Advanced search

Message boards : Graphics cards (GPUs) : GPU overvolting danger

Author Message
Keegan
Send message
Joined: 25 Feb 14
Posts: 15
Credit: 53,918,886
RAC: 0
Level
Thr
Scientific publications
watwatwatwat
Message 42588 - Posted: 10 Jan 2016 | 20:09:09 UTC

To what extent would I be degrading the life of a 780Ti by overvolting it?

In my system I have 2 780Ti graphics cards, both by evga, the first one and top card is a super clocked model which I could push to 1256Mhz core and +84Mhz memory stably; The second and lower card was a normal, but used card I had added for extra power when I got a 4k monitor. When putting them in sli config, their clocks change to 979 and 1006Mhz. So I tried putting setting the offset to +140 core and +84 memory (the highest stable I had on the original) and when benchmarking, the top card rarely touches the 106.3% power limit or voltage limit and keeps it's temp in the low 70s; the bottom card can handle the memory offset fine but the core offset is to much for it, it stays around 2 degrees cooler, fans about 8% higher, and regularly hits the power limit and voltage limit and is not very stable. (using most recent precisionX)

As for cooling I am using the reference style blower cards, power supply intakes air from the bottom and out the back, my cpu uses an h80i also set up to push air out the back, I have about 12% room to play with on the cards fans.

It will be probably 2 or 3 years before I can upgrade again, so in that time how high could I reasonably turn up the voltage? would I be better of setting a more aggressive fan profile? would it be very advantageous to create a profile with different overclocks on each card to use for crunching with sli disabled for when I'm not planning on gaming? How significant is the power draw on the cards lights?

Also, how bad is it for the server for a workunit to fail? unfortunately, the only way to know if it's stable on gpugrid is to test it.

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2076
Credit: 15,113,965,076
RAC: 4,954,502
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 42589 - Posted: 10 Jan 2016 | 22:39:59 UTC - in response to Message 42588.

To what extent would I be degrading the life of a 780Ti by overvolting it?
Much of the degradation comes from the temperature rise caused by the overvolting.
Thus if you can keep the temperature of your cards as it was before overvolting them, their lifetime won't be significantly shorter.
Take into consideration, that the power draw of the GPU is in direct ratio of its frequency multiplied by its voltage squared. So 10% rise in voltage causes 21% rise in power draw. (1.1*1.1=1.21) If you rise the frequency by 10% as well, the total power draw increase will be 33.1%.

As for cooling I am using the reference style blower cards,
It limits your possibilities. Severe overvolting is only possible on water (or more exotic) cooling.

I have about 12% room to play with on the cards fans.
That wouldn't improve cooling too much.

Would I be better of setting a more aggressive fan profile?
Yes.

would it be very advantageous to create a profile with different overclocks on each card to use for crunching with sli disabled for when I'm not planning on gaming?
Yes.

How significant is the power draw on the cards lights?
It's negligible.

Also, how bad is it for the server for a workunit to fail? unfortunately, the only way to know if it's stable on gpugrid is to test it.
It's no problem, as the failed workunits get reassigned to another host. It's worse for you, as you loose the time it took to process in vain. If it happens too often, the overclocking is counterproductive. You should also know that different batches tax the GPU to different extent, thus some batches tolerate less overclocking. (for example the GERARD_A2AR)

Keegan
Send message
Joined: 25 Feb 14
Posts: 15
Credit: 53,918,886
RAC: 0
Level
Thr
Scientific publications
watwatwatwat
Message 42592 - Posted: 12 Jan 2016 | 18:20:17 UTC

I hope you don't mind if I throw a few more batches of questions your way.

When you say severe overvolting, do you mean outside the limits of precisionX (75mV)?

Also, I've found that the cards will reach their max boost speeds (not overclocked) independent of each other when in sli and running the gpugrid application only if I have the power target maxed.
____________

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2076
Credit: 15,113,965,076
RAC: 4,954,502
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 42595 - Posted: 12 Jan 2016 | 23:18:52 UTC - in response to Message 42592.
Last modified: 12 Jan 2016 | 23:22:10 UTC

When you say severe overvolting, do you mean outside the limits of precisionX (75mV)?
Yes.

Also, I've found that the cards will reach their max boost speeds (not overclocked) independent of each other when in sli and running the gpugrid application only if I have the power target maxed.
SLI is mostly a risk factor in GPU crunching, as the two GPUs are working on their own task, not like when they rendering the next frame of a game together.

Keegan
Send message
Joined: 25 Feb 14
Posts: 15
Credit: 53,918,886
RAC: 0
Level
Thr
Scientific publications
watwatwatwat
Message 42797 - Posted: 10 Feb 2016 | 19:21:49 UTC

Turns out that the superclocked card had a faulty vrm and in retrospect I should have noticed this a long time ago, but because nothing crashed and the clocks would reset themselves after restarting the computer I figured that it was an overclocking issue. When I realized that the same thing was happening with stock clocks, I wrote the company and they told me to send it back.

To my surprise, the good people at evga replaced it with a 980 superclocked with acx cooler.

KSUMatt
Avatar
Send message
Joined: 11 Jan 13
Posts: 216
Credit: 832,764,443
RAC: 0
Level
Glu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 42798 - Posted: 10 Feb 2016 | 20:14:32 UTC - in response to Message 42797.

To my surprise, the good people at evga replaced it with a 980 superclocked with acx cooler.


That's why I only buy EVGA. I've never had anything but good experiences dealing with them. Their customer service tends to be excellent.

ATF
Avatar
Send message
Joined: 5 Jan 12
Posts: 117
Credit: 40,721,760
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwat
Message 43073 - Posted: 24 Mar 2016 | 10:39:06 UTC - in response to Message 42798.

Damn daniel! I was going to recommend taking off the stock fan and applying some 20 dollar thermal paste that isnt electrically conductive so it has lower chances of frying your system. Then you got new gear, so I wouldnt really risk it. Anyways, I know the old 570s I used to get I was unable to remove the fans cause the screws seemed saudered even tho they were philips head. But new fans, add some thermal paste, I would say out of limited experience to just the chip itself, drop your gpu temps, fastest way to do it in my opinion. I always buy gpus with 3 fans.

Post to thread

Message boards : Graphics cards (GPUs) : GPU overvolting danger