EVGA

Hot!X299 Dark Stability Overclock Notes/Info

Page: < 1234 > Showing page 2 of 4
Author
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/02 09:53:33 (permalink)
GGTV-Jon
geninfo are you up for testing other Thermo Grizzly products - for Science!!
 
 Thermal Grizzly Conductonaut - Thermal Conductivity of this product is 73 W/mk and thermal resistance of 0.0021 K/W.
https://www.newegg.com/Product/Product.aspx?Item=9SIA4RE6M28090&cm_re=Thermal_Grizzly-_-2S7-00A9-00013-_-Product

 
Don't change EVGA CLC 280 pre-applied thermal paste. Still advice, but see #33 below for different results.
 
Removed Kryonaut.  Applied thin, even coat Conductonaut to top of IHS.  For science.
 
CPU temp increased 4C. :(
 
Checked cooler tightening.  Looser only raised temp further. (See #36)
 
Estimate CPU temps now 7C hotter than CLC pre-applied thermal paste.
 
 
post edited by geninfo - 2018/02/02 13:31:49
#31
bcavnaugh
The Crunchinator
  • Total Posts : 38105
  • Reward points : 0
  • Joined: 2012/09/18 17:31:18
  • Location: USA Affiliate E5L3CTGE12 Associate 9E88QK5L7811G3H
  • Status: offline
  • Ribbons : 276
Re: X299 Dark Bench Notes/Info 2018/02/02 10:45:00 (permalink)
geninfo
No idea why this forum or Kingpin forum thread does not reference, but:
https://xdevs.com/guide/e299ocg/
Courtesy jeter9209 in another thread.

 
Thanks for the update. Are you Planning on Overclocking your X299 Dark under LN2?

Air/water cooling clock/voltage/temperature range

With a good aircooling or watercooling setup that can handle full load temperatures under 80 °C common Skylake-X processors can reach stable frequency around 4.9-5.3 GHz on all four cores. Special loading cases using AVX-instructions drop this bar to about 4.6-4.8 GHz due to higher stress on processor. These numbers are valid only after processor delidding and replacement of stock TIM to high-performance grease, such as Kingpincooling.com KPx.
 
Funny that they are running Bios 0.00. Without at least 2.05 I would be carful.

 
post edited by bcavnaugh - 2018/02/02 11:28:20

Affiliate Code: E5L3CTGE12 & Associate Code: 9E88QK5L7811G3H


 
#32
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/02 12:38:53 (permalink)
GGTV-Jon
geninfo are you up for testing other Thermo Grizzly products - for Science!!

 
Proper application of liquid metal between IHS and cooler can lower temp.
 
Earlier results not matching theoretical science:
Kryonaut 12.5 wm/k
Conductonaut 73 wm/k
6x thermal conductivity leading to higher temp!?
 
Resort to desperate empirical science:
Remove CLC heat sink.
Tape sides of CPU bracket where open socket edges exposed.
Slather heat sink AND IHS with additional liquid metal.  No puddles (see #39 below).  Plus very small dot in middle of IHS.
Reattach heat sink. Screws tight.
 
Result
Loaded temperature down 18C (from 89C to 71C)!
Estimate 11C down from stock paste.
 
Note
Advice remains do not remove EVGA CLC 280 pre-applied paste.
Unless you dare instant motherboard and CPU destruction without warranty support.
Conductive liquid metal in socket = over.
 
Will liquid metal ooze when I orient vertical in case?
Stay tuned for tears.
post edited by geninfo - 2018/02/04 18:58:52
#33
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/02 12:57:34 (permalink)
bcavnaugh
Thanks for the update. Are you Planning on Overclocking your X299 Dark under LN2?

 
No.  Working on 24/7 stable creative/scientific workstation.
Theoretically custom loop next ($480), then chiller loop ($900), but EVGA CLC 280 only $130 (plus Conductonaut/Tape +$20).
 
 
bcavnaugh
Air/water cooling lock/voltage/temperature range
With a good aircooling or watercooling setup that can handle full load temperatures under 80 °C common Skylake-X processors can reach stable frequency around 4.9-5.3 GHz on all four cores. Special loading cases using AVX-instructions drop this bar to about 4.6-4.8 GHz due to higher stress on processor. These numbers are valid only after processor delidding and replacement of stock TIM to high-performance grease, such as Kingpincooling.com KPx.
 

 
Not for 7980XE.  From Kingpin guide:
 
"High-core count processors such as i9-7960X or i9-7980XE may drop few hundred MHz from [4500Mhz] maximum stable clock, mostly due to the thermal limitations"
 
So, they say probably 4.2Ghz max stable 7980XE+Dark without exotic cooling. 
post edited by geninfo - 2018/02/02 13:32:19
#34
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/02 13:06:00 (permalink)
Some graphs. Prime95 24K, in-place FFT, 5 minutes per test for 33 minutes.  Liquid temperature start  normalized; no significant ambient temp delta. 
Experimental 1.30 VCore at 44Ghz with 3200 DDR4 on XMP profile at 1.35V, with 1.166 VCCIO.
 
 First liquid metal application:

 
Second liquid metal application:

 
post edited by geninfo - 2018/02/02 13:12:09

Attached Image(s)

#35
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/02 13:25:32 (permalink)
Importance of tightening down heat sink demonstrated.
Captured with first liquid metal application (see above).
 
89C = Fully tightened down.
91C = Loosened all screws two turns.
94C = Loosened all screws two more turns.
89C = All screws tightened back up.
 
Fully tightened = finger strength tightened with screwdriver; not wrist strength tightened.
 
Read as possibly  more thermal interface material needed.
 
 
 
post edited by geninfo - 2018/02/02 13:37:39

Attached Image(s)

#36
bcavnaugh
The Crunchinator
  • Total Posts : 38105
  • Reward points : 0
  • Joined: 2012/09/18 17:31:18
  • Location: USA Affiliate E5L3CTGE12 Associate 9E88QK5L7811G3H
  • Status: offline
  • Ribbons : 276
Re: X299 Dark Bench Notes/Info 2018/02/02 13:45:07 (permalink)
geninfo
bcavnaugh
Thanks for the update. Are you Planning on Overclocking your X299 Dark under LN2?

 
No.  Working on 24/7 stable creative/scientific workstation.
Theoretically custom loop next ($480), then chiller loop ($900), but EVGA CLC 280 only $130 (plus Conductonaut/Tape +$20).
 
 
bcavnaugh
Air/water cooling lock/voltage/temperature range
With a good aircooling or watercooling setup that can handle full load temperatures under 80 °C common Skylake-X processors can reach stable frequency around 4.9-5.3 GHz on all four cores. Special loading cases using AVX-instructions drop this bar to about 4.6-4.8 GHz due to higher stress on processor. These numbers are valid only after processor delidding and replacement of stock TIM to high-performance grease, such as Kingpincooling.com KPx.
 

 
Not for 7980XE.  From Kingpin guide:
 
"High-core count processors such as i9-7960X or i9-7980XE may drop few hundred MHz from [4500Mhz] maximum stable clock, mostly due to the thermal limitations"
 
So, they say probably 4.2Ghz max stable 7980XE+Dark without exotic cooling. 

This still Applies for the Higher End Processors as well.
"These numbers are valid only after processor delidding and replacement of stock TIM to high-performance grease, such as Kingpincooling.com KPx."
Yes you do have a "7980XE (delid, 4.3Ghz bin)"
On my Core i9 7940X is only running at 4500MHz on a custom Water Loop and higher then I have to push the vCore and for 24 Hour Operating Time it would kill the CPU after a year or two.
 
"Working on 24/7 stable creative/scientific workstation" Mine X299 Dark ATM is pushing really hard on PG.
 
Man the Forum is really crappy today or the network is being overloaded by all the new members going after their Elite Membership
post edited by bcavnaugh - 2018/02/02 13:53:24

Affiliate Code: E5L3CTGE12 & Associate Code: 9E88QK5L7811G3H


 
#37
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/02 13:48:29 (permalink)
bcavnaugh
 
This still Applies for the Higher End Processors as well.
"These numbers are valid only after processor delidding and replacement of stock TIM to high-performance grease, such as Kingpincooling.com KPx."



Agreed.  All results in this thread are delidded 7980XE.
#38
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/02 13:56:26 (permalink)
Latter half of video here demonstrates liquid metal application and avoiding puddles:
 
https://www.gamersnexus.net/guides/3068-how-to-delid-intel-i9-cpu-and-apply-liquid-metal
 
Note, Steve demonstrates delidding  (between CPU and temporarily removed IHS).  
But same concepts apply when putting between IHS and cooler.
post edited by geninfo - 2018/02/02 14:02:52
#39
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/02 14:00:40 (permalink)
bcavnaugh
 Man the Forum is really crappy today or the network is being overloaded by all the new members going after their Elite Membership



Easier to buy and register CLC and PSU.  I did to get Dark day one! 
#40
GGTV-Jon
FTW Member
  • Total Posts : 1811
  • Reward points : 0
  • Joined: 2017/11/25 14:11:43
  • Location: WA, USA
  • Status: offline
  • Ribbons : 18
Re: X299 Dark Bench Notes/Info 2018/02/02 14:44:52 (permalink)
My Dark was delivered today, won't be able to see it till I get home from work - still working on getting all the peices for the build though
#41
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/03 10:50:11 (permalink)
New 4.5Ghz stable overclock found.  Updated in first post. 
 
This post archives 4.3Ghz overclock info originally in first post.
 
Build
Thermal Grizzly Kryonaut Replacing CLC Stock Paste (NOT RECOMMENDED #21)
 
BIOS
1.06 (on BIOS2)
Vcore: Override 1.10
CPU Multipler Settings: 43x
AVX2 Negative Offset: -3
AVX3 Negative Offset: -5
Memory: XMP Profile 1 (DDR4-3200 16-18-18-36-1.35V)
VCCIO: 1.166

OS
Windows 10.0.16299.125
 
Driver
Samsung_NVM_Express_Driver 2.3
 
EVGA Flow Control Fan Curve for *Coolant*
20/25
30/25
33/40
42/70
48/100
60/100
Software

EVGA Flow Control 2.0.4.0 (Firmware Version 2.10.0.0)
 
Note
MemTest86 in Legacy BIOS 4.3.7 mode 1 pass overnight at JEDEC (2133).
MemTest86 in Legacy BIOS 4.3.7 mode 1.5 passes overnight at XMP Profile 1 (3200).
Overnight Prime95 Blend test no failures.
Prime95 8K FFT in-place 30 minutes AVX on, CPU temp 90-92C, system watts 580, no throttling observed.
Must run EVGA Flow Control for fans to follow coolant curve. #19
#42
GGTV-Jon
FTW Member
  • Total Posts : 1811
  • Reward points : 0
  • Joined: 2017/11/25 14:11:43
  • Location: WA, USA
  • Status: offline
  • Ribbons : 18
Re: X299 Dark Bench Notes/Info 2018/02/03 11:38:27 (permalink)
@geninfo - now that you have done all of that Der8auer goes and releases the direct contact frame to get the IHS out of the picture
https://www.overclockers.co.uk/der8auer-skylake-x-direct-die-frame-hs-005-dr.html
 
Video with GamersNexus
https://youtu.be/DJOVdmd7NHQ
#43
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/03 11:39:48 (permalink)
Explanation and graphs of Prime95 stability test.
 
Different people recommend test times for Prime95 stability check.
Ranges from 15 minutes to 72 hours!
 
To balance time with certainty, used this Prime95 torture test:
Custom
Threads = 36 (2 * cores)
Min FFT = 8
Max FFT = 4096
Run FFTs in-place = Checked
Time to run each size = 5
 
Runs 70 different tests in set order.
Overall time 350 minutes.
Repeats until stopped.
700 minutes, all tests run twice.
But 5 minutes per test not exact wall time.
Budget 400 minutes wall time per test cycle, 800 minutes for two test cycles.
 
Graph of 4.5Ghz stability run (762 minutes):

 
Prime95 test sequence order runs small FFT (high heat) tests in first and third quarter of test cycle.
Repeated patterns in graph show two full test cycles.
Almost.  Actually manually stopped test 2 tests short of full second cycle. :(
But close enough! :)
post edited by geninfo - 2018/02/03 16:14:46

Attached Image(s)

#44
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/03 11:44:04 (permalink)
GGTV-Jon
@geninfo - now that you have done all of that Der8auer goes and releases the direct contact frame to get the IHS out of the picture
https://www.overclockers.co.uk/der8auer-skylake-x-direct-die-frame-hs-005-dr.html
 
Video with GamersNexus
https://youtu.be/DJOVdmd7NHQ



Yes, am aware of that.
Cringed when I saw because I thought you may dare me. :p
For science? 
#45
GGTV-Jon
FTW Member
  • Total Posts : 1811
  • Reward points : 0
  • Joined: 2017/11/25 14:11:43
  • Location: WA, USA
  • Status: offline
  • Ribbons : 18
Re: X299 Dark Bench Notes/Info 2018/02/03 11:47:01 (permalink)
geninfo
GGTV-Jon
@geninfo - now that you have done all of that Der8auer goes and releases the direct contact frame to get the IHS out of the picture
https://www.overclockers.co.uk/der8auer-skylake-x-direct-die-frame-hs-005-dr.html
 
Video with GamersNexus
https://youtu.be/DJOVdmd7NHQ



Yes, am aware of that.
Cringed when I saw because I thought you may dare me. :p
For science? 





ALWAYS - For SCIENCE!!
#46
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/03 12:33:48 (permalink)
GGTV-Jon
ALWAYS - For SCIENCE!!



Signed up for availability notification list.
May do for second build so I can get workstation into use.
Rumor has it you have a Dark now.  What CPU?  Maybe you should advance science first? ;)
#47
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/03 13:54:58 (permalink)
Not into competitive benching, but know some may want info.
With new 4.5Ghz stable CPU overclock and GPU +50 core overclock.
Not yet convinced GPU stable at 2025Mhz, even though guaranteed.  Needs testing.
 
Time Spy https://www.3dmark.com/3dm/24973727
Fire Strike https://www.3dmark.com/3dm/24974462
Furmark 2160 http://www.ozone3d.net/gp.b/score.php?which=520371
Furmark 720 http://www.ozone3d.net/gp.b/score.php?which=520377
 
Thanks bcavnaugh that scores have links online even with 3DMark Basic.
post edited by geninfo - 2018/02/03 14:14:12
#48
bcavnaugh
The Crunchinator
  • Total Posts : 38105
  • Reward points : 0
  • Joined: 2012/09/18 17:31:18
  • Location: USA Affiliate E5L3CTGE12 Associate 9E88QK5L7811G3H
  • Status: offline
  • Ribbons : 276
Re: X299 Dark Bench Notes/Info 2018/02/03 16:03:58 (permalink)
Not bad and Great Scores
https://www.3dmark.com/compare/spy/3246791/spy/2043335
https://www.3dmark.com/compare/fs/14848012/fs/13077938
 
I really have to work on my X299 Overclocking and really need to mess with the AVX2 and AVX3 Settings.
4GHz https://valid.x86.fr/zszbmd  5GHz  https://valid.x86.fr/6inny9
I did reduce my memory from 3200 to 3000 and seems to say a lot cooler all-around.
post edited by bcavnaugh - 2018/02/03 16:14:38

Affiliate Code: E5L3CTGE12 & Associate Code: 9E88QK5L7811G3H


 
#49
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/04 11:14:15 (permalink)
Phantom throttling may undermine performance at higher overclock.
 
Theory
Phantom throttling = apparent CPU frequency stays same, but performance drops.
Internal CPU frequency divisor not sensed by HWiNFO64.
Detectable by temp or power draw falling.
Caused by insufficient power for overclock.
Higher overclock defeated if performance drops.
 
Example
Increase Mesh Ratio from 30 to 32 in known stable config.
Hangs within one minute of Prime95 test.
Vmesh:
1.19 Override Hang 6 Minutes
1.20 Override BSOD Code 124 9 Minutes
-- Raise VCore from 1.170 to 1.175 --
1.20 Override BSOD Hang 8 Minutes
1.20 Adaptive BSOD Hang 18 Minutes
1.22 Adaptive BSOD Hang 19 Minutes
1.25 Adaptive Pass!

BUT, look at temp/power graph!
 
Prime95 40K, in-place FFT, 5 minute per test, 30 minute run (manually stopped)
 


Temp/power draw usually remains constant across test.
Graph shows temp/power draw bouncing high and low.
Infer phantom throttling.
 
Solve
With CPU VIN = Auto, BIOS shows 1.838V.
Set CPU VIN to 1.9V, BIOS shows 1.937V (after restart).
Drop VCore to 1.170 to decrease core temp/power since 1.170->1.175 did not help.
Rerun test.
 
Prime95 40K, in-place FFT, 5 minute per test, 30 minute run (manually stopped)
 

 
Back to constant temp/power draw inferred as constant performance.
Overall stability unknown, but passes this test.
 
Refer
http://www.overclock.net/...hantom-throttling.html
post edited by geninfo - 2018/02/06 20:00:58

Attached Image(s)

#50
GGTV-Jon
FTW Member
  • Total Posts : 1811
  • Reward points : 0
  • Joined: 2017/11/25 14:11:43
  • Location: WA, USA
  • Status: offline
  • Ribbons : 18
Re: X299 Dark Bench Notes/Info 2018/02/04 14:43:37 (permalink)
At what frequency was the CPU during all of that?
The package wattage - is that for all cores or per core?
 
I looked at your overclock.net link, mention of load line calibration made me think of this video - throwing this out there for those that have net seen it
https://youtu.be/Z8nFdFpuVBg
 
post edited by GGTV-Jon - 2018/02/04 14:56:24
#51
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/04 18:37:40 (permalink)
GGTV-Jon
At what frequency was the CPU during all of that?
The package wattage - is that for all cores or per core?
 
I looked at your overclock.net link, mention of load line calibration made me think of this video - throwing this out there for those that have net seen it
https://youtu.be/Z8nFdFpuVBg



Thanks, good video!
 
4.5Ghz.
 
Package wattage neither all core nor per (18) core.
 
Roughly:
15W Package Power = ~150W Wall
50W Package Power = ~550W Wall
 
I interpret as relative CPU power unit to compare across tests.
 
Have not yet got 32X Mesh Ratio stable.  Found reproducible crash at 128K Prime95 FFT in-place.
Pulling too much power?  Goes to 675W from wall (59W Package Power).  Amazed cooler can hold temp at 87C.
Kingpin guide mentions high Mesh Ratio sensitive to CPU temperature.  Need direct die. :p
post edited by geninfo - 2018/02/04 18:41:54
#52
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/05 19:36:29 (permalink)
Working mesh ratio.
Saw interesting image.
 
Prime95 filling 64GB of RAM in 45 seconds

 
Prime95 test settings:
Custom
Threads = 36 (2 * cores)
Min FFT = 4096
Max FFT = 4096
Run FFTs in-place = Unchecked
Use Memory = 60,000
Time to run each size = 5
 
Must throw lots of data around mesh network between cores and memory controllers?
60,000 instead of 64,000 to prevent SSD memory paging since OS using 2-3GB RAM.
 
What can fill RAM faster or test mesh ratio better than Prime95?
post edited by geninfo - 2018/02/05 23:01:44

Attached Image(s)

#53
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/06 11:07:43 (permalink)
Found two potentially stable mesh ratio overclocks.
 
All other settings see first post.
 
Mesh Ratio = 31x
VIN = 1.90V
VMesh = 1.29V Adaptive
VSA = Auto
VCCIO = 1.19V
 
Mesh Ratio = 32x
VIN = 1.91V
VMesh = 1.29V Adaptive
VSA = 1.05V
VCCIO = 1.19V

30x->31x
Very minor improvement in Aida64 Engineer trial memory/cache tests.
+ ~100 (~0.8%) in Time Spy CPU score.
 
31x->32x
Very minor improvement in Aida64 Engineer trial memory/cache tests.
+ ~110 (~0.9%) increase in Time Spy CPU score.
 
Note
Time Spy comparisons by average of three runs at each mesh ratio.
AIDIA64 memory latency decrease ~1.25ns from 30x to 32x.
 
Both pass Prime95 128K in-place FFT (45 minutes) and 4096K moving FFT (15 minutes).
 
32x passed Prime95 tests at VIN 1.90, but crashed in Time Spy until VIN 1.91.
Increased VSA voltage taking from core/mesh voltage?
Fixed by more voltage to CPU via VIN?
 
Quad-channel RAM with 16GB DIMMs driving VCCIO?
Need 1.116V+ VCCIO for any stable configuration.
 
EVGA BIOS save/load profile feature very useful for changing overclocks!
#54
Myersjames93
Superclocked Member
  • Total Posts : 131
  • Reward points : 0
  • Joined: 2016/06/09 19:15:38
  • Status: offline
  • Ribbons : 0
Re: X299 Dark Bench Notes/Info 2018/02/06 19:20:37 (permalink)
Whats the VRM temps
#55
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/06 19:58:20 (permalink)
Myersjames93
Whats the VRM temps



HWiNFO64 records "VR T1" sensor.  VRM temp?
From 47C to 58C under heavy Prime95 load in 32x Mesh Ratio config above.
VRM heat sink on, heat sink fans running default fan curve.
 
Prime95 128K in-place FFT 5 minute per test cycle (46 Minutes, Manually Stopped)
 

 
post edited by geninfo - 2018/02/09 03:02:13

Attached Image(s)

#56
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/08 09:04:13 (permalink)
Check Windows Event logs Viewer for WHEA CPU internal parity errors indicating overclock instability.
 
Theory
Hardware reports errors to Windows Hardware Error Architecture.
Hardware reports corrected errors (recoverable) or uncorrected errors (not).
WHEA errors in Windows Event Viewer logs.
HWiNFO has WHEA error count sensor.
Skylake-X cores detect internal parity errors.
Parity error means data in a register or cache did not get stored correctly?
Skylake-X cores fix some internal parity errors.
Fixing data before processing prevents possible core crash or memory corruption.
Ideal no parity errors.
Parity errors indicate slightly unstable overclock.
Increase VCore or lower CPU frequency.
Less common, reduce memory overclock?
 
Example
Overclocking memory, first to 3400.
All Prime95 spot tests pass.
Prime95 stability test fails at 254 minutes.
Cannot replicate Prime95 stability failure with spot tests at failing sizes (1280K->56K).
RealBench, Aida64, Time Spy, Fire Strike, Cinebench all pass.
 
Notice Time Spy demo sometimes crashes.
Time Spy application error in Windows Event log, exception code 5.
Exception code 5 means access violation - Time Spy attempts to read wrong memory?
Time Spy demo crashing more often with memory overclock?
Memory overclock at fault?
Boost DIMM voltage, VCIN, memory frequency (to 3466, in case better ratios).
Still 1 out of 10 crash on Time Spy demo.
 
Review old test data and Windows Event Viewer log:
Time Spy demo application errors in Windows Event Viewer log before memory overclock.
Update to 1.06 BIOS (Spectre fix) four days before start of parity errors.
From 0 to 5 WHEA parity errors each day.
Time Spy application errors usually within minutes of WHEA parity errors!
Parity errors start when shifting from 44x to 45x core multiplier.
 
So:
Probably not just memory overclock - WHEA and Time Spy application errors started before.
Probably not Spectre fix - WHEA errors not immediately after.
Probably WHEA parity errors and Time Spy Demo crash related - times close.
Insufficient VCore for 45x core multiplier?
 
WHEA error usually appears by 5th Time Spy demo run after restart
Start Vcore +0.001 and run 5 Time Spy demos.
1.174Vcore, Time Spy demo stops crashing, still WHEA errors.
1.177Vcore, one Time Spy demo crash; still WHEA errors.
1.180Vcore, no Time Spy demo crash; still WHEA errors. :(
Drop frequency to 44x, no Time Spy demo crash; no WHEA errors; 10x Time Spy demo runs.
Frequency back to 45X, 1.90Vcore, Time Spy demo crash; still WHEA errors. :(!
1.120VCore, no Time Spy demo crashes; no WHEA Errors; 10x Time Spy demo runs! :)
1.195VCore, no Time Spy demo crashes; no WHEA errors; 10x Time Spy demo runs.
Prime95 56K in-place FFT 5 minute test cycle comparison temps +5C (91C @ 16 minutes).

Config
From 32x Mesh Ratio config above:
VIN: 1.92
VCore: 1.1195 (can go down to 1.1191?)
VDIMM1: 1.375
VDIMM2: 1.386
Fast Boot: Off
Force Memory Retraining: Enabled
 
VDIMM2 different because BIOS actual voltage lower than VDIMM2 set in BIOS.
Those VDIMM1/VDIMM2 settings create same BIOS actual voltage for both DIMM banks.
Fast Boot off, retraining on in case lack of memory training problematic.
 
VIN and VDIMM increase while troubleshooting WHEA errors not needed?
 
May go back:
VIN: 1.91
VDIMM1: 1.35
VDIMM2: 1.35
Fast Boot: On
Force Memory Retraining: Disabled
 
Tip
Create two custom views in Windows Event Viewer.
WHEA = All events by source WHEA-Logger.
Application Error = All events by source Application-Error.
Check frequently when overclocking.
 
Refer
WHEA
WHEA Overclock Partial Instability 
WHEA Error Overclock Instability Full Discussion 
WHEA Errors And CPU Frequency
WHEA Errors Caused By Spectre Fix? 
Intel WHEA Error Decoding
Intel WHEA Error Discussion & Diagnostic
post edited by geninfo - 2018/02/09 10:13:21
#57
GGTV-Jon
FTW Member
  • Total Posts : 1811
  • Reward points : 0
  • Joined: 2017/11/25 14:11:43
  • Location: WA, USA
  • Status: offline
  • Ribbons : 18
Re: X299 Dark Bench Notes/Info 2018/02/08 23:47:20 (permalink)
LOL, Good work, and a hell of a lot of stuff to read - TY
 
CPU is now in hand - just waiting on picking up the rest of the stuff....
#58
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/09 05:22:14 (permalink)
Check stress test, benchmark, or app crash with BIOS defaults (stock) settings.
Furmark crash on results submit probably not overclock problem.
 
After WHEA adjustment above, Furmark crashes with access violations after 4-5 passes and submits.
4.5Ghz still not stable or memory overclock problem?
Vcore raise to 1.20V; still crashes.
BIOS defaults (stock); still crashes.
Stock confirm saved time adjusting Vcore or memory clock further.
 
More info at Furmark forum:
http://www.geeks3d.com/fo...9.msg6976.html#msg6976
#59
geninfo
Superclocked Member
  • Total Posts : 191
  • Reward points : 0
  • Joined: 2017/12/18 11:22:15
  • Status: offline
  • Ribbons : 2
Re: X299 Dark Bench Notes/Info 2018/02/09 07:18:37 (permalink)
bcavnaugh
Not bad and Great Scores
 



bcavnaugh, more benchmarks for you before GPU change.
 
-> indicates improvements from Mesh Ratio, memory, and GPU overclocking.
 
Config
Candidate final stable config as in WHEA post above.
Candidate final stable GPU overclock +75/+428.
 
Benchmark
Time Spy 10772-> 11188
Fire Strike 27011-> 28078
Furmark 720 Bench 16129-> 17017
Furmark 2160 Bench 7043 -> 7516
Cinebench R15 CPU 4515
RealBench version 2.43 299370 (image 208183, encode 642498, opencl 119718, multitask 322857)
AIDA64 Engineer Trial version 5.95.4500 (see screens below)
 
Note
Time Spy/Fire Strike in top 50/60 3DMark results for single 1080 Ti including LN2 and Kingpin results.

RealBench score would #4 leaderboard (defunct?) and top for single 1080 Ti against Apex/Extreme boards.
RealBench image editing test does not launch on first RealBench test after restart, but no errors.
RealBench image editing test runs fine if RealBench stress test run briefly once.
 Impatient.  Wait several minutes for gimp to show first RealBench run.  Discard first result.
 
Furmark test version 1.19.0.1 -> 1.20.0.1.
3DMark version 2.4.4180.
 
AIDA64 Cache & Memory 

 
AIDA64 GPGPU
 
post edited by geninfo - 2018/02/09 16:41:08

Attached Image(s)

#60
Page: < 1234 > Showing page 2 of 4
Jump to:
  • Back to Mobile