So, uh, my GPUs failed. Again.

Playing EU4 earlier, suddenly the display went off as the display drivers crashed. It recovered, but then there's ridiculous graphical artifacting all over the screen, and the display drivers are crashing literally every few seconds, eventually crashing the entire system.

Rebooted and the artifacting persisted, both in Windows and in game. Disabling SLi did nothing. A complete wipe and clean installation of the drivers did nothing. Flashing back to default vBIOS (from the svl7 vBIOS) did nothing.

No issues at all when running on the Intel IGP. 100% sure that this is GPU hardware failure.

As those who have seen my past posts will know, this is now the THIRD set of bad cards I've had - the first were 7970Ms that failed, the second were 7970Ms with a graphical artifact in Crossfire, and this is the pair of 680Ms that they replaced them with because I complained about AMD's unreliability.


At least there's one potential upside. I'm currently on holiday in Orlando, and I'm here for another week, so hopefully I'll get good support from the American teams instead of the crap I had to deal with back in the UK. Maybe they'll give me 780Ms and upgrade the rest. Going to phone them and see. Hopefully I can keep this, because I really don't want replacement with an 18 (with its ridiculously hot CPU and the 880M's weird, crappy performance problems)

I'm guessing that this many GPU failures isn't normal and that I'm just incredibly unlucky.

 

I think you are very unlucky. Have you tried the time tested, card swapping method of finding out if it's hardware failure? Swap one GPU and test, then another, test, see which one is defective. Also are you using the same adapter across all three sets of GPUs?

Good luck, we hope for the best!

 

Yea I remember your last thread. Sorry to hear that I hope you get this taken care of soon.

Sent from my Nexus 5 using Tapatalk

 

On the phone with them right now. They have to pull some strings to actually offer hardware support as I'm an international customer, but they're doing it. Very impressed so far.

Diagnostics suggest the VRAM is bad, it failed validation.

 

Here's hoping the 4th set works out for you, Evil! I've had the 680Ms before and they were tough cards, hopefully you'll get the same quality in return.

 

Maybe its just me... but if so many cards had failed, I wouldn't be flashing modded vbioses onto them and overclocking them. I just had my 880Ms replaced and I got the machine in May and they were on a stock vbios most of the time (including when they failed).

At least you were a) able to flash them back to stock, and b) get Dell to help you though.

Have you investigated the possibility that your motherboard is causing the failures? Just seems odd, especially with both AMD and nVidia cards crapping out.

 

Yeah, you know it takes a very special case to kill 3 sets of cards. I have full confidence in Evil's OC abilities, but I'm starting to wonder if the problem is elsewhere. That's why I asked about the possibilities of the AC adapter maybe being faulty and you bring up a good point that it could be the motherboard as well just frying these poor GPUs.

 

Replacement authorised, new pair of 680Ms will be installed by a technician within the next two days. Certainly can't complain about the service here, apparently the guy had to pull a lot of strings just to be able to arrange this as the main US AW team can't normally service internationally registered computers.

The cards were rarely overclocked, the overclocks were always very mild compared to those that others here have pushed, and temperatures were never close to being dangerous at any point. The main reason I actually flashed the modded vBIOS was not for overclocking but because the first card refused to boost clock, it was a few dozen MHz below what it should have been when gaming.

I do occasionally run a memory overclock as the 680Ms seem to stutter in a fair few games for me (probably due to the smaller memory bus compared to the 7970Ms and the much lower memory clocks compared to the desktop cards), but I never pushed it beyond an extra 100MHz on top of the base 900MHz which shouldn't have caused any issues.

The motherboard probably isn't the cause, the other GPU failures all happened on a different board; the board was replaced to try to attempt to fix the visual artifacting I was seeing with the 7970Ms after the first pair failed.

Well, I've been using the same adapter from day one...maybe? The 7970M failures weren't so unusual as they were known for not being particularly reliable. The artifacting issue...I still have no idea what that was, because it happened on two pairs of 7970Ms, two different xfire cables, and two mobos...it was only fixed with the switch to 680Ms. And 680Ms aren't unknown to fail either, even if they are more reliable.

I kind of feel that it's unlikely that the power supply could have killed several different graphics cards in several different ways without also damaging anything else in the system. It's a possibility though...I'll see if any further problems arise.

 

I hear you on the modded vbios, i am tempted to flash these 880Ms just to get stock clocks to stick but with the heat... Meh.
Something is obviously causing an issue though or you have the absolute worst luck with graphics cards. Glad to hear they took care of you though

Sent from my HTC One_M8 on Tapatalk

 

Yeah, please keep us updated on your progress. Hopefully it's just a string of bad luck and not a larger underlying problem. Nice work getting your replacements ASAP, gotta give Dell a hand in this as well

 

I've been running modded vBIOS on 580M, 680M and 780M with no issues at all... I flash a modded vBIOS on day 1 in the life of my GPUs. And, you've all seen my overclocks. So, nah, I don't believe that using a modded vBIOS has any relationship to this problem. A vBIOS mod is only a software tweak, not a hardware mod. I think it's just a fluke coincidence. If something is causing it, a messed up AC adapter is a legitimate guess.

Awesome that they are taking care of things while you are here in the states. Maybe mention the AC adapter to them and see if they think it's worth replacing as a precaution.

 

Both GPUs replaced. At first, the graphical artifacting was still happening along with the crashes. Now it's not displaying video at all on the dedicated cards, including POST, even though it's booting into Windows.

Waiting for the tech to enter notes in the system, and then I'm going to have a very serious talk with Dell about the possibility of replacing this with a decked out 18. After two screen replacements, two mobo replacements, ODD replacement, and the replacement of three pairs of graphics cards, I think it's about time that this thing is replaced entirely. I seriously have to ask how much more needs to be replaced before I have a computer that functions as intended.

 

Dell should have done (bold text) that for you a while back, I feel. For them to have kept on flogging the proverbial 'dead horse' this long is not doing the poor ole horse any good at all......even pets get put out of their pain and suffering more mercifully than this!!

Out of interest, did they replace the SLI bridge as that might be a possible culprit. I feel for ya, Bro.....Dell need to man up and just kit you out with a new 18. Period. Replacements are (or should) be based of the original spec machine you ordered, so bear that in mind if you've had any 'bumps' in the meantime.....

 

Considering the number of hardware replacements this computer has had, I'm honestly surprised that they haven't already replaced it.

The technician called me back not long ago, he's talked to Dell and now they apparently want to replace the GPUs and motherboard. Again. For those not counting, this will be the fifth pair of GPUs and fourth motherboard in this computer.

About to call them up to see what they can actually do. I doubt they can do much while I'm here in America, given that they had to pull strings just to service it and that I've only got a week to wait, but I dread contacting the manager who dealt with my issues in the EMEA region. He'll probably want me to send it to the depot for a second time.

In terms of a replacement, the bare minimum they can give me is a 4810MQ as the 4710 turbos to just 3.5GHz, 100MHz short of the default on my current CPU, but I'm going to try to push for one of the higher processors after all of the crap I've had to go through. They'll also have to provide a pair of 880Ms, as that's the only comparable config to my current setup. But here's hoping I can push for the best CPU, a bit more RAM, perhaps a larger SSD and a blu-ray drive.

 

For your sake, I hope they give you 780M instead of 880M. You don't want the nightmare that is the 880M

Sent from my HTC One_M8 on Tapatalk

 

Well, the American teams can't do anything for me because my computer's registered in the UK and because I'm only here for another 6 days. If anyone has any suggestions for anyone I could contact outside of the standard department to try to make it happen, I'm all ears, but otherwise I'm going to have to email the manager I dealt with in the EMEA region and hope that he doesn't want me to send it in to the depot again.

I'll ask, but if it's a new system, it's doubtful that they'll be able to produce it with 780Ms.

And I'd still rather have the nightmate that is the 880M than the turd that is the R9 M290X (i.e. a 7970M with more VRAM).

I got the impression that the 880M's throttling issues have been at least mostly fixed with a modded vbios? Or am I talking out of my backside?

 

The card runs hotter with the modded vbios (it forces the cards to run their boost clocks all the time instead of their standard throttle the core after sustained 82C which often causes the slave to run under 954MHz - I have seen as low as 888MHz) but the modded vbios is the only way to get them to even match a 780M in performance.

The modded vbios runs the cards to around 90C max in my Sager with IC Diamond while stock they don't pass 82C with max fans.

So your options are either run the stock vbios and accept some dynamic excessive throttling in exchange for low 80s for temps or run the modded vbios and get full speed but 88-90C temps. Even dropping the clock to the stock 954MHz it will still get into the upper 80s within an hour.

If you're brave, you can put liquid ultra on and stay in the low 80s

Sent from my HTC One_M8 on Tapatalk

 

Ah. Brilliant. So my choices are to stick with a lemon computer and pray that nothing else breaks (assuming that they can actually fix it) or go with a replacement with a GPU that's quite possibly not as quick due to throttling issues or will kill itself through overheating if the throttling is removed. Or to go with the AMD turd option which will probably barely function thanks to crappy drivers and then die in a puff of smoke after not being pushed very hard at all.

Assuming they can't offer one with 780M's of course, which I'm praying they will.

 

Technically speaking, even when throttled, the 880M is much faster than the M290X and the 680M. 780M it beats at stock speeds but pretty much nobody keeps their 780M stock. If you're going based on overclocking, the 780M will beat the 880M for max stable every day overclock because it runs much cooler. Otherwise the 880M is still pretty damn fast.

This is stock, unmodified, CPU not overclocked, just pulled it out of the box from repair and tested it http://www.3dmark.com/fs/2600484

Keep in mind that 3dmark gets throttled right off the bat. It never boosts and barely averages 954MHz on the stock vbios.

This is the modded vbios, again no overclocking. http://www.3dmark.com/fs/2614610

Sent from my HTC One_M8 on Tapatalk

 

In all honesty, I wouldn't even mind just running the modded vbios and underclocking it slightly. I'm not particularly interested in overclocking, on the 880M there's likely never going to be any need to do so. But I would prefer a constant, reliable clock speed.

Is there any way to play with the voltages on the 880Ms, or rather, can they be decreased slightly? Dropping the voltage a bit would help a lot with temperatures if it's possible (and if it's still stable).

 

The 880M is an overclocked 780M with 8GB of RAM. It runs 954/993MHz @ 1.012/1.018v stock and the mod runs them at 993MHz @ 1v, there is not a lot of room to undervolt them. I was able to take 12.5mv off of both and pass a Valley run but 25 caused a driver crash and since I don't know if these cards have that hardware flag everyone was talking about that cripples them after too many crashes, I'm not willing to push it. I'm just accepting that it's gonna be hot.

I actually logged the stock vbios voltages:

993MHz - 1.018v
980MHz - 1.006v
967MHz - 0.993v
954MHz - 0.987v
941MHz - 0.981v
928MHz - 0.968v

I suppose that each step might have another 12.5mv of wiggle room? At least they will run at those voltages. I'm gonna play with that today actually. That was the slave card which seems to use a bit higher voltage than the master.

Sent from my HTC One_M8 on Tapatalk

 

Well, if I end up with a pair of 880Ms, I'll get a chance to play around with the voltages and clock speeds and find something that works. Is it possible to modify the vBIOS myself to change the clock speeds and voltages permanently?

Just sent an email to the manager, so I'll see what he says about it. I'm not sure what I'll say if he asks me to send it to the depot again, but my response will not be pleasent...

 

If you ask svl7 to make you a custom mod, he might do it for a donation. Otherwise you can make profiles in inspector and save them on the desktop. I don't know if EVGA profiles save voltage changes or not

Best of luck with Dell man, that really sucks.

Sent from my HTC One_M8 on Tapatalk

 

So if the new cards they installed are causing the exact same issue... i would assume the issue is beyond faulty cards. A complete replacement would be ideal, because the cause has not been pinpointed as of yet (it could be a faulty power brick, etc.. for all anyone knows)

 

Yeah, at this point, push for system replacement. This charade has gone for long enough. You're a trooper for sticking with it for this long.

 

So a couple of days ago, before the engineer turned up, I switched it back from IGP to dedicated graphics to demonstrate the problem. Except instead of booting into Windows with no video output, it no longer booted at all, with the 8 beep error code. Now it's completely unusable. Brilliant.

Even better, the technician turned up, unpacked the replacement motherboard, and it was filthy. The ports were clogged with dust, there were hairs stuck to it, and there was what appeared to be a liquid stain in one corner. The tech advised against installing it for those reasons. We went through basic troubleshooting, even found a loose cable under the palmrest, but nothing helped, it still won't POST. Oh, and the HDD ribbon cable clamp broke off in the process.

Another dispatch was supposedly created to replace the motherboard (again), GPUs (again) and something to do with the screen, but I heard nothing more about it from Dell yesterday.

The EMEA manager emailed me back yesterday, he's "raised a query" with the product and engineering team and I'll supposedly receive a decision by the end of Monday. And if that decision isn't a brand new fully loaded 18 to replace my system, I'll be annoyed. (I probably need the extreme CPU just to be able to run it as fast as my current one thanks to how crap Haswell is.)

 

Sorry to hear the troubles continued, and have gotten worse!

At least now you have a serious chance of getting a replacement.

 

If Dell dont build you a new 18 after this latest botch, I'd be telling them where their future business can go.......

 

While you wait and if it's apart...
Look at the pcie slots and make sure there is nothing in there.
Check the sli cable as well.
And check the backs of the cards with the brackets touch the pcb. If you do not have yellow tape anywhere protecting it.. Chances are the brackets are causing the cards to die sooner than later. And make sure the bracket is actually on correctly.

 

I had a conversation with a Mario and Armando (sp) and they said that automatically send out a replacement computer after 3 failed repairs of the same part. So like if you say that the cards work, but then they break a day later, then the cycle resets.

All this to say... make sure you stress test and work the unit hard before you let dell know that the unit works.

 

While the 880Ms are fairly crippled, they still aren't completely awful. You should be able to run them with your M18xR2 without much issue. Ethrem usually does not use max fans on the clevo, which hurts his cooling; out of the box your laptop's cooling should surpass his auto fan clevo; so you may not want to die about the 880Ms yet really. They're sub-par to 780Ms and that's unacceptable; there's no denying this, but it's not like they're the worst card on the planet to have.

Also, haswell is a blast furnace, but you should be easily able to maintain stock speeds/voltage on them. A 4810MQ's 3.6GHz at stock (4 cores), though I would fight a little for a 4900MQ/4910MQ as it's equivalent to your 2820QM.

Seriously though, good luck with your machine.

 

Don't speak about me like you know what you're talking about. This is the third time now I've told you this - I run max fans when I use the modded vbios, its not necessary with the stock (because it doesn't pass 87C and when it hits 87C they get kicked down to 80-82C while max fans creates unacceptable noise and shaves 1-2C off) and the stock will throttle no matter the temperature because it's a power limit, not a thermal one.

Stock vbios my machine gets slightly better scores than notebook check got on their SLI review (with my 4940MX set to function as a 4810mq) so stock vs stock they are better than 780M, you just won't ever squeeze the overclocks out of them that 780M gets.

Sent from my HTC One_M8 using Tapatalk

 

Sorry about that then. I really didn't remember. But of course stock vs stock they'll beat 780Ms, they literally have at best a 220MHz OC compared to 780Ms (my 780Ms never hit 823MHz like they should have). Also, it actually surprises me that power limitation is a reason, because you have enough power for higher clocks than a 780M SLI with 120Hz on your screen at stock, even if it throttles a bit; my 780Ms would artifact games and not even boost at all on stock vBIOS. By that logic, you should be artifacting some; it's weird.

Anyway, even so, max fans for me (with modded vBIOS) can make a change of 5 to 10 degrees (game- and current temperature dependent). I still think something's weird with your heat. But you RMA'd the slave card, didn't you?

 

I RMA'd the machine and they replaced both cards (as confirmed by the having different ASIC quality percentages from the old ones). They function exactly the same as the old cards. I do need to redo the stock paste job but haven't gotten around to it just yet but the clock behavior is exactly the same as the old pair only the slave hasn't puked like my other one did and the system has stopped shutting off during gaming.

NVidia messed up the vbios and set the power target too low to actually run the cards at full stock clocks. I suspect they messed up boost too because I was toying with undervolting and overclocking on the stock vbios and found less core throttling when the slave card was exactly +14 on the core. Unfortunately the stock voltage isn't high enough to sustain the boost clock that pushed so they started to crash and I had to revert to the stock settings.

The modded vbios has issues that pop up out of nowhere. I was using the mod just fine, temps under 90C, and then one day it decided that if I rebooted the machine it would get stuck in a loop where it would show the modded vbios message and then shut off and restart before it would actually POST and it would just loop shutting off and restarting. The only way to fix that was to pull the power cord and start the machine on battery and plug it in after POST or shut the machine down (off) and turn it back on instead of using the restart option. I went back to the stock vbios, I can't risk bricking these cards.

Sent from my HTC One_M8 using Tapatalk

 

I have had this exact issue as well. What I noticed however was that this only happened after I fiddle with Intel XTU, and once I stopped using XTU (in favour of my partially unlocked BIOS for undervolting) the problems ceased happening entirely. In my case, as I said, the modded vBIOS is actually 100% crucial to operating my system at 120Hz, so it's not like I can drop it. But that's my notices. To be fair, I never tried XTU after receiving my machine back from mythlogic the first time, though the machine never showed the modded vBIOS logo when booting (apparently they had UEFI boot turned on? When I cleared the CMOS to attempt fixing my shutdown issue it booted showing the modded vBIOS message but wouldn't boot into windows until I changed the UEFI to boot win 8. Not sure if it was a basic Mythlogic BIOS update or what).

 

It may well be that it's my hybrid booting that is the problem. I set it to non-windows 8 UEFI mode so I could have both UEFI and BIOS booting options.

Interesting that XTU was the cause in your case. I wonder if that's what did it to me because I changed my chip to have a 28x multi instead of a 33x to drop heat in games that don't demand the power of the CPU. But it rebooted when I made that change and I didn't have the issue but subsequent reboots I'm not so sure. Interesting. Thanks for the heads up, never would have thought to blame XTU. Since I have decided I prefer my CPU to run cooler than to run 3.9GHz all cores @ 92C, I don't need the ease of XTU.

Sent from my HTC One_M8 using Tapatalk

 

Yeah, I noticed because when I left XTU as default and closed the entire program outright, I didn't get the issue. I noticed the issue when I started playing around with undervolting in december at first, and thus the link was almost instant in my mind. I fiddled over and over and honestly just dealt with it for a few months. But after that, I looked at my BIOS and realized what I wanted to do with my CPU I could do in my BIOS, so I did it there and left XTU off. Never got the problem again. Interesting, isn't it?

But that's me, I check EVERYTHING till I get a fix XD. You should see the list of things I checked to fix my shutdown bug >_>

 

Or just use Win7

 

Sure you can, you just don't know how.

 

Well it fixed it. And I now have the core running @ 849MHz @ 0.896v with a max temp of 72C. I'm quite happy with that. Thanks man, never would have associated the two!

I meant for every day overclocking. I got my old pair up to 1156 before the brick overloaded. There was probably more room after that even. They do overclock very well, they just put out too much heat for my comfort level and as I already had to replace them with new cards once, I'm going to baby the hell out of them for now.

Sent from my HTC One_M8 using Tapatalk

 

Well see? There ya go. You should have mentioned that problem in the first place before, I would have had you fixed up as early as march XD. now you've got a solid 780M clock with less voltage and your cards be cool and quiet, no? Enjoy your machine =D.

 

That is one hell of a sacrifice I would never make. You are a better man than I in that aspect. Machine would have got sent back if I had to accept anything other than what I wanted it to do.

 

He already sent it back I believe he said. Got new 880Ms.

 

He can get 10 new 880m's but that wont fix the problem though. . It's hard to explain why they don't work but they all do work. You just have to do far more than the norm to get them to work.

 

Well he did actually have a problem; his slave was hitting 90 deg constant without any OCing when using the modded vBIOS, and he said his new slave doesn't do that, so at least that's sorted. I just hope I can get my temps down for some 950/6000 daily clocks when I get some gelid paste or something.

 

Also his system was shutting off during gaming so there's that

Sent from my Nexus 5 using Tapatalk

 

Also, it wasn't working right before it got to all that. There is that as well....

 

Bingo but this is more a case of accepting the fact that the machine works and compromising some speed.

The machine doesn't shut off anymore and I can run them at full boost, I just prefer not to have the temps that high when I don't have any games that aren't maxed out with them running at 780M clocks. 954MHz runs @ 80-82C but a 10C drop in temp from what I get with a stock vbios with only a 6% performance penalty is acceptable for now.

Sent from my HTC One_M8 using Tapatalk

 

Yeah I am still running the stock VBIOS on mine for the time being, I may decide to flash it later down the line but for the time being the performance is more than adequate. I do get those odd moments of lag in high geometry areas in Assassins Creed Black Flag but it's tolerable for the graphics settings that I have basically maxed out in that game.

 

I think black flag is just unoptimized.