I9 9900K Cinebench R20 issue

[Sajeev]

Hi guys, I OC’d i9 9900K to 5.0GHz with a core voltage of 1.28V, now the problem is that Aida64, Prime95 all run fine(tested for about 30 mins). Cinbench R15 runs fine too but as soon as I launch Cinebench R20, I get a blue screen error, temps seem to be ok at 75-79 degrees.

 

Any ideas what could be causing the issue?

 

Hardware config

CPU: i9 9900K

Motherboard: Maximus XI Extreme 

RAM: 64GB Corsair Dominator Platinum @3200MHz

Storage: M.2 Samsung 970 Evo Pro

GPU: Radeon VII

PSU: evga 2000 watt 80 Gold Standard

Cooling: Custom Liquid Cooling 

 

Could the above issue be because of voltage? I have got a mixed reaction from people around me, some say they are getting same speeds at 1.25V and feel I am adding too much voltage and some are telling me to increase the same, not sure what to do, any guidance please?

 

Cheers

Sajeev

 

[GMart84]

@Sajeev SO I tried R20 last night for the first time too and I needed to pull back the OC to get it to run stably.  In R15, I cna benchmark at 5.4 Ghz with a BCLK of 101.95 set in the bios and a multiplier of 53, but that would not complete with the R20 version.  I had to back it off to 101 BLCK in order to get it to complete the test.  I am sure there are some differences in how the new version is testing.

 

I run my 9900K at 1.29 vCore in bios for my 5.1 Ghz OC that I use everyday, and 1.39 vCore for my 5.35-5.4 Ghz Benchmarking OC.

 

Check your VCCIO and VCCSA as I have never been able to get my OCs stable unless I have modified those.

[tarfeef101]

R20 uses more instructions, even AVX-512 on supported CPUs. Not sure if your CPU supports it, or if your other programs use it (depends on versions of that software), but it is probably a much more intensive program now. 

I'd check your AVX offset for your OC, and see if you can get it stable by increasing the offset, so it doesn't affect your performance for non-AVX workloads (ie gaming)

 

Edit: according to intel ark, it doesn't seem to support AVX-512. But it still probably will use AVX2, which hits your CPU about as hard as possible. 

Edited March 13, 2019 by tarfeef101

[Zando_]

Just now, tarfeef101 said:

R20 uses more instructions, even AVX-512 on supported CPUs. Not sure if your CPU supports it, or if your other programs use it (depends on versions of that software), but it is probably a much more intensive program now. 

I'd check your AVX offset for your OC, and see if you can get it stable by increasing the offset, so it doesn't affect your performance for non-AVX workloads (ie gaming)

^^^ It's much harder to run than CB15. I can push 4.74GHz or so in CB15, can't get any higher than 4.56Ghz in CB20 without a crash then BSOD (X5670 on an EVGA Classified X58 SLI 4 way and custom loop). I don't think X58 supports an AVX offset, at least I haven't seen one, but adjusting that on the 9900K should keep it from crashing. 

[tarfeef101]

Just now, Zando Bob said:

^^^ It's much harder to run than CB15. I can push 4.74GHz or so in CB15, can't get any higher than 4.56Ghz in CB20 without a crash then BSOD (X5670 on an EVGA Classified X58 SLI 4 way and custom loop). I don't think X58 supports an AVX offset, at least I haven't seen one, but adjusting that on the 9900K should keep it from crashing. 

Yeah, I should give it a shot on my 7700K. Maybe I can actually hit a good score with my system before the real OCers upload good ones. 

Only problem is I run windows 8.1 on that rig so HWBot won't accept it

[Sajeev]

7 hours ago, tarfeef101 said:

R20 uses more instructions, even AVX-512 on supported CPUs. Not sure if your CPU supports it, or if your other programs use it (depends on versions of that software), but it is probably a much more intensive program now. 

I'd check your AVX offset for your OC, and see if you can get it stable by increasing the offset, so it doesn't affect your performance for non-AVX workloads (ie gaming)

 

Edit: according to intel ark, it doesn't seem to support AVX-512. But it still probably will use AVX2, which hits your CPU about as hard as possible. 

So you mean I should try putting the AVX offset to 3 or 4?

 

Rgds

Sajeev

[Sajeev]

11 hours ago, tarfeef101 said:

R20 uses more instructions, even AVX-512 on supported CPUs. Not sure if your CPU supports it, or if your other programs use it (depends on versions of that software), but it is probably a much more intensive program now. 

I'd check your AVX offset for your OC, and see if you can get it stable by increasing the offset, so it doesn't affect your performance for non-AVX workloads (ie gaming)

 

Edit: according to intel ark, it doesn't seem to support AVX-512. But it still probably will use AVX2, which hits your CPU about as hard as possible. 

I set the offset to 1 and now it's working fine, got the R20 to run flawlessly a couple of times without any hiccups. Been running the OOCT stress test for about 18 minutes now, plan to run it for about an hour and see the results. Temps seem to be varying quite a bit mostly between 70-80, with once in a while hitting 86 on one core.

 

You guys have been great, thank you so much...

 

Cheers

Sajeev

[Sajeev]

14 hours ago, GMart84 said:

@Sajeev SO I tried R20 last night for the first time too and I needed to pull back the OC to get it to run stably.  In R15, I cna benchmark at 5.4 Ghz with a BCLK of 101.95 set in the bios and a multiplier of 53, but that would not complete with the R20 version.  I had to back it off to 101 BLCK in order to get it to complete the test.  I am sure there are some differences in how the new version is testing.

 

I run my 9900K at 1.29 vCore in bios for my 5.1 Ghz OC that I use everyday, and 1.39 vCore for my 5.35-5.4 Ghz Benchmarking OC.

 

Check your VCCIO and VCCSA as I have never been able to get my OCs stable unless I have modified those.

Can you please tell me what VCCSA value are you using? I got most of the tests stable except for the Asus Realbench which kept on crashing. Would be greatful to you if you could share your VCCSA and VCCIO values.

 

Rgds

Sajeev

[Falkentyne]

15 hours ago, Sajeev said:

Hi guys, I OC’d i9 9900K to 5.0GHz with a core voltage of 1.28V, now the problem is that Aida64, Prime95 all run fine(tested for about 30 mins). Cinbench R15 runs fine too but as soon as I launch Cinebench R20, I get a blue screen error, temps seem to be ok at 75-79 degrees.

 

Any ideas what could be causing the issue?

 

Hardware config

CPU: i9 9900K

Motherboard: Maximus XI Extreme 

RAM: 64GB Corsair Dominator Platinum @3200MHz

Storage: M.2 Samsung 970 Evo Pro

GPU: Radeon VII

PSU: evga 2000 watt 80 Gold Standard

Cooling: Custom Liquid Cooling 

 

Could the above issue be because of voltage? I have got a mixed reaction from people around me, some say they are getting same speeds at 1.25V and feel I am adding too much voltage and some are telling me to increase the same, not sure what to do, any guidance please?

 

Cheers

Sajeev

 

I never had any problems running R20 on my 9900K.  No AVX offsets.  Set voltage to 1.340v, 5.1 ghz, cache x47 and did 5 repeated loops (without stopping) of R20, and checked for WHEA Errors in HWinfo64 and there were none.  R20 only seemed to use about 5 more watts than R15, still not nearly as much as Prime95 29.6 build 7 smallest FFT (AVX disabled).

I have absolutely no idea how R20 would -instantly- BSOD as soon as you start it.  Usually a program will at least run until the system gets too hot and reaches the unstable point, then crash, not just insta-BSOD.

What loadline calibration (LLC) are you using?  You never told us.  You should be using LLC5 or LLC6, but not higher (LLC6 will have less vdroop than LLC5 but worse transient response; LLC5 will have more vdroop, meaning you will need a higher bios voltage to compensate for the required "true load" target voltage, but voltage stability will be better).

OCCT was showing a vcore of 1.14v, and I know that is not accurate at all.

You don't have your CPU Vcore sensors visible in your HWinfo screenshot (VID is NOT Vcore!).

[Sajeev]

29 minutes ago, Falkentyne said:

I never had any problems running R20 on my 9900K.  No AVX offsets.  Set voltage to 1.340v, 5.1 ghz, cache x47 and did 5 repeated loops (without stopping) of R20, and checked for WHEA Errors in HWinfo64 and there were none.  R20 only seemed to use about 5 more watts than R15, still not nearly as much as Prime95 29.6 build 7 smallest FFT (AVX disabled).

I have absolutely no idea how R20 would -instantly- BSOD as soon as you start it.  Usually a program will at least run until the system gets too hot and reaches the unstable point, then crash, not just insta-BSOD.

What loadline calibration (LLC) are you using?  You never told us.  You should be using LLC5 or LLC6, but not higher (LLC6 will have less vdroop than LLC5 but worse transient response; LLC5 will have more vdroop, meaning you will need a higher bios voltage to compensate for the required "true load" target voltage, but voltage stability will be better).

OCCT was showing a vcore of 1.14v, and I know that is not accurate at all.

You don't have your CPU Vcore sensors visible in your HWinfo screenshot (VID is NOT Vcore!).

LLC6

[Sajeev]

2 hours ago, Falkentyne said:

I never had any problems running R20 on my 9900K.  No AVX offsets.  Set voltage to 1.340v, 5.1 ghz, cache x47 and did 5 repeated loops (without stopping) of R20, and checked for WHEA Errors in HWinfo64 and there were none.  R20 only seemed to use about 5 more watts than R15, still not nearly as much as Prime95 29.6 build 7 smallest FFT (AVX disabled).

I have absolutely no idea how R20 would -instantly- BSOD as soon as you start it.  Usually a program will at least run until the system gets too hot and reaches the unstable point, then crash, not just insta-BSOD.

What loadline calibration (LLC) are you using?  You never told us.  You should be using LLC5 or LLC6, but not higher (LLC6 will have less vdroop than LLC5 but worse transient response; LLC5 will have more vdroop, meaning you will need a higher bios voltage to compensate for the required "true load" target voltage, but voltage stability will be better).

OCCT was showing a vcore of 1.14v, and I know that is not accurate at all.

You don't have your CPU Vcore sensors visible in your HWinfo screenshot (VID is NOT Vcore!).

To reply to your few concerns, BSOD in R20 appears when it runs for about a few seconds 3-4, how ever I solved it by setting the AVX offset value to 2 and it works just fine now, however the only thing I am not sure about is why does the Asus Realbench keep on crashing, no idea, as per @GMart84 above it has something to with VCCIO and VCCSA. In my case my bios shows VCCIO voltage to be higher than that of VCCSA.. strange. Tried reducing the values but still of no use, just want this thing to run smooth... It actually crashes when it starts H.264 Video Editing test, OpenCl and Heavy Multimedia.., Stress test too desn't work and crashes within 5 seconds.. just a little frustrated now..:(( 

 

 

Best Regards

Sajeev

[GMart84]

@Sajeev My settings on the Gigabyte Aorus Z390 Extreme are:

 

CPU vCore: Normal

DVID: -0.50

VCCIO: 1.050v

VCCSA: 1.100v

 

When I had my Asus Maximus X Hero (Wi-FI ac) with the 9900K

 

CPU Voltage: Offset

Offset Direction: -

Offset Voltage: 0.08

LLC Level 6

 

Those settings with my chip had it hovering right around 1.29-1.30v for my 5.1 Ghz OC.

[Sajeev]

48 minutes ago, GMart84 said:

@Sajeev My settings on the Gigabyte Aorus Z390 Extreme are:

 

CPU vCore: Normal

DVID: -0.50

VCCIO: 1.050v

VCCSA: 1.100v

 

When I had my Asus Maximus X Hero (Wi-FI ac) with the 9900K

 

CPU Voltage: Offset

Offset Direction: -

Offset Voltage: 0.08

 LLC Level 6

 

Those settings with my chip had it hovering right around 1.29-1.30v for my 5.1 Ghz OC.

Not working for me, for some weird reason both VCCIO and VCCSA keep going back to 1.2500 V after bios restarts.. Have no clue what could be the issue.. Any idea? Oh yes I was saving them before exiting the bios, also tried creating a profile...

 

Rgds

Sajeev

[GMart84]

@Sajeev That seems high given my experience.  I mean you really do not need to worry until you hit 1.3v or higher on those values so the 1.25v isnt too bad.  Do you have one of the automatic OC features activated.

 

[Sajeev]

44 minutes ago, GMart84 said:

@Sajeev That seems high given my experience.  I mean you really do not need to worry until you hit 1.3v or higher on those values so the 1.25v isnt too bad.  Do you have one of the automatic OC features activated.

 

Nope, none of the OC features are activated... 

 

Tks for helping me out man.. really appreciate it.

 

 

Rgds

Sajeev

[tarfeef101]

14 hours ago, Sajeev said:

I set the offset to 1 and now it's working fine, got the R20 to run flawlessly a couple of times without any hiccups. Been running the OOCT stress test for about 18 minutes now, plan to run it for about an hour and see the results. Temps seem to be varying quite a bit mostly between 70-80, with once in a while hitting 86 on one core.

 

You guys have been great, thank you so much...

 

Cheers

Sajeev

Feel free to mark this as answered

Glad you got everything working!

[Sajeev]

7 hours ago, tarfeef101 said:

Feel free to mark this as answered

Glad you got everything working!

Everything works but for Asus Realbench....

 

Rgds

Sajeev

[Falkentyne]

3 hours ago, Sajeev said:

Everything works but for Asus Realbench....

 

Rgds

Sajeev

So let's try something.

Set your LLC to 6 and manual vcore voltage (not adaptive/offsets) to 1.35v static voltage.

Set your AVX offset to 0.

 

Now run realbench 2.56.

Does it crash?

[Sajeev]

38 minutes ago, Falkentyne said:

So let's try something.

Set your LLC to 6 and manual vcore voltage (not adaptive/offsets) to 1.35v static voltage.

Set your AVX offset to 0.

 

Now run realbench 2.56.

Does it crash?

Let me give it a try.. currently running memtest86 to check if the ram has some issues.. will get back asap. Tks a ton man for helping me. Really appreciate it.

 

Rgds

Sajeev

[tarfeef101]

One thing to be aware of: Running your RAM at higher speeds can make OCing your CPU harder (make it unstable w/ the same settings as when your RAM was stock). On intel mainstream, if you drop your RAM by a few hundred MHz you may be able to get your CPU stable with very little difference in performance.

Also, if you do end up having RAM instability (you mentioned you were checking it), you may need a bit more System Agent voltage, as that includes the memory controller. But, do not push it very far, as it is rather sensitive, and pushing that as far as your CPU is not really a good idea (may damage/degrade it). I've heard a few numbers tossed around as a "limit", usually between 1.2 and 1.3V

[Sajeev]

Guys Finally Finally Finally I got Realbench running too and finish successfully with the report saying EVERYTHING IS OK (though it shows failed to parse the results).Actually the only thing that I changed was Max CPU capability to Auto from 140% that I was running previously and didn't disable the Intel Speed Shift Technology (and yes I have not changed my CPU Core/Cache Voltage, it is still 1.29 with an AVX offset of 2, Ram voltage is 1.35 set by Auto), both VCCIO and VCCSA are at 1.2500 V (by default), that's it everything as of now running fine. Passed R20 Cinebench and intel Burn Test. I am going to run OCCT stress test for about good 7-8 hours to check if the system is stable, any temperature issue, etc etc.. If everything is fine I will post all my settings here for you guys.
 

@GMart84 @tarfeef101 @Zando Bob @Falkentyne 

 

Once again I would like to thank each one of you who helped me with this issue . Tks Guys, Tks a Ton

 

Best Rgds
Sajeev

[Sajeev]

11 hours ago, tarfeef101 said:

One thing to be aware of: Running your RAM at higher speeds can make OCing your CPU harder (make it unstable w/ the same settings as when your RAM was stock). On intel mainstream, if you drop your RAM by a few hundred MHz you may be able to get your CPU stable with very little difference in performance.

Also, if you do end up having RAM instability (you mentioned you were checking it), you may need a bit more System Agent voltage, as that includes the memory controller. But, do not push it very far, as it is rather sensitive, and pushing that as far as your CPU is not really a good idea (may damage/degrade it). I've heard a few numbers tossed around as a "limit", usually between 1.2 and 1.3V

I guess different ram's have different power requirements:

 

CPU System Agent: This is important because it handles the IO between the CPU and other domains. For overclocking, this can help stabilize higher memory frequencies. I want to point out since I am on AUTO, the UEFI is showing a much higher number than what should usually be the healthy range. If possible, do not go outside of 1.35V as a maximum. I must adjust mine accordingly for memory stability. In Windows 10, AI Suite 3 reports 1.12V for System Agent (AUTO) and is fully stable for my 3600MHz kit. This may be a bug with the first BIOS release.

Setting Voltage in increments is always a good idea when overclocking, the values I used are what worked for me, and are just a starting or reference point for others. Overclocking and voltage will vary from system to system so any values I suggest are just to help others try to achieve their overclocking goals and are not 100% guaranteed.


• DDR4 frequency range: DDR4-2133 - DDR4-2800 Required CPU VCCIO Voltage range: 1.05V - 1.15V Required CPU System Agent Voltage range: 1.05V - 1.15V
• DDR4 frequency range: DDR4-2800 - DDR4-3600 Required CPU VCCIO Voltage range: 1.10V - 1.25V Required CPU System Agent Voltage range: 1.10V - 1.30V
• DDR4 frequency range: DDR4-3600 - DDR4-4266 Required CPU VCCIO Voltage range: 1.15V - 1.30V Required CPU System Agent Voltage range: 1.20V - 1.35V
 

Source of the above info: https://www.overclock.net/forum/18088-hardware-reviews-program/1711158-9900k-z390-soldered-overclocking.html

 

Since mine is a 3600 MHz, the system has automatically given it a value of 1.25, no matter what I do it just doesn't change. 

 

Best regards

Sajeev

[alatron978]

Cinebench R20 is very demanding, it requires 50mv more then other avx workloads which is a very significant amount.

[Sajeev]

13 hours ago, alatron978 said:

Cinebench R20 is very demanding, it requires 50mv more then other avx workloads which is a very significant amount.

Ohhh.....no wonder it was crashing.

 

Tks

Rgds

[Sajeev]

@GMart84 @tarfeef101 @Zando Bob @Falkentyne @alatron978 guys sorry to bother you guys, just want to understand one thing, The VRM on my MoBo get's really hot and its actually impossible to touch, you can actually feel the heat... any suggestions on this..

 

PS: currently my MoBo is still not inside a case but on an open bench without any direct airflow towards the VRM. Also I have changed my CPU Core/Cache voltage to 1.255 and it's really stable..

 

Tks & Regards

Sajeev