TR Forums

hey you guys - I have a 470 from Gigabyte that's acting up and I want advice on how to make sure it is the video card causing problems. I bought in new in Nov. 2010, put it to pretty heavy use in C4D and ZBrush, less demanding in Premiere Pro.

Symptoms started 2 weeks ago - a few "exceeded Time-Out limit" errors when the card hit >70C caused the driver to fail then reload itself. This weekend I got the first BSOD and artifacting - even with temps below 50C. bluescreenview says the NV driver is the culprit on all 3 BSODs.

When the first symptoms started I pulled the card and applied new TIM (AS-5). Everything appeared perfectly fine. I guess the naked eye won't reveal a broken trace but fwiw the board looks straight and the electronics look intact.

I've tried rolling back the driver a few times - it seemed to help for a few days, then the Time-Out errors began again. I tried running OCCT and it spat out errors almost immediately.

I wonder if there's anything you recommend before I try baking this card (for the hell of it; i don't really expect it to cure it). I haven't run memtest(86) in awhile but I really doubt this is RAM-related. The PSU is acting normal, no hiccups ..

Is there anything besides the video card possibly causing these errors and BSOD...?

Thanks you guys.

Yeah, it is the card. Usually it is the memory starting to go bad as it is rarely as well cooled as the GPU, but it also can be issues with the power circuitry on the card. Ive seen that looking burnt on a few cards and the damage makes the card unstable as it can not deliver clean enough power to keep the card running normally.

thanks SF - makes sense - I can run in safe mode no problem - lower res, driving just 1 display instead of 2.

The only way to conclusively prove it's the card is to reproduce the issue with it in another machine.

However, as SF says, it's very likely to be the card. Try your IT department at work if you don't have a spare machine to test it in at home.

yeah I already asked - the guy is petrified my card will somehow "ruin" his machines... it's a dead end.

You think it's likely I fried this card when I started running a fluid/fire sim plugin for C4D? The plugin can use either the CPU or the GPU to render to the viewport; naturally the GPU is much faster with all its cores, even an older card like my 470 so that's what I used.

But the card shouldn't let itself get fried right? I was always careful about watching the temps too....the worst I ever saw was 80C.

just weird how the card gave me no trouble for 3.5 years, then suddenly after pushing it harder it decides to die. Eh...maybe not so weird. ?

Dunno, We had a quartet of light-duty Rhino3D workstations using GTX285's and they all failed due to overheating eventually. Computational rendering is like Furmark to some extent, it can be like a power virus and tip ageing components past their limits.

Mine had similar symptoms - "time out" driver crashes only though, and only while running 3D games. I think it was caused by minuscule board flex. I could try baking it, but I heard I had to douse the board with solder flux, which I don't have, so I replaced it with a spare XFX Radeon HD 6870 instead. That card survived mining for a year straight, and it's still working fine, so I have high hopes for this one.

Chrispy_ wrote:

... like a power virus and tip ageing components past their limits.

I think that's what happened to my trusty 470. Worked great in Adobe's suites, C4D using OpenGL rendering in the viewport was reliable...but once I started running TurbulenceFD (smoke/fire sim plugin) - even though I cranked the fan and kept temps down to <75 - it must've just pushed Fermi over the edge.

I'll be buying a new card ASAP. Since I can't afford Quadro or Tesla - and don't run a 10-bit color pipeline anyway (plus i'm not a full-time pro mograph artist) - is there any point in looking for higher DP numbers?

TurbulenceFD recommends a 4G card and requires CUDA 2.0 but I can't find out whether it uses double-precision math - or benefits from better DP ability.

I guess it doesn't really matter since I'll be sticking with NV and I can't afford Titan. It'll prob be a 4G 770.

Thanks for your help C I appreciate it! If you have any recommendations I'd love to hear them.

canoli wrote:

TurbulenceFD recommends a 4G card and requires CUDA 2.0 but I can't find out whether it uses double-precision math - or benefits from better DP ability.

http://forum.jawset.com/viewtopic.php?f=4&t=1143
Only single precision. Get your 770!

thanks CSC I've read a lot on the TFD forums. All the posts are from end-users and I have no reason to doubt them but I wish manufacturers/designers/programmers (in general) would tell us more about how best to run their software.

I guess it must be common knowledge that only specialized software (and programmers) need DP for FP numbers.

I may wait and save for the 6G 780-Ti. 336GB/sec is s-i-c-k ... I hope the rumor is true and it gets released soon (but not too soon; i need to save $)

Thanks again CSC I appreciate it.

canoli wrote:

I guess it must be common knowledge that only specialized software (and programmers) need DP for FP numbers.
.

I just wish AMD and Nvidia wouldn't artificially cripple the DP performance of their non-FireGL/Quadro cards. Pretty much all the software you'd consider "pro" software these days uses SP for FP and in terms of 3D CADCAM OpenGL has been all but abandoned as a result of the software needing to be compatible with laptop IGPs

For those rare instances when you want DP throughput, it's irritating that your hardware is running at 1/8th or 1/64th of it's potential speed just because of some unnecessarily artificial (and incorrect) product segmentation. It's as stupid as intel's K-series not having VT support....

Chrispy_ wrote:

canoli wrote:

I guess it must be common knowledge that only specialized software (and programmers) need DP for FP numbers.
.

I just wish AMD and Nvidia wouldn't artificially cripple the DP performance of their non-FireGL/Quadro cards. Pretty much all the software you'd consider "pro" software these days uses SP for FP and in terms of 3D CADCAM OpenGL has been all but abandoned as a result of the software needing to be compatible with laptop IGPs

For those rare instances when you want DP throughput, it's irritating that your hardware is running at 1/8th or 1/64th of it's potential speed just because of some unnecessarily artificial (and incorrect) product segmentation. It's as stupid as intel's K-series not having VT support....

I hear ya, though maybe it's not "incorrect" from NV's POV. They have a business model they stick by and god forbid they apply common sense once in awhile. Then...if you decide you need a professional card, the Quadro line doesn't even begin to surpass Geforce until you spend >$1500. Oh sure the drivers are superior but the hardware on the <$1500 Quadros is kind of a joke. I haven't checked them in awhile, maybe it's better these days I dunno.

But crippling the DP performance on the 780 and the 780-Ti is just inexcusable imo. The 780-Ti is a >$700 video card and they still couldn't give us the complete "big kepler" experience? I'm sure if/when the 6G version comes out it will be the same DP spec as the 3G.