Originally Posted by HyperMatrix
Hey. I just faced a similar panic. I posted on the llc thread about it but it is very relevant to what you're saying. Here is a copy/paste. Does this sound like what happened to you?
Guys...just had an incredibly scary moment after using this mod. I have to share. I'm running 3 Titans in SLI. First couple of days, I had a lot of fun with this. Was already using the 144% power limit bios. Played around with LLC enabled and disabled. I couldn't max them out at 1.325v or even 1.3v without my 1250W PSU completely dying. So I decided I'd get a NEX1500 as an upgrade in a few days. Anyway I set it up at around 1.26v-1.27v, and average of 1200MHz across the board, and memory at 7500MHz. It worked beautifully after I lowered the voltage and clocks so my games wouldn't shut down. Played some Crysis 3 amongst others, and really roasted the cards for hours without any crashes.
Today, I was working on setting up Wake On Lan for my PC so I could remotely power it on using Splashtop on my iPad/iPhone. After a lot of issues thanks to the Maximus V Extreme needing you to set the network stack as your primary boot device for Wake On Lan from Ethernet to function (which takes an extra minute or two before switching to boot option 2, the SSD) I gave up. I reverted bios back to previous saved settings. Booted in to Win8.1. Decided to play some BF3, which isn't really a heavy workload for Titans. 5 seconds into the map loading, game froze. Hmm. Odd. Decided to drop the clocks by 15MHz as these were completely playable after hours of roasting earlier. Again...froze.
I tried removing memory overclock. Upping voltage. Slightly reducing clocks again. Still kept freezing with a driver crash. At this point I was panicking. Obviously my first assumption was I had damaged my VRM. But how could I find out? So I set out disabling SLI and ran my single card with those settings. Seemed to work fine. I was relieved. Thought it was just an SLI setting issue. So then I enabled SLI for Card 1 & 2 only (set 3rd as dedicated PhysX). Loaded Valley to see if it'd crash, and it didn't crash initially but I almost immediately saw artifacting that I didn't get with the single card (#1). Then I switched to Card 1 & 3 SLI. And it froze. So at this point...Card 2 = slightly damaged. Card 3 = pretty well toast. I had closed MSI afterburner as I didn't want any settings to be picked up. So I rebooted my PC and disabled PCIE ports for card 1 & 2 and decided to focus test on Card 3. I even swapped power cables from Card 1 to Card 3 in case there was some issue there.
So I loaded the benchmark with default card settings. And it crashed yet again. I have no idea why. Next step was downclocking. Lowered power limit to 57%. -105MHz on GPU clock. -500MHz on memory. This time it didn't crash and I could see the throttling from the power limit setting. So I slowly moved that up. At 100% power limit and 0 increase to gpu core, I crashed again. But as I crashed...I noticed afterburner was now displaying 1202MHz on the GPU core. So I was like what the hell...reconfirmed settings, saved again, and same thing. Then I had a sigh of relief. Obviously if the GPU is trying to run at 1202MHz without a voltage boost, that would cause crashing. So I increased voltage and Voila. Set to 1.3v, Disabled LLC, upped memory by +750MHz. At 1.325v, +144% power, and +83MHz on the GPU clock, I was now getting 1280MHz on Card 3. Without crashing. Going to 1292 caused a crash, which just means it's lacking juice. Went back down to 1280MHz, and all is stable in a 5 minute bench so far. Previously it'd crash in around 5 seconds.
Now...yes, I figured this out. And I still may have issues with card 2 as I haven't had time to test it out. But...here's the kicker. Previously, to get about 1200MHz from card 3, I had to set the GPU clock to +245MHz or so. And now...the software was doing the same thing with only +83MHz on the slider. This happened unexpectedly. And freaked the hell out of me. I didn't want to be out $2000, and also have to deal with the hassle of redoing my loop. Any ideas how or why this could happen? Has anyone else noticed it?