Overclock.net banner

Status
Not open for further replies.
321 - 340 of 679 Posts

·
Premium Member
Joined
·
2,704 Posts
Discussion Starter #321
So what you're saying is, VID does nothing once you're in manual OC via the bios? I'm worried because I'm trying out the Per-CCX utility and it does seem to adjust my VID. And VID keeps jumping round. Again just to be sure, in manual mode, VID is ignored even though it jumps around?

I attached a screenshot with some graphs open. The first run is just booted up with 1.275 VID and 1.325 vCore set in the BIOS. Then I set 4000MHz on CCX0 and somehow this jumps the VID to 1.488, I run Cinebench and it drops to 1.4'ish. The other two voltages (vCore and SVI2) stay the same. Then in Ryzen master, I put the voltage back on 1.275, and again only VID changes.



It just seems so counter-intuitive that the voltage setting in ryzen master when doing manual OC would adjust VID if that didn't do something, so in that regard seeing 1.40v VID under full load does worry me.
Where can I find this tool?
 

·
Registered
Joined
·
94 Posts
Don't touch VID. The CPU will work out what VID is required`- you just set VCore. Most accurate measurement of that is the CPU Core SVI2 TFN in HWInfo.
Yeah but then why does it change VID when applying a per-CCX OC to 1.48, with VID settling at 1.40, all that while SIV2 TFN stays completely unchanged? That tells me that either CPU VID does nothing or SIV2 TFN is not actually a measure for the voltage on the core. They cannot both be true, can they?

For the record, I'm not actively changing VID or intending to do so when pressing that button on the Per-CCX OC tool. It just happens.

Where can I find this tool?
I found it in der8auer's video description of this one
. The direct link to the tool is http://bit.ly/30zhbMz
 

·
Registered
Joined
·
94 Posts
[doublepost, sorry]

Well I'll edit this then, from a pretty sketchy test it seems VID indeed does nothing in manual OC mode. When doing Small FFT AVX on this 3.8GHz setting, temps instantly jump to 94-95C, so I quit the test immediately. That was on VID 1.275. Setting VID to 1.20 and 1.10 (where it drops to 1.00) does not scale at all with either power usage or temperature (confirming that the power usage number is correct).

Edit2// Confirmed that least lower VID and/or Ryzen Master voltage setting in general does nothing when on Manual OC in the bios. Setting VID to 1.1v still allows me to get CCX1 to 4.5GHz.

Why setting VID is even a thing (either in BIOS or Ryzen Master) is beyond me.
 

·
Premium Member
Joined
·
2,704 Posts
Discussion Starter #324
In short: Unless you are certain that you are using manual voltage mode (VRM controller, from the bios) do not use the application ("worktool").

It is a CPU killer if the VRM controller is running in offset-mode.

I tried it with the voltage mode set to offset (or "Auto") and setting 4200MHz for CCX0/1 and 4050MHz for CCX2/3 resulted in 1.405V actual Vout during Cinebench R20.

The voltages during the default OC-Mode remain sane, regardless of the VRM controller operating mode.
This hack however causes the voltage governor to go insane.

I have no idea why someone thought it would be a good idea to release this to the public...
 

·
Registered
Joined
·
248 Posts
Interesting Scaler Boost on my 3800x

Not surprisingly I am seeing higher vCore boosts on average with the PBO Scaler at 3x.
 

Attachments

·
Meddling user
Joined
·
7,423 Posts
@crakej

So last night I did benches and was happy with them. Rerun profile for length testing again and all well.

This morning before leaving home I setup the lowered SOC/VDDG profile, nothing else was changed. Now I benched it and was horrified as I lost performance, even though stability was not lost.

I then went back to initially used SOC/VDDG, performance was back. Lowered SOC/VDDG and gone again.

Last nights benches, room ambient ~20C.


Reload profile and test, room ambient ~26C.


Reload lowered SOC/VDDG profile, room ambient ~26C.

 

·
Registered
Joined
·
2,194 Posts
In short: Unless you are certain that you are using manual voltage mode (VRM controller, from the bios) do not use the application ("worktool").

It is a CPU killer if the VRM controller is running in offset-mode.

I tried it with the voltage mode set to offset (or "Auto") and setting 4200MHz for CCX0/1 and 4050MHz for CCX2/3 resulted in 1.405V actual Vout during Cinebench R20.

The voltages during the default OC-Mode remain sane, regardless of the VRM controller operating mode.
This hack however causes the voltage governor to go insane.

I have no idea why someone thought it would be a good idea to release this to the public...
Because they knew you would save everyone? Classic super hero lure?
 

·
Registered
Joined
·
1,639 Posts
@crakej

So last night I did benches and was happy with them. Rerun profile for length testing again and all well.

This morning before leaving home I setup the lowered SOC/VDDG profile, nothing else was changed. Now I benched it and was horrified as I lost performance, even though stability was not lost.

I then went back to initially used SOC/VDDG, performance was back. Lowered SOC/VDDG and gone again.

Last nights benches, room ambient ~20C.


Reload profile and test, room ambient ~26C.


Reload lowered SOC/VDDG profile, room ambient ~26C.


That's what I said and noticed myself. System can be stable BUT you are loosing performance so I think best would be to run benchmarks and find at what voltage one maintains performance and going up does not give more but going down cuts it. What You think ??
 

·
Registered
Joined
·
1,639 Posts
In short: Unless you are certain that you are using manual voltage mode (VRM controller, from the bios) do not use the application ("worktool").

It is a CPU killer if the VRM controller is running in offset-mode.

I tried it with the voltage mode set to offset (or "Auto") and setting 4200MHz for CCX0/1 and 4050MHz for CCX2/3 resulted in 1.405V actual Vout during Cinebench R20.

The voltages during the default OC-Mode remain sane, regardless of the VRM controller operating mode.
This hack however causes the voltage governor to go insane.

I have no idea why someone thought it would be a good idea to release this to the public...
I'w noticed same thing and its pita to use anyhow.
One question I was playing with Ryzen master and changing mhz per core in 1 ccx. Noticed something weird. If I run lets say core 1 4350 core 2 4400 core 3 4325.... The slowest set core gets dropped to 3500 not sure if its a bug or what. Any idea ??

Would be great if Bios had option to overclock/ccx atm heh.
 

·
Meddling user
Joined
·
7,423 Posts
That's what I said and noticed myself. System can be stable BUT you are loosing performance so I think best would be to run benchmarks and find at what voltage one maintains performance and going up does not give more but going down cuts it. What You think ??
I agree :) .

I'm gonna stay at SOC: 1.062 VDDG: 1.013, instead of 1.037/0.986, just rerun AIDA64 again once and back where I should be.


I'm better off spending the time lowering some subtimings, going for SCL 3 (if it stable/perform), then finding out what value between the ~25mV difference of SOC/VDDG is break point for performance.
 

·
Premium Member
Joined
·
2,704 Posts
Discussion Starter #331
I'w noticed same thing and its pita to use anyhow.
One question I was playing with Ryzen master and changing mhz per core in 1 ccx. Noticed something weird. If I run lets say core 1 4350 core 2 4400 core 3 4325.... The slowest set core gets dropped to 3500 not sure if its a bug or what. Any idea ??

Would be great if Bios had option to overclock/ccx atm heh.
Yeah, ideally this would be implemented in the bios, given it can be done without messing the voltages.

Setting different frequencies to the cores belonging to a same CCX will mess up the frequencies.
The L3 is shared between the cores of the same CCX and because of that there are limitations. So set the frequency per CCX, in case you find the need to control them separately.
 

·
Registered
Joined
·
108 Posts
I'w noticed same thing and its pita to use anyhow.
One question I was playing with Ryzen master and changing mhz per core in 1 ccx. Noticed something weird. If I run lets say core 1 4350 core 2 4400 core 3 4325.... The slowest set core gets dropped to 3500 not sure if its a bug or what. Any idea ??

Would be great if Bios had option to overclock/ccx atm heh.
Cores on same ccx share same FID.
DID is usually 2 default.
So 4000mhz=8000/2
Did in 0.5 steps
So if 1 core is 4000, other one either (8000/2.5) or 8000/1.5 or synch 8000/2
FID decided by highest clocked core, and often it seems when there's a change to one core, the other cores drop a step as a fail safe first since the user has not requested the other cores in the same ccx to switch.
So the step diff is a bit too big within a ccx
 

·
Computer Enthusiast
Joined
·
2,850 Posts
Has no one else had problem of fastest cores not being used?
 

·
Registered
Joined
·
1,308 Posts
@crakej

So last night I did benches and was happy with them. Rerun profile for length testing again and all well.

This morning before leaving home I setup the lowered SOC/VDDG profile, nothing else was changed. Now I benched it and was horrified as I lost performance, even though stability was not lost.

I then went back to initially used SOC/VDDG, performance was back. Lowered SOC/VDDG and gone again.

Last nights benches, room ambient ~20C.


Reload profile and test, room ambient ~26C.


Reload lowered SOC/VDDG profile, room ambient ~26C.

Glad you mentioned this, I was running some real low SoC voltage and VDDG voltages just a moment ago and still am to a extent as I test out what works and doesn't right now.
I see this might mainly be a SoC voltage thing. I can seemingly run real low VDDG voltages without issue. Started to test even lower ones to rule it out.
Seems my SoC can be lowest ~1.050V and run Prime95 without issue. But VDDG can be real low.
Been ok with 900mv... I was trying to test 850mv but seems it was ignored and set 900 instead. I'll have to recheck that.

EDIT: got it correct now, now to just test 850mv VDDG.
 

·
Registered
Joined
·
94 Posts
In short: Unless you are certain that you are using manual voltage mode (VRM controller, from the bios) do not use the application ("worktool").

It is a CPU killer if the VRM controller is running in offset-mode.

I tried it with the voltage mode set to offset (or "Auto") and setting 4200MHz for CCX0/1 and 4050MHz for CCX2/3 resulted in 1.405V actual Vout during Cinebench R20.

The voltages during the default OC-Mode remain sane, regardless of the VRM controller operating mode.
This hack however causes the voltage governor to go insane.

I have no idea why someone thought it would be a good idea to release this to the public...
Thanks for the due diligence and the fair warning. I guess it's a tool for experienced overclockers, probably it's not the best idea that he put it in the video description of his popular channel where less experienced overclockers will frequent.
 

·
Computer Enthusiast
Joined
·
2,850 Posts
Which workload?
Cinebench R15, R20 and at least X265 automatically control the affinity in runtime.
Might not be optimal for CPPC preferred cores.
It was mixed workload, included a few mins of resolve rendering a short scene, loaded up lots of other stuff at same time to see what it would do. Yes, CB15 and 20 as well.

Can you recommend anything that might better demonstrate if the CPU is behaving as expected? Even in the mixed workload, if I watch it in RM, the fastest cores certainly look like they're being underutilized.

About to spend another couple of hours chasing 3800MTs. I was up really late as just couldn't put it down! Didn't want to sleep until I made that breakthrough!
 

·
Registered
Joined
·
9 Posts
3800MHz CL16 GDME seems to be beating out your setup, I think some of the subtimings adjusted may gain you some more.

That's weird really, you only got twr and trtp tighter, I'm wondering what's wrong here or where to improve, subtimings are pretty much as tight as they get, tighter would be unstable. Does ddr4 perform a bit worse in terms of bandwidth when a bit too warm? Both my sticks are around 50C but were still stable for 18k % karhu.

I even got 2 subtimings tighter than you, tfaw, trrdl and 3 primaries tighter.

Not sure where to go from here.
 

·
Manual breath enabled
Joined
·
1,261 Posts
Got 1900Mhz ram/IF 1:1 to work.
Chaising 60.xxx in readspeed xD

Still the boost thing is still not really working.
While ramstress or cinebench the CPU voltage wont go under 1,45V. Dont know if this is ok for the long run.

Wich PBO/XFR settings are ok?
Have PBO Disabled
XFR->PBO enabled
 

Attachments

·
Registered
Joined
·
1,639 Posts
Cores on same ccx share same FID.
DID is usually 2 default.
So 4000mhz=8000/2
Did in 0.5 steps
So if 1 core is 4000, other one either (8000/2.5) or 8000/1.5 or synch 8000/2
FID decided by highest clocked core, and often it seems when there's a change to one core, the other cores drop a step as a fail safe first since the user has not requested the other cores in the same ccx to switch.
So the step diff is a bit too big within a ccx
That makes it all clear. Iw spwnd whole friday evening testing core by core what they can do. And as alwaya found weak links. So at least i know what ccxs worst core can work at l. Just settinf by ccx in bios would be great hehe
 
321 - 340 of 679 Posts
Status
Not open for further replies.
Top