects anything untill it reaches point where it starts to be maxed out.

Any insights from those who experimented?

 

[KungFuSpaghetti]

I've not tried reducing PPT yet but hwinfo isn't showing total package power correctly... unless it really is allowing it to pull nearly 160w? That's with scalar set to x10, hits 86 degrees on 360mm aio... auto seems the same speed but maxed out to only 72 degrees. Performance difference is negligible too, 658 it pulled single core on cb20, 246 auto but much lower temps, feels like laws of diminishing returns now.

 

[BigBANGtheory]

So did you guys come to conclusion that lowering PPT limit scales down voltage as well? otherwise I cant see how PPT limits effects anything untill it reaches point where it starts to be maxed out.

Any insights from those who experimented?

It does yes from my experience, but the issue is still stability. I think part of the problem is opportunistic boosting where core's you've lowered the curve on still seem to accept a high boost frequency but can't handle it.

 

[Dreamic]

.

 

[Guest2]

So did you guys come to conclusion that lowering PPT limit scales down voltage as well? otherwise I cant see how PPT limits effects anything untill it reaches point where it starts to be maxed out.

Any insights from those who experimented?

It certainly does for me
Using the following at the moment and not seeing more than 77c on heavy benching (AIDA64 stability testing CPU/FPU/Cache)
AIDA64 really batters the CPU, a lot more than multiple runs of Cinebench R20 or R23. Cooling is a Noctua U14-S.

125 W PPT
85 A TDC
125 A EDC

Max CPU boost clock override - 150MHz
Scaler - Auto
Curve - -20 all cores, except -5 on best two.
SoC - 1.1750V
VDDP - 0.950V
VDDG CCD - 0.950V
VDDG IOD - 1.125V
NB LLC - Mode 1
NB Switching frequency - 1000KHz

Left at PBO default and no overclocking temps reach upto 90c in AIDA64

 

[KungFuSpaghetti]

So did you guys come to conclusion that lowering PPT limit scales down voltage as well? otherwise I cant see how PPT limits effects anything untill it reaches point where it starts to be maxed out.

Any insights from those who experimented?

Finally had a chance to mess with PPT limits, did as per post above, same exact settings but on my Asrock taichi, but for new performance regressed. Left alone with no limits, 6107 on cb20, then reboot with PPT limits and 6001.

Currently messing with the curve optimiser setup per core, see if it nets ant extra performance.

 

[Guest2]

Finally had a chance to mess with PPT limits, did as per post above, same exact settings but on my Asrock taichi, but for new performance regressed. Left alone with no limits, 6107 on cb20, then reboot with PPT limits and 6001.

Currently messing with the curve optimiser setup per core, see if it nets ant extra performance.

I wouldn’t use my settings. I had a random crash the other night in Warzone.
Limiting power isnt great if the power is needed but isnt there.
I’ve gone back to PBO auto and just using my RAM and IF overclock

 

[KungFuSpaghetti]

I didn't have any stability issues, but it literally was one run on cb20 multi and single test and back to auto. Temp stayed the same strangely but performance was worse, but the asrock bios can be weird.

 

[alec]

Does curve optimiser work on Zen 2?
What exactly does it do?

Haven't heard it working for Zen 2
What exactly it does nobody knows.

There is a frequency/voltage curve factory hardcoded for every core in cpu.
Curve optimizer shifts that curve, so that for any given frequency core would ask for less (or more if shift is positive) voltage.
As result of negative shift, cores are more likely to boost higher. Essentially same voltage as before but higher frequency.

 

[Towel]

hi all, made an account after reading this thread as its a gold mind however a very scattered one

just finished building a new pc with a 5800x, 4000mhz CL16 Ripjaws with a 360mm EK AIO on the x570 tomohawk.

tried to use XMP couldn't post so after a quick google found 4000mhz isn't stable, have gone with 3800mhz with 1900 fclock.
booted up and have also seen 5MHz which is amazing coming from a 3900x.

from reading this is clear the 5800x is a hot chip but, my idle temps are 45-50, cinebench 1st run 85, 2nd 89, 3rd 90, these seem really hot.

Interested in getting some performance and lower temps (unless I'm mistaken).

Whats the start point? Undervolt, overide, eco-mode, curve + more mhz... so many diffrent things in the thread im unsure where to start

cheers for helping a bot like me.

 

[KungFuSpaghetti]

hi all, made an account after reading this thread as its a gold mind however a very scattered one

just finished building a new pc with a 5800x, 4000mhz CL16 Ripjaws with a 360mm EK AIO on the x570 tomohawk.

tried to use XMP couldn't post so after a quick google found 4000mhz isn't stable, have gone with 3800mhz with 1900 fclock.
booted up and have also seen 5MHz which is amazing coming from a 3900x.

from reading this is clear the 5800x is a hot chip but, my idle temps are 45-50, cinebench 1st run 85, 2nd 89, 3rd 90, these seem really hot.

Interested in getting some performance and lower temps (unless I'm mistaken).

Whats the start point? Undervolt, overide, eco-mode, curve + more mhz... so many diffrent things in the thread im unsure where to start

cheers for helping a bot like me.

You have a top end AIO there, so you'll definitely be able to get the temps down a little, whilst also getting more performance (assuming case airflow is decent and it's not an oven!).

Worth pointing out too that each mobo manufacturer seems to have wildly different implementations of the AGESA code so what works on one system may not work on another.

In the BIOS I have PBO set to advanced
PBO Limits disabled
Scalar AUTO (tried it with x1 all the way to x10, just seemed to make it run a hotter)
Curve Optimiser set to individual cores with negative offset - I would start out using all core on this, try -15 as a starting point.
Boost I would go straight in at +150MHZ personally.

If that's all stable and no WHEA event errors, blue screens or random reboots then maybe look at your core quality order in HWINFO and then apply negative curve optimisations to each core manually. I have my worst cores at -25 and my two best at -15. On my board this appears to be the sweet spot. If you do get the odd random reboot this appears to be down to LLC when Ryzen has a short burst of volts and a sharp temp increase in lightly threaded workloads or background tasks. I've not experienced this myself, but it's been discussed here and other threads.

You can also lower PPT and EDC values to further drop temps. I had a play about with it, but sadly temps stay the same on my system whilst performance inexplicably drops. I think this is more down to ASRock's bios though.

 

[Towel]

You have a top end AIO there, so you'll definitely be able to get the temps down a little, whilst also getting more performance (assuming case airflow is decent and it's not an oven!).

Worth pointing out too that each mobo manufacturer seems to have wildly different implementations of the AGESA code so what works on one system may not work on another.

In the BIOS I have PBO set to advanced
PBO Limits disabled
Scalar AUTO (tried it with x1 all the way to x10, just seemed to make it run a hotter)
Curve Optimiser set to individual cores with negative offset - I would start out using all core on this, try -15 as a starting point.
Boost I would go straight in at +150MHZ personally.

If that's all stable and no WHEA event errors, blue screens or random reboots then maybe look at your core quality order in HWINFO and then apply negative curve optimisations to each core manually. I have my worst cores at -25 and my two best at -15. On my board this appears to be the sweet spot. If you do get the odd random reboot this appears to be down to LLC when Ryzen has a short burst of volts and a sharp temp increase in lightly threaded workloads or background tasks. I've not experienced this myself, but it's been discussed here and other threads.

You can also lower PPT and EDC values to further drop temps. I had a play about with it, but sadly temps stay the same on my system whilst performance inexplicably drops. I think this is more down to ASRock's bios though.

Yeah hence why when i booted i was shocked with the temps,

i just followed this video as a base Ryzen 5000 Undervolting with PBO2 – Absolutely Worth Doing - YouTube

100% enabling PBO with limits on AUTO was the issue, the limit was 500w and 220m EDC, which explains a lot.

I will try individually changing curves, how do you check core quality order in HWINFO?


would change the LLC work when playing with the curves? I imagine they would work against each other

Using the default PBO limits dropped cinebench temps down to 80

 

[KungFuSpaghetti]

Yeah hence why when i booted i was shocked with the temps,

i just followed this video as a base Ryzen 5000 Undervolting with PBO2 – Absolutely Worth Doing - YouTube

100% enabling PBO with limits on AUTO was the issue, the limit was 500w and 220m EDC, which explains a lot.

I will try individually changing curves, how do you check core quality order in HWINFO?


would change the LLC work when playing with the curves? I imagine they would work against each other

Using the default PBO limits dropped cinebench temps down to 80

That vid is great, that was my starting point, plus seeing what the likes of @gerardfraser had done with his as well as others.

The core quality will show in Ryzen master too, but I dont bother with that or even have it installed. In HWINFO your cores are ranked in quality, eg: Core 1 Clock (perf #1/1) is my best core, then core 4 is (perf #1/2) etc. I would start out with an all core negative curve though, then go from there. Should gain multi and single core performance over stock as well as lower temps. I max out in gaming around 65 and 72 seems to be the high point after a slew of CB20 runs.

I dont know for sure as I've not had the issue with the random restarts, but people have observed this on here and other sites and seems to be when the system is idle. My assumption is that it's the short bursts you get for background tasks are causing rapid voltage and temp increases before quickly dropping. I dont think this would or should work against the curve optimiser though, but I leave my LLC on auto at all times, I've never had the need to change it so far. If I do encounter reboots randomly though, this would be the first thing I'd be looking to alter.

 

[heartburnron]

I'm in the same boat regarding temps hitting 90 at full load with everything at stock. I've been playing around with the curve optimizer and PBO limits and whilst I'm not yet satisfied I've reached the sweet spot or fully understand everything, what I can say for certain is that the single biggest impact I get on temps is lowering the PPT limit in the PBO settings. The max multi-core boost clock also increases and decreases in line with adjustment to the PPT so I guess you need to find the balance of the two you're happy with. For now, I've left the PBO settings at 127, 95, 125, -20 curve on all cores except the best two (-5). This gives me a multi-core boost up to 4550 Mhz and max temps of 82.

 

[Joxeon]

Yeah hence why when i booted i was shocked with the temps,

i just followed this video as a base Ryzen 5000 Undervolting with PBO2 – Absolutely Worth Doing - YouTube

100% enabling PBO with limits on AUTO was the issue, the limit was 500w and 220m EDC, which explains a lot.

I will try individually changing curves, how do you check core quality order in HWINFO?


would change the LLC work when playing with the curves? I imagine they would work against each other

Using the default PBO limits dropped cinebench temps down to 80

Try setting manual limits with PPT125 EDC125 TDC88 and then see how much negative curve you can run -10~15~20etc as this should match or exceed stock performance while bringing temps down by another 10C.

I've left the PBO settings at 127, 95, 125, -20 curve on all cores except the best two (-5). This gives me a multi-core boost up to 4550 Mhz and max temps of 82.

From my experience doing a per core curve just seems to default the all core boost to the lowest value so say you set all cores to -30 then got a 4700 all core boost then go back in and change one core to -5 then all core will only boost to 4550 etc, not sure if this is a bug with it but would be nice to see individual cores scaling with different frequencies depending on how much offset they have.

 

[heartburnron]

Try setting manual limits with PPT125 EDC125 TDC88 and then see how much negative curve you can run -10~15~20etc as this should match or exceed stock performance while bringing temps down by another 10C.



From my experience doing a per core curve just seems to default the all core boost to the lowest value so say you set all cores to -30 then got a 4700 all core boost then go back in and change one core to -5 then all core will only boost to 4550 etc, not sure if this is a bug with it but would be nice to see individual cores scaling with different frequencies depending on how much offset they have.

Agreed, that is the behaviour I think we're all expecting to see on the individual cores, one to keep an eye on for sure. I've been focusing on trying to have more reasonable temps without compromising significant performance as apposed to gaining it. I still haven't been able to break 15000 on CB R23 though regardless. I'll give your settings a go and see what happens...

UPDATE: Just set all cores to -20 rather than having -5 on the two best and my clocks jumped to 4650 benching so same behaviour @Joxeon has seen. My R23 score dropped 300+ points though so not necessarily better in terms of performance the higher clocks...

 

[Towel]

so far ive got +150mhz and -15 curve on all with default PPT, TDC, EDC

this gets me 4650 mhz in cinbench r20 at around 80-82 degrees which is much better, i will try the lower PPT now