Overclock.net banner

81 - 100 of 359 Posts

·
Always Learning
Joined
·
1,194 Posts
Quote:
Originally Posted by Johan45 View Post

OK then, 1700 @ 4.0 DMM 1.52v P95/Aida again 20 minutes each ram is the same as 1700x
YD1700BBM88AE
9GU3920N70533
UA 1705 SUT
China with the "D" marking the 1700x had it as well
What is your bios voltage setting on? Wondering how large the discrepancy is between bios voltage (or offset if you're using that) and DMM. Also whether or not you're using LLC levels.
 

·
Meddling user
Joined
·
7,423 Posts
Discussion Starter #83
@Johan45

Many thanks
smile.gif
.

@finalheaven

On recommendation from @mus1mus @SuperZan @josephimports

We have Y-Cruncher.

Same room temp as x264 / RB stress mode (~22°C), this is ~5°C hotter for same OC, cooled by TR Archon SB-E X2.

Start



Just before I end



All mentioned have been using it and like x264 this is harder on CPU than RB, see josephimports post in owners thread, link.

What I can't believe is how quiet a fan profile I can maintain on the R7 [email protected] vs my i5 [email protected] Yeah large clock difference, but it's 8C/16T vs 4C/4T.
 

·
Always Learning
Joined
·
1,194 Posts
@gupsterg

Hmm... what does it mean then that I can pass 10 loops of x264 easily, but fail one hour of RealBench? Because on my system, I have to raise volts to pass realbench compared to x264. Are the people you mentioned using high speed ram? I am testing at 3200 14-14-14-34.
 

·
Premium Member
Joined
·
2,738 Posts
Using LLC at least on ASUS C6H is definitely not required or recommended.

Even at "Auto" setting there is no droop what so ever, no matter how hard you push the CPU. Infact generally the actual voltage output is 10-20mV (depending on the specimen) higher than the actual set point. Voltage overshoot hurts just as badly as undershoot (droop). The whole point is to minimize the variation between the different load conditions. If you need to increase the LLC to achieve stability, then your voltage set point is incorrect (insufficient) to begin with.
 

·
Premium Member
Joined
·
2,738 Posts
For stability testing I would recommend Prime95 28.10 (which generally isn't working properly on Ryzen).
With specific settings it is possible to make it stress the CPU properly: Select Custom -> Set the minimum and the maximum FFT size to 128 -> Select run FFTs in place.
 

·
Stock *ahem*
Joined
·
6,382 Posts
I heard AMD has for a long time implemented a different form of vDroop than Intel does.
 

·
Premium Member
Joined
·
2,738 Posts
Quote:
Originally Posted by Quantum Reality View Post

I heard AMD has for a long time implemented a different form of vDroop than Intel does.
The significantly tighter droop specification applies only in cases where the dLDO is used. dLDOs are not used for B1 stepping consumer parts, so the load-line specification does not apply. Infact the AM4 load-line specification is looser than it was for AM3+ for example (1.425mOhm vs. 1.3mOhm).
 

·
Always Learning
Joined
·
1,194 Posts
Quote:
Originally Posted by The Stilt View Post

Using LLC at least on ASUS C6H is definitely not required or recommended.

Even at "Auto" setting there is no droop what so ever, no matter how hard you push the CPU. Infact generally the actual voltage output is 10-20mV (depending on the specimen) higher than the actual set point. Voltage overshoot hurts just as badly as undershoot (droop). The whole point is to minimize the variation between the different load conditions. If you need to increase the LLC to achieve stability, then your voltage set point is incorrect (insufficient) to begin with.
My Vdroop according to my settings in bios and HWiNFO is more like 40-45mV. I guess I can live with that still.
 

·
Registered
Joined
·
481 Posts
I still stand by my belief that If you are using power saving features of the CPU, you should not be using LLC. LLC is good for setting records where the chip needs to run at max freq all the time. Not so good for the life of the CPU, and in my experience, causes instability under idle tasks. I was able to get my 3930K running 24/7 prime95 with high levels of LLC at 4.8GHz. I cannot run that for CPU life, so settle with 4.5 without LLC and using power saving features with a much lower max cpu voltage.

I have my 1700 in now, still waiting on my board. Only going to OC using P states, no LLC.
 

·
Premium Member
Joined
·
2,738 Posts
Quote:
Originally Posted by finalheaven View Post

My Vdroop according to my settings in bios and HWiNFO is more like 40-45mV. I guess I can live with that still.
Monitoring isn't generally very accurate.
You may see heavy undershoot in software monitoring, while measuring the actual voltage shows some overshoot
wink.gif
 

·
A fish with teeth!
Joined
·
1,076 Posts
Where do you guys measure the Vcore with a DMM? Is it practical to do while installed in a case like this, or only while it's outside of a case, say sitting on cardboard or some non-conductive surface?



If I have shaky hands and would be scared of shorting something do I need special tiny alligator clips?

EDIT: nvm...My meter is no good for this sort of thing. Only 0.1V DC resolution.
 

·
Always Learning
Joined
·
1,194 Posts
Quote:
Originally Posted by gupsterg View Post

@finalheaven

On recommendation from @mus1mus @SuperZan @josephimports

We have Y-Cruncher.

Same room temp as x264 / RB stress mode (~22°C), this is ~5°C hotter for same OC, cooled by TR Archon SB-E X2.

Start



Just before I end



All mentioned have been using it and like x264 this is harder on CPU than RB, see josephimports post in owners thread, link
I'm stable at +0.1875 offset (1.375v) in both Y-Cruncher (5-iterations) and 10 loops of x264. Betting it could go a lot longer.

However, not stable with 1 hour of RealBench unless i am +0.21875 offset (1.406v).

Is this why Silicon Lottery uses RealBench to test chips for binning them into categories?
 

·
Meddling user
Joined
·
7,423 Posts
Discussion Starter #94
@The Stilt

+rep and thank you for your posts
thumb.gif
.

@S1L3N7D3A7H

+1 on not using LLC
thumb.gif
.

I've settled for 3.8GHz ACB @ ~1.330V on DMM when CPU underload, this isn't too bad considering stock is 3.2GHz ACB ~1.089V. So +18.8% CPU clock gain for +22.1% VCORE gain. 3.9GHz was crazy IMO, +21.9% CPU clock gain for +35% VCORE .

Temps on 3.8GHz are at least 10°C lower than 3.9GHz (~60°C vs ~70°C for same tests). Just amazed how quiet a fan profile I can keep on Ryzen vs it [email protected] when stability testing. For gaming Ryzen vs i5 4690K = same quietness.

@finalheaven

It means your particular CPU sample destabilises OC quicker with RealBench. For me it is opposite to you, x264 showed instability quicker vs RealBench. If you wish to speed up OC profiling, use what destabilises your CPU 1st as stability test, then other tests will fall into place.

Perhaps the people mentioned are using higher RAM, but maybe not. I would check their pasts posts, SuperZan and Joesphimports are in the DB in my sig and have not advised of any updates.

No idea why Silicon Lottery only use RealBench. Like I said before I don't believe 1 stability test qualifies as covering all aspects for me.

The way I tested all OCs is not only stability tests that max clocks/usage but ones where there is varying load as well, as in normal usage varying load is more like how I use my PC.

The stability test that keeps CPU at MAX usage tells you is the that upper clock is fine. As we can see when we change PState 0 on our R7 1700 that idle clock has changed within the value shown in PState 2. This to me would mean even other states between idle and ACB have shifted. So by using a varying load stability test we will in effect check those for stability.

IMO the varying load stability test will play more havoc with "LLC effect", as the CPU keeps changing clocks/states the voltage is changing, therefore voltage changes occur more often and there is more "LLC effect" being tested.

I have sometimes read posts where a member says"I can pass 12hrs P95" and when I do x lighter usage case PC crash. To me that just means they did not do enough different cases of stability testing or are just "gunning" for too high an OC for daily use.
 

·
Registered
Joined
·
1,367 Posts
Okey guys, round 1 of my test. Did only test with 4000mhz and Stock settings for now.

Cpu: 1800x
Memory: G Skill - F4-3200C14D-16GTZR

Round 1:

  • 4000Mhz
  • Pstates0
  • Vcore Offset + 0.125
  • 3200mhz Dram
  • Stable
  • LLC Level 3
  • On Load - Software read: 1.395v - 1.417v // DMM read: 1.417v - 1.423v

  • 4000Mhz
  • Pstates0
  • Vcore Offset + 0.125
  • 3200mhz Dram
  • No Stable
  • LLC Auto
  • On Load - Software read: 1.373v - 1.395v // DMM read: 1.373v - 1.376v

  • 4000Mhz
  • Pstates0
  • Vcore Offset + 0.125
  • 3200mhz Dram
  • No Stable
  • LLC Level 1
  • On Load - Software read: 1.373v - 1.395v // DMM read: 1.365v - 1.368v
  • Stock Settings
  • IDLE Max peaks to 1.570v
  • On Load - Software read: 1.286v - 1.308v // DMM read: 1.298v
Im trying to test more situations. Keep updating. If you need specific settings let me know!
 

·
Always Learning
Joined
·
1,194 Posts
Quote:
Originally Posted by majestynl View Post

Okey guys, round 1 of my test. Did only test with 4000mhz and Stock settings for now.

Cpu: 1800x
Memory: G Skill - F4-3200C14D-16GTZR

Round 1:

  • 4000Mhz
  • Pstates0
  • Vcore Offset + 0.125
  • 3200mhz Dram
  • Stable
  • LLC Level 3
  • On Load - Software read: 1.395v - 1.417v // DMM read: 1.417v - 1.423v

  • 4000Mhz
  • Pstates0
  • Vcore Offset + 0.125
  • 3200mhz Dram
  • No Stable
  • LLC Auto
  • On Load - Software read: 1.373v - 1.395v // DMM read: 1.373v - 1.376v

  • 4000Mhz
  • Pstates0
  • Vcore Offset + 0.125
  • 3200mhz Dram
  • No Stable
  • LLC Level 1
  • On Load - Software read: 1.373v - 1.395v // DMM read: 1.365v - 1.368v
  • Stock Settings
  • IDLE Max peaks to 1.570v
  • On Load - Software read: 1.286v - 1.308v // DMM read: 1.298v
Im trying to test more situations. Keep updating. If you need specific settings let me know!
Awesome post! What software are you using btw to read the volts and what is bios stating as the volt (it moves around too i know) when you are doing +0.125 offset with no LLC? Thanks!
 

·
Hardware Abuser
Joined
·
3,451 Posts
Today I wrestled with the second R7 1700 and I mean wrestled. I'm coming to the conclusion that all the "BIOS" issues people are seeing are more CPU related that board related. I know you have seen the results I had with the two other CPUs I have clocked on the CHVI @gupsterg
To start with on BIOS 0902 I kept losing my SSD , so I decided to flash the 1101 also tried 0038 that didn't make a difference with the SSD issue. After setting a static SOC voltage of 1.0 the issue with the disappearing drive was gone. But if there was a thumb drive left in the system after reboot it would always try to boot from it regardless of boot settings in BIOS.
On versions 1001 and 0038 with this CPU the voltages were crazy, showing super high in BIOS, CPU overvoltage warnings etc.
Reverted back to 902 and seemed a bit more normal. One thing to note. Using Win7x64 and my DMM reading would match exactly what I set in BIOS. SW including HWinfo read too high. example 1.25v core in BIOS and DMM would read 1.33v in HWInfo.
Memory was another battle, tried different straps,timings and voltage combinations but it will not go past ~ 2700 with any kind of stability, as you know the other two had no issues with up to 3300 and tight timings.
In the end I got to 4.0 @ 1.425v stable. Ram at 2700 14-14-14-14-28 1.35v SOC 1.1
Sp back to my initial statement, I feel it has more to do with the CPU than it does BIOS. Many of the "weird" issues people are seeing in the CHVI thread I saw today and only difference was the chip.
I'll leave this here for you actual voltage is 1.55 DMM

 

·
Meddling user
Joined
·
7,423 Posts
Discussion Starter #99
@Johan45

+rep, cheers mate
thumb.gif
.

@S1L3N7D3A7H

+rep, thank you
thumb.gif
.

@finalheaven

I had been having a situation, after I increased RAM speed from 2133MHz to 2400MHz for my 3.8GHz profile some tests where failing, when repeated for similar settings. The issue to me seems it's lack of VCORE in my case.

So say originally on 3.8GHz / 2133MHz I used +106.25mV / SOC: [Auto] (~0.838V), this passed x264, RB, [email protected] I upped to +112.5mV when set 2400MHz RAM and mobo with SOC: [Auto] (~1.050V). This seemed to work but then had intermittent [email protected] errors. So increased to +118.75mV with SOC: 0.950V, this passed all but [email protected] again an issue, next I went +118.75mV with SOC: 0.925V all passed except Y-Cruncher. Now I have been on +125mV with SOC: 0.925V , so far [email protected] has passed ~7hrs.



I'm upping the length of stability testing, I will do the same as my i5 4690K, 8hrs x264, 4hrs RB stress mode, 12hrs+ [email protected] and gonna throw in some 4hrs or so of Y-Cruncher.

The above 3.8GHz settings are meaning my actual VCORE with +125mV offset is ~1.344V
redface.gif
, I won't achieve a higher OC on this CPU for 24/7 use IMO, as I wish to remain at ~1.35V.

My current 3.8GHz 2400MHz UEFI settings changes from defaults.








 
81 - 100 of 359 Posts
Top