Question 5700x and curve optimizer

[kurdtnz]

Ive run both the curve optimizer on Ryzen Master and the optimizer on the HYDRA software and they both set all my cores to a -30 offset, how reliable is this? Ive not had any issues gaming, running benchmarks etc. Also, I didnt know if this is a thing with Ryzen Master but it use to show your top 2 preferred cores but since using it again with this processer it hasnt highlighted any core with the little stars?

 

[kurdtnz]

Forgot to mention, I wanted to know the reliability of these results so I could use the offsets in my motherboards BIOS.

 

[kerberos_20]

ryzen master already edits your bios
hydra ignores bios and it ignores amd based overclocking tables, its software based
if you use hydra, then ryzen master is useless as it gets overwritten by hydra
that also answer your best two cpus question, amd provided tables are ignored with hydra

 

[kurdtnz]

ryzen master already edits your bios
hydra ignores bios and it ignores amd based overclocking tables, its software based
if you use hydra, then ryzen master is useless as it gets overwritten by hydra
that also answer your best two cpus question, amd provided tables are ignored with hydra

Thank you, I just wanted to make sure those offsets were reliable. I dont run RM and HYDRA together, just wished to see if they both came up with the same optimizations. (to be honest, I stopped using HYDRA a while ago as it makes my system very unstable.) I mentioned doing the offsets in the BIOS as I also wanted to see if I could increase the max boost frequency by +100 - +150 but the only thing im unsure about is what to set the 'pbo limit' to? some say just disable it?

 

[drea.drechsler]

Forgot to mention, I wanted to know the reliability of these results so I could use the offsets in my motherboards BIOS.

I'd personally be a bit skeptical of using -30 for all cores but maybe 7nm is mature enough to be that good now. You can look for a utility called CoreCycler to run a test on the CPU, one core at a time. It will look for stability on each core with an aggressively bursty workload using one of several common stress test utilities. For maximum assurance let it run 4 or 6 complete cycles through all cores. If a core fails then back off the setting for it 2 or 4 clicks and try again.

Once you get a pass on all cycles, best stability is guaranteed by backing off all settings by 1 or 2 clicks just to be safe. But if you like to be aggressive then go with it.

If you get a random crash while gaming (or anything else for that matter) look in the event log for a WHEA error coinciding with the time of the crash event, then in the details to find the thread the error occurred on. Back off the core that thread runs on 1 or 2 clicks and go back in gaming (or whatever).

This is overclocking/undervolting which can be a bit iterative like this if pushing to the edge of silicon quality.

I'd also suggest uninstalling RyzenMaster. If using an AMD GPU disable the RyzenMaster service if you want to overclock the GPU. That's because Radeon drivers install the RM service too and will mess up your CPU overclock settings considerably every time you set a GPU overclock.

 

[Unolocogringo]

CO is good but not always accurate. It appears to be more accurate on single chip processors. I had already did the manual testing of CO on my 5600x. Then I tried the software one.
Manual -29 software -30 which was not 100% stable. Core 5 is my problem core. It needs .5v more voltage than my other cores to be stable @4.65ghz.
Running Core Cycler and normal usage testing is the only way to check for 100% stability.
What might be stable for stress tests, may or may not be stable with varied usage simulated by CoreCycler
All silicon is different and behaves differently, . You have to determine what works for your processor.

 

[kerberos_20]

I'd personally be a bit skeptical of using -30 for all cores but maybe 7nm is mature enough to be that good now

im on zen 2 and have -0.1v on all cores, no issues here

 

[drea.drechsler]

im on zen 2 and have -0.1v on all cores, no issues here

Many people have been able to run even larger negative offsets on Zen 2 series processors since their release, although usually with reduced performance being a result. OP's situation is different since Curve Optimizer is tweaking the actual V/f curve used to govern the processor's boosting algorithm, which can be on a per-core basis. Only Zen 3 processors (on AM4) can do this, with PBO2.

-30 all-core is such an extreme CO setting because there's usually a very high negative offset from the factory for the gold star (best) cores. So they tend to crash pretty easily in gaming if pushed that far negative in CO. But then a 5700X isn't a 5800X for a reason, and possibly the 7nm node is just that good now.

 

[drea.drechsler]

One other thing: If using an AMD GPU, Radeon Drivers install the RM service too. So you have to disable the RyzenMaster service after every GPU driver update if you want to overclock the GPU. That's because it will mess with your CPU overclock settings considerably every time you set a GPU overclock.

 

[Roland Of Gilead]

Ive run both the curve optimizer on Ryzen Master and the optimizer on the HYDRA software and they both set all my cores to a -30 offset, how reliable is this? Ive not had any issues gaming, running benchmarks etc. Also, I didnt know if this is a thing with Ryzen Master but it use to show your top 2 preferred cores but since using it again with this processer it hasnt highlighted any core with the little stars?

Hey there,

I've found very similar results, with each of Hydra, Bios CO, and Ryzen Master. I also tried CTR 2.1, and oddly enough, it gives the best results for me (4.65ghz all core - 4,85ghz one to two cores)

I'd agree with above. -30 across all cores is possibly going to cause instability.

 

[Roland Of Gilead]

Thank you, I just wanted to make sure those offsets were reliable. I dont run RM and HYDRA together, just wished to see if they both came up with the same optimizations. (to be honest, I stopped using HYDRA a while ago as it makes my system very unstable.) I mentioned doing the offsets in the BIOS as I also wanted to see if I could increase the max boost frequency by +100 - +150 but the only thing im unsure about is what to set the 'pbo limit' to? some say just disable it?

Hey kurdtnz,

I had a similar issue with Hydra. Essentially I went to do the diagnostic, and yielded worse results than manual, or PBO/CO, and was unstable. The one thing I had changed during this process, was a bios update. When I reverted to the previous bios I was running, that instability disappeared. If you updated the bios whilst testing, it could be the new bios has shifted the voltage tolerances. At least that's what I found. Just thought I'd mention it. Sometimes new Bios revisions bring some good performance jumps, but in my case with AGESA 1.2.0.7 that's where my problems came from. I've settled on 1.2.0.6/b as it gives me best results.

 

[Unolocogringo]

All silicone is different.
5600x is 4.65ghz stock. 3.9-4.1 all core load. After CO 4.5-4.65ghz all core load and lower temps. Best core 1.22v worst core 1.38 all core load.
Anything above 4.65 ghz is unstable.

My 3600 does 4.4 all core boost @1.29v. static voltage.
Much better that the 1.4-1.5 voltage spikes in my usage scenario. 24/7/365 always loaded and runs cooler.
PBO uses too much voltage to be stable on all processors by default.

 

[drea.drechsler]

...
My 3600 does 4.4 all core boost @1.29v. static voltage.
Much better that the 1.4-1.5 voltage spikes in my usage scenario. 24/7/365 always loaded and runs cooler.
PBO uses too much voltage to be stable on all processors by default.

Often overlooked but very important: the harmful effects of voltage are cumulative. The average core voltage my 3700X sees is actually less than 1.29V even though it's using a fairly aggressive PBO setup. When running a heavy rendering workload it's much less as it's running a safer clock around 4100-4200Mhz. That's part of the fallacy of fixed overclocks being "better" for Zen 2 and Zen 3 processors (probably Zen 4 too).

 

[Unolocogringo]

I run folding with Nvidia cards, which uses a just in time compiler.
This makes the workload constantly jump from core to core and uses 1 core per video card if your processor is fast enough. So it constantly shifts between 2 cores.
All core voltages were higher than my static voltage on the 3600 which feeds 2 video cards so 4 cores constantly swapping.

Both processors still boost/sleep cores as normal , just at lower voltages achieved different ways.
My 5600x will not overclock at all. but by lowering voltage all core boost frequency went up .55 to .85 ghz on all core workloads AND runs cooler.
Just a little tinkering for better performance and cooler temps for 24/7/365 runtimes.
It takes 2 full cores to feed a GTX 1070 with a I5 6600 with 16 gig 3200 ram.
the 3700x is already a better die. that will boost higher with lower voltage compared to a 3600.

 

[kurdtnz]

I'd personally be a bit skeptical of using -30 for all cores but maybe 7nm is mature enough to be that good now. You can look for a utility called CoreCycler to run a test on the CPU, one core at a time. It will look for stability on each core with an aggressively bursty workload using one of several common stress test utilities. For maximum assurance let it run 4 or 6 complete cycles through all cores. If a core fails then back off the setting for it 2 or 4 clicks and try again.

Once you get a pass on all cycles, best stability is guaranteed by backing off all settings by 1 or 2 clicks just to be safe. But if you like to be aggressive then go with it.

If you get a random crash while gaming (or anything else for that matter) look in the event log for a WHEA error coinciding with the time of the crash event, then in the details to find the thread the error occurred on. Back off the core that thread runs on 1 or 2 clicks and go back in gaming (or whatever).

This is overclocking/undervolting which can be a bit iterative like this if pushing to the edge of silicon quality.

I'd also suggest uninstalling RyzenMaster. If using an AMD GPU disable the RyzenMaster service if you want to overclock the GPU. That's because Radeon drivers install the RM service too and will mess up your CPU overclock settings considerably every time you set a GPU overclock.

Ive not tried this utility yet but will at some point, thanks for the suggestion. Ive still been running an all core -30 offset and have had no issues at all yet and have been gaming pretty hard with it, to be honest, have been really impressed with this 5700x (over my 5600x) The only reason I got it was because it was really cheap at the time and Ive swapped out my son's 3600 for the 5600x opportunity knocks and all that. Owing the fact the 5700x is also a 65w cpu too was a bonus. Im not a big overclocker, just wanted to optimize the system I have and to see what the 5700x could do. Thanks everyone!

 

[drea.drechsler]

.... Owing the fact the 5700x is also a 65w cpu too was a bonus. ....

That could also be part of what leaves it stable. If AMD's using the same die that's capable of fitting a 105W TDP 5800X but limiting it to a 65W TDP they're not stressing nearly hard enough to make it squeak out of the box. So there is a lot of margin to dial back the CO settings.

If using CO and PBO, you could also adjust PPT, TDC and EDC. Try using the same settings that a 5800X does (142W PPT, 95A TDC and 140A EDC) to see if it picks up performance. You probably won't see higher boost clocks but it may hold a higher average clock in a rendering task like Cinebench, which is where the real money is, if cooling is good enough.

 

[John Chesterfield]

That could also be part of what leaves it stable. If AMD's using the same die that's capable of fitting a 105W TDP 5800X but limiting it to a 65W TDP they're not stressing nearly hard enough to make it squeak out of the box. So there is a lot of margin to dial back the CO settings.

If using CO and PBO, you could also adjust PPT, TDC and EDC. Try using the same settings that a 5800X does (142W PPT, 95A TDC and 140A EDC) to see if it picks up performance. You probably won't see higher boost clocks but it may hold a higher average clock in a rendering task like Cinebench, which is where the real money is, if cooling is good enough.

All of this is good advice. It makes sense that a 5700x would have more headroom due to be tuned to do 65w, the 5700x is where the smart money goes in terms of performance per dollar.

And as also said above, all silicon is different. This 5900x went straight to -30 on all cores using CO, with a few tweaks to PPT, TDC and EDC it stays around 72 Celsius doing 4.5Ghz all core Cinebench R23. That's my happy medium of performance vs temps.