Overclock.net › Forums › Intel › Intel - General › Devil's canyon relationship with RAM timings/speed and voltage.
New Posts  All Forums:Forum Nav:

Devil's canyon relationship with RAM timings/speed and voltage.

post #1 of 2
Thread Starter 
I've been for an entire week, 24/7 stress testing my system each half day with different settings, I want to get the lowest possible timings on the highest possible frequency on my RAM with 4790K OCEd to 4.4GHz and cache to 4.2GHz. (I would raise that multiplier if I could but with these RAM Speeds and timings it doesn't seem possible)

I could OC my CPU to 4.7GHz with 1.27v on the cores, which is pretty good I think, this with a 1333MHz RAM, 24 hours Prime95 stable.

Now I installed my G.Skill 2400 C10 through XMP with all primary timings and even secondary timings being correctly in place, after this I couldn't even get 4.5GHz under the same voltage, not even 4.4GHz for some reason, not even at 1.3v, I'd always get rounding error on Prime95 so I decided it was the memory controller, raised I/O and SA voltage to 1.2v and still had the same issue.

This is when I decided it wasn't worth weeks of Priming95 for days and I lowered the RAM speed to 2200MHz by using BCLK:DRAM ratio set to 100:100 and then picking 2200MHz RAM speed, this is when I started overclocking the RAM in a different manner, timings.

See, now my question is, do tightening the timings actually increase the need of extra voltage for CPU Cores and cache? I set it up to 2200MHz 9-11-11-20-1T-tRFC 230 and could reach 100% stability on Memtest after 10 entire passes on my 16GB dual channel with the following settings.



9-11-11-20-1T and 230 tFRC.

This is when I ran another 24 hours of Prime95 with my CPU set to 4.4GHz with Core voltage of 1.18v and Cache set to 4.2GHz with Cache Voltage set to 1.14v with custom settings on Prime95 going from 8K FFT Size to 4096k with In-Place FFTs and 15 minutes length, making sure that in 21 hours Prime95 would test every single FFT Size available in the software for 15 minutes each for absolute 100% stability, the test came out stable, not a single error after 24 hours.

The thing is, in for a penny in for a pound.

I started changing every single one of the over 20 different 2ndary and third timings, each and every single one of them to the lowest possible value where Memtest86+ would complete a pass, this took me around 2 days, after that I thought it was enough and I got at least 15 different 2ndary and third timings to the lowest possible value (For each single one of them as they were individually tested) after tweaking all these timings I ran a 24 hours Memtest86+ to ensure the stability of all those tweaked timings combined, 20 Passes later it was all good and stable, Diamond Solid Stable.

The thing is, I went back to Windows with these new RAM settings and I decided to run Prime95 again with the same Prime95 settings (Thank God I did because I can be OCD with this) just as before for 24 hour of testing just to be absolutely sure, problem is that Prime95 along with my PC BSODed in 1 hour after running the test with my already proven 100% 24 hours Prime95 stable 4.4GHz 1.18v core and 4.2GHz 1.14v cache, I was obviously baffled as to what was happening, this was my absolute Diamond Solid CPU overclock and it was crashing, I couldn't even blame the new RAM timings because it passed 24 hours of MEMTest86+ proving to be stable as a Diamond, also Prime95 gives rounding errors instead of BSODs when there are RAM issues, also this BSOD was "WHEA_UNCORRECTABLE_ERROR" which I'm sure can't be attributed to the RAM... So everything pointed out to the CPU, my Diamond Stable CPU, for some reason... "IT MAKES NO SENSE" I thought to my self.

This is when I thought that MAYBE tightening the RAM timings to such an EXTREME extent (2 entire days of testing each value to the lowest possible number, individually, for a total of 15-17 different timings) could in turn somehow increase the need for Core Voltage AND/OR Cache Voltage for my 4.4GHz CORE and 4.2GHz cache on my 4790K.

Look and behold, following my suspicious I raised the core voltage from 1.18v... (1.18v the value I used which was diamond solid stable 24 hours Prime95 before I tightened the RAM timings), to 1.19v after tightening up every single 2ndary and some third RAM timings and BAAM, I could hold Prime95 for 4 hours before crashing instead of one!

This is when I raised the Cache voltage from 1.14v to 1.16v while keeping the 1.19v new voltage for my core (Core that once again was 100% 24 hours Prime95 stable with 1.18v before tightening up the RAM timings) and now I'm 8 hours in Prime95 with no errors and still running while writing this and will leave it overnight.

CONCLUSION:

Scientifically speaking, tightening DRAM latency on all of my tests (At least extremely tight timings like I did) made my VCORE and CACHE need for voltage increase, I want to know if there's information about this, papers, explanations, WHY.

Any insight would be appreciated, Thank you.
Edited by Dargonplay - 7/15/16 at 7:23pm
post #2 of 2
...only two days. PFFFFT!

"He who tries the most goes the fastest." Onepagebook.

Keep working.
Refresh rate and Maximum Asynchronous Latency are the keys to stability with low latencies.

No, there's no place to read about this.
Most people don't share, as the people that have figured this out are the people getting hand picked ran from the suppliers and using it to make money in benching.
My System
(15 items)
 
  
CPUMotherboardGraphicsRAM
FX6300 Black M5A99X EVO R2.0 Nvidia GTS450 Team Vulcan PC3 12800 
Hard DriveOptical DriveCoolingOS
Samsung 840 PRO Asus DRW-1608P (x2) Custom Water Cooling Win7 (Ult), Win 8.1 & Win Server 2012 R2 
MonitorKeyboardPowerCase
2 X Samsung 915N Ducky Shine III, Blue Cherry/Blue LEDs PCP&C 1kw Lian Li PC-71 (W/Window) 
MouseAudio
Logiteck G400s none 
  hide details  
Reply
My System
(15 items)
 
  
CPUMotherboardGraphicsRAM
FX6300 Black M5A99X EVO R2.0 Nvidia GTS450 Team Vulcan PC3 12800 
Hard DriveOptical DriveCoolingOS
Samsung 840 PRO Asus DRW-1608P (x2) Custom Water Cooling Win7 (Ult), Win 8.1 & Win Server 2012 R2 
MonitorKeyboardPowerCase
2 X Samsung 915N Ducky Shine III, Blue Cherry/Blue LEDs PCP&C 1kw Lian Li PC-71 (W/Window) 
MouseAudio
Logiteck G400s none 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Intel - General
Overclock.net › Forums › Intel › Intel - General › Devil's canyon relationship with RAM timings/speed and voltage.