Are OCCT Errors BS?

[OGKRG]

48mins into testing.  No BSOD's. I have it set to stop testing on first ERROR. 

[porina]

You can have computation errors that don't result in something as severe or noticeable as a BSOD, lockup or reboot. I'm not so familiar with OCCT but have used Prime95 for similar testing, as I have more control over that to identify where the problem might be.

[OGKRG]

13 minutes ago, porina said:

You can have computation errors that don't result in something as severe or noticeable as a BSOD, lockup or reboot. I'm not so familiar with OCCT but have used Prime95 for similar testing, as I have more control over that to identify where the problem might be.

Does Prime 95 show you what the error is? 

 

I have done some back to back testing with OCCT to identify if errors are flaws in the programs code.

 

1st test was throwing errors back to back at different times frames under an hour.

4.1GHz @ any voltage less than +0.102v which is less than 1.2v give or take.
 

2nd test was Stable 2-3 times over 1 hour length.

4.1GHz @ + 0.102v
Min - 1.164v
Max - 1.224v
Avg - 1.195v

 

Testing today.

4.3GHz @ +0.174v (Failed) (48:21 - Error) 1.29v-1.3

 

Current testing is 4.3GHz @ +0.204v, Max voltage showing is 1.33v, 20mins in no errors.

 

I think OCCT is pretty accurate. I only test Large Data Sets, Extreme, Auto Threads, AVX2. I heard Small Sets test cache more, but Large Sets test the Die itself more. Need some verification on this. Large Sets are closer to real world usage maybe. 

 

Also what do you think max voltage safe for this chip is? I have a Ryzen 5 3600. Precision boost is off of course. With it on, I was getting 4.2GHz at around 1.4v max give or take higher or lower at times. That is way to high. Obviously good points of OCing this chip is figuring out if I can undervolt and also achieve higher than stock boost speeds, which so far shows a favorable outcome.

[Alex Atkin UK]

2 minutes ago, OGKRG said:

I think OCCT is pretty accurate. I only test Large Data Sets, Extreme, Auto Threads, AVX2. I heard Small Sets test cache more, but Large Sets test the Die itself more. Need some verification on this. Large Sets are closer to real world usage maybe. 

Bear in mind that errors that do not cause a crash, may be invisibly causing data corruption.  So I'd take all errors seriously.

[OGKRG]

2 minutes ago, Alex Atkin UK said:

Bear in mind that errors that do not cause a crash, may be invisibly causing data corruption.  So I'd take all errors seriously.

I also had this paranoia, also hearing that maybe errors due to calculation errors could result in under performing and frame drops in games. I love overclocking, but I also love gaming. It's a double edged sword lol. 

 

And yo! You're a brony? Have you ever heard of SMOR3S? the musician brony?

[porina]

17 minutes ago, OGKRG said:

Does Prime 95 show you what the error is? 

No but I understand how to read the output to help point me into the likely areas, combined with other selective testing.

 

17 minutes ago, OGKRG said:

I only test Large Data Sets, Extreme, Auto Threads, AVX2. I heard Small Sets test cache more, but Large Sets test the Die itself more. Need some verification on this. Large Sets are closer to real world usage maybe. 

I wonder if this is similar to Prime95 FFT size. There at small FFT sizes data can fit in CPU caches, and thus stress the cores more. Bigger tasks spill over into ram. Due to the much lower performance of ram this is less stress on the cores, but will stress the IMC and ram.

 

My impression having previously owned Zen 2 CPUs is they're tuned very close to the limit as sold. For practical purposes I wouldn't bother to under-volt them if you want stability in all use cases. Only safe overclocking is relaxing power limit, in other words PBO. The power usage varies too much for old school fixed clock/'voltage style OC unless you don't care about stability in all situations.

[OGKRG]

5 hours ago, porina said:

No but I understand how to read the output to help point me into the likely areas, combined with other selective testing.

 

I wonder if this is similar to Prime95 FFT size. There at small FFT sizes data can fit in CPU caches, and thus stress the cores more. Bigger tasks spill over into ram. Due to the much lower performance of ram this is less stress on the cores, but will stress the IMC and ram.

 

My impression having previously owned Zen 2 CPUs is they're tuned very close to the limit as sold. For practical purposes I wouldn't bother to under-volt them if you want stability in all use cases. Only safe overclocking is relaxing power limit, in other words PBO. The power usage varies too much for old school fixed clock/'voltage style OC unless you don't care about stability in all situations.

 

TL;DR - Prime 95 is cool, performance boost sucks, OC 4 LIFE! 

 

Blend mode on Prime 95 will stress both CPU and RAM. Don't get me wrong I like it, and have used it on and off, the worker windows are hideous and the program needs and overhaul. (I merge all worker windows into one lol.) I just stick to OCCT and RealBench/Cinebench and have had good translations with overclocks. OCCT errors do seem real. I am at 4.4GHz 40mins stable *fingers crossed* @ 1.38v-1.40v an AIO would be nice, since I am reaching max temps but still have some headroom with the wraith prism, def an upgrade from the stealth for sure. I usually just do Large data sets with AVX2 on extreme. I don't see the point of Small data sets for the extreme heats, same with Linpack Extreme, just seems a bit unrealistic. I like RealBench stress test for 1 hour after I pass stable on OCCT for an hour as icing on the cake, and Cinebench to see performance numbers. I like running benchmarks after, couple times over to make sure I am not degrading performance.

 

Is stock performance boost good? It's hard to say. If you want an out of the box solution it'll work, but you still have to worry about thermal throttling. My 3600 was running 4.2GHz boost @ around 1.45v+ I am getting 200MHz higher with a little less voltage. It's not a huge increase like my 7700K where I was hitting 4.9GHz @ 1.4v-1.5v but it's an accomplishment, and I like ocing, and I hate just slapping a CPU in my computer and not being able to really call it my own. I think the VRMs on the AORUS M aren't to bad either. The BIOS are really weird for adjustments though, you would have to see it to believe it lol. I'ma see how far I can take this bad boy though. It def needs to be cooler to hit over 4.6GHz I can prob get 4.5GHz if I am lucky on the same voltage or a little higher on air.