Aida 64 will pass, RealBench will not pass

[Hawkstorm]

@pleot Not from my own experience the temps in OCCT are accurate, but the voltages etc may not be as it doesn't report Haswell-E correctly in this version, that'll prolly be version 5.

And running 2 sensors side by side is a good way to get incorrect readings, only use one.

[Pleot]

While AIDA and RealBench passed 8 hours on my reference 4300 hz OC, with OCCT (ver 4.1.1) I had to pump up the voltages on Vcore from 1.193750 to 1.25 to be able to pass 1 hour 30 min with OCCT (now I will try to get it as low as possible)

Real world may not be enough because a problem can occur at the most inappropriate time, and if one is about to break the record on the game and win the championship and the computer BSODs due to bad OC right on the finish line it is very bad

But it is frustrating having AIDA 64 passing 9 hours and OCCT and RealBench saying something is not right, so I tend to stick to the more demanding, for me OCCT

But I had to disable temps warnings on OCCT because the are not accurate, genarally I only use the OC pannel display to check temps...

[Menthol]

As long as you don't over ride the on board OV, OC, protection in the bios theoretically your CPU will either down-clock from high temps or the motherboard protection will shut down or restart your computer

[Qwinn]

Have you done anything to input voltage, aka vccin? If not, try setting it at 1.9v which is very safe (even 1.95 is generally safe), it can increase stability significantly and let you lower your vcore voltage a little or even possibly increase ratio a tick.

[Pleot]

I have tried that some time ago and it did improve stability a lot, cant remember the exact proportion. But was afraid it could damage my board or CPU so I lowered it from 1.9 back to 1.85 for safety reasons. Are you sure it is that safe like you say?

Right now Im on the lowering vcore path, I will find the limit between 1.25 and 1.2. When I do I will have a rock solid OC on the 4300 hz on a 5930k (I mean pass on all the stress tests being OCCT the harder).

Afterwards I will attemp to get to the 4500. I had the 4500 running 9 hours AIDA but Realbench said no go on the first 15 min... Not even with 1.3 volts on the core.

Are you sure 1.9 is safe enough for the input voltage? (will have to test if it is efective anyway)

[Qwinn]

From everything I have read, and I've read damn near every thread on these topics I've found including threads over 300 pages long on overclock.net, even 1.95v is considered completely safe. If anyone disagrees (or agrees), please chip in, cause if I'm wrong I'd really like to know.

One thing I've read a few times is that setting input voltage to 1.9v is the *first* thing you should do when you begin trying to find a stable overclock. I've also read that once you find a stable overclock, you should try swapping higher input voltage for lower vcore until you reach stability at a lower vcore or your input voltage reaches 1.95v.

Note that the way your input voltage actually behaves under load will be affected by your Load Line Calibration (LLC) setting. I'm running at LLC 7. When I was trying to get stable at 4.5Ghz at an acceptable 24/7 voltage/temperatures (which ultimately proved unsuccessful), I wasn't able to lower the LLC at all from it's Auto setting, which I believe generally runs at LLC 9.

[Pleot]

So I dont understand if LLC values are better higher or lower - llc7 better than llc9?

I have very acceptable temps and voltages for an 4300 hz OC on corsair 3300 hz ram (running ram at 2800 very tight timmings that prove to be as fast as 3300 defaults)
105 bclk ratio 41
ram 2799 14 14 14 33 1t
vcore was stabe at 1.193750 on aida and RB but OCCT is going to make me go between 1.25 and 1.2 (still testing)
Dram voltage 1.405 just to boot
Eventual 1,290 to run ram after boot (this is actually not bad value for the ram voltage)

input voltage 1.850
System agent 1.05

I will try increasing input voltage when I go for the 4500 hz later

Been at 4500 hz 9 hours AIDA with values not much different form these, but RB said no go
Anyway its not bad for a PC even if it stays like this

[JackNaylorPE]

I stopped using synthetic tests ... and saved myself a lot of time. I find that I can be AIDA 64, Prime95, Intel XTU stable and still fail in RoG Real Bench. When OC'ing, I run the 4th (multitasking) test and if it passes, I know I am perty darn close. Eliminates many, many hours as I would detect unstable Ocs in 2 minutes rather than 40+ on a synthetic.

And when ya think about it, did you build ya PC to run programs and / or games or did ya build it to run synthetics ? Not to say that RB is the be all and end all of testing; I still find that I can pass all synthetics and RB; run all my CAD programs w/o issue and system will still crash in certain games . And yes, it's the same games on every box. Use the profiles BIOS feature to set up profiles for:

Stock (for troubleshooting)
Stock w/ XMP (for troubleshooting)
24/7/365 OC
Gaming OC (a notch higher than above)
Battlefield * (a notch significantly lower than 24/7/365)

* and a few others

I only have time to play 1-2 games a year but my son is a pilot and uses box for flight sims and various games. Most times, it's on the 24/7 profile tho I do use the Gaming OC once and a while just to see what extra I get.

[Pleot]

I completely agree, stress tests seem to have a life of their own and not agree to what stability is , and if demanding programs and demanding real life processing routines dont fail why give so much credit to stressing - Ive found an argument around that is respectfull enough, seems like if the machine is not stable enough it will alocate a lot of errors on system files that will become drastic over time and render the SO useless ,

So my ideal is to try passing as many stresses as possible until I reach a point of satisfation - then will simply leave it there forever and not worry with OC anymore

[Dr. Zchivago]

will simply leave it there forever and not worry with OC anymore

Unless, you get the temperature-tweaking bug.

Z