Why factory GPU OC are so conservative?

[lucashuber]

Hello! My doubt is pretty auto explicative, why factory GPU OC are so conservative?!!

I'm considering this cause i just got a stock PNY GTX 1060 3GB (with some decent cooling system as u can see). So i decided to research some pre-studied OC made by factories and overclockers... After some tests i was amazed by somethings:

1) My stock GPU clock was much higher than the one listed in PNY site despite i didn't make any changes (1834Mh on my tests vs 1704 stock setting)

2) 1834 Mhz is higher than almost all factory OC i've seem and almost the same as the biggest factory OC i've found - EVGA GTX 1060 FTW+ 3Gb has a 1860Mhz OC ).

3) Some simple tests i did were able to put my GTX 1060 to work at 2025Mhz GPU clock and 4400Mhz of memory (actually 8800Mhz), using all power and voltage stock settings. With this config, my power limit didn't pass 94,5% of its maximum. (this is less than most OC i've saw in internet, that were able to push core clock till 2060 Mhz and memory clock till 9000 Mhz only with 108% max power lvl)

4) Almost none factory OC i've found had memory OC. Almost all of them presents 8008Mhz clock, and one model from Gigabyte uses 8100Mhz.

5) IF GPU overclock is safe as users use to say (if u don't mess up with core voltage), why factories are so carefull? I've made a very simple OC considering the factories OC above stock settings (around +100Mhz for EVGA core clocks, and +100Mhz for the gigabyte memory clock) and even using this settings, my gpu didnt pass the 90% maximum power limit. Besides that, my GTX only have a simple 6 pin power connector, while most factory OC GTX 1060 use 8pin connectors.

That's all!

TY!!!!

 

[photonboy]

It's quite simple.
Every GPU (the processor) and memory chip is different thus there's a silicon lottery to how high the chip can go.

The closer they get to the theoretical maximum frequency the higher the statistical chance of FAILURE the average card will have.

So you might go from say 2% failure (due to GPU) to 10%.

You also reduce the LIFE of the card even if it works initially. For most this won't matter but again you are still going to have more cards fail within the Warranty period the closer you push towards the theoretical maximum.

(there's also NOISE, power components, binning of GPU's between card tiers that changes the cost... some GPU's for example may fail above 1900MHz so are put into cards of lower frequency)

 

[k1114]

Those higher factory oc will also turbo higher so you can't compare your actual turbo vs their non actual turbo speeds. If you want an explanation, I forget why they state lower turbo speeds when they actually go higher but it's easy to google.

Besides the lottery as mentioned, it's also better stability in different uses/situations. They also don't want to be running at the limit as overtime, they won't be able to run stable at those clocks and would need to run lower. Then you'd have lots of complaints of "failing" components. Tying in with reducing life, that max will also drop quicker running at the limit. Memory ocing is even more of a lottery so will be more conservative.

The 8 vs 6 pin power connector allows for more headroom for power and more stable power. Again stability being a reason.