OCCT 3.1.0 shows HD4870/4890 design flaw - they can't handle the new GPU test !

[Tetedeiench]

I hope you don't think I was flaming you with my source code comment. I am just saying that we can't necessarily rule it out either. Even the best programmers make mistakes and it helps to have extra eyes looking at it.

If it only happens on GDDR5 cards, what about the 4870x2 or 4770? I would test it on my x2 except that I sold it. I'll be sure to test it on the 4890 when I get it - it better pass at 1.45v and 1100mhz core .

What kind of water blocks were used? Full cover?

I don't rule it out completly, but the possibility has been made very, very unlikely at that point. But truly, i only use very basic shader instructions, and DirectX9 functions. Nothing fancy.

I don't have any reports on the HD4770 card at the moment. The 4870x2 cards encountered the same problem, but it happened on one card only, and in crossfire mode, which multiplies the problem factor by 3 or 4.

I don't think it was a full cover. the watercooling didn't increase the vrm cooling, but we increased the VRM cooling by adding fans, without success.

[Tuvok-LuR-]

well 4770 has GDDR5 and castrated VRMs, could be interesting to test if the 40nm can compensate for this

[Ch@pS]

Wow, guy posts and asks for help, community shoots him down in flames and calls his stress test a power virus. Well done members!!

Chalk up another card failure, a sapphire 4870.

[Shintai]

There are also certain stability tests that can permanently damage some Intel CPUs, and nothing really stresses them that much in real-world use.

But one could argue that doing everything you possibly could at once is not a meaningful mode of operation.

Thats pure BS.

There is no way you can damage a VIA/AMD/Intel CPU with any software/power virus.

Because unlike GPUs, CPUs and VRM designs are made to handle anything. Its all about GPU designers going the cheapskater way.

How can a power virus be considered valid software instructions?

How do we know that the software is giving it valid instructions without looking at the source code?

If it can run it its valid. And if the GPU/Card dont have any protection against it it means the GPU/card maker FAILED.

[p2501]

the watercooling didn't increase the vrm cooling, but we increased the VRM cooling by adding fans, without success.

That's what I was saying, if HS contact is bad it doesn't matter how many fans you slap on it, it'll still crash. I'm trying to be constructive here.

[tajoh111]

I hope you don't think I was flaming you with my source code comment. I am just saying that we can't necessarily rule it out either. Even the best programmers make mistakes and it helps to have extra eyes looking at it.

If it only happens on GDDR5 cards, what about the 4870x2 or 4770? I would test it on my x2 except that I sold it. I'll be sure to test it on the 4890 when I get it - it better pass at 1.45v and 1100mhz core .

What kind of water blocks were used? Full cover?

This is a common problem with the xspc ones, I know this for sure and its actually one of the reason beside just putting the project aside, that I haven't installed mine yet.

If this was an NV card, noone would be insulting the OP, infact, alot of people would cheer on the OP as a hero, who threw another spear at the beast known as NVIDIA.

[Tetedeiench]

Two pictures of the VRMs :

The one without any bugs :
http://mangafranceworld.free.fr/Divers/DSC00550.JPG

The one with the bug :
http://mangafranceworld.free.fr/Divers/DSC00551.JPG

Interesting, isn't it ?

[LordEC911]

Wow, guy posts and asks for help, community shoots him down in flames and calls his stress test a power virus. Well done members!!

Chalk up another card failure, a sapphire 4870.

Crying wolf by calling this a design flaw is asking for help?
By definition it is a power virus, so there is nothing wrong with calling it that.

[Tetedeiench]

Crying wolf by calling this a design flaw is asking for help?
By definition it is a power virus, so there is nothing wrong with calling it that.

Well, how do you explain test 7 how my report then ? a HD4870 works fine, you witch the card, poof, same syptoms as the whole bunch of cards replaced everywere.

I know you do want the cards to be bug-proof. But please, let's be constructive. My goal here is to narrow down the cause.

The test we have done narrow the problem to be most probably a problem with the cards, already, and not the software (reread the summary on the previous page to convince yourself). We're trying to get it pinned down now : what is the exact cause of this failure ?

[Shintai]

Two pictures of the VRMs :

The one without any bugs :
http://mangafranceworld.free.fr/Divers/DSC00550.JPG

The one with the bug :
http://mangafranceworld.free.fr/Divers/DSC00551.JPG

Interesting, isn't it ?

So its another "going cheap". That kinda takes you back to the review samples of HD4770 vs retail. I wonder how many other cards are like that. Its tempting to think on NVidia since they "block" the view.

[Tuvok-LuR-]

Which of the following seem most likely to you:
1) The test uses close to 100% of the transistor in the die. In game it doesn't happen cause they aren't optimized for a specifical gpu (as on consoles).

2) The test doesn't bear any cpu worload, so the gpu hasn't to wait for geometry data and has no time for idling between operations. In game it doesn't happen cause there's always a cpu workload

3) Both.

[Shintai]

Which of the following seem most likely to you:
1) The test uses close to 100% of the transistor in the die. In game it doesn't happen cause they aren't optimized for a specifical gpu (as on consoles).

2) The test doesn't bear any cpu worload, so the gpu hasn't to wait for geometry data and has no time for idling between operations. In game it doesn't happen cause there's always a cpu workload

3) Both.

After seeing the same cause for 1000s of people in an MMO I dont believe in any of the above.

And considering driver profiles for games are filled with idle states, its pretty clear that the GPU cant handle certain games normally either. Hence why renaming .exe files.

[Tetedeiench]

Which of the following seem most likely to you:
1) The test uses close to 100% of the transistor in the die. In game it doesn't happen cause they aren't optimized for a specifical gpu (as on consoles).

2) The test doesn't bear any cpu worload, so the gpu hasn't to wait for geometry data and has no time for idling between operations. In game it doesn't happen cause there's always a cpu workload

3) Both.

To me ? 2. I'd almost said "both", as my shader code has been kept simple to be easily optimized for all architectures. But as i haven't done specific codes for specific GPUs, i won't say that of my code. It's a generic code that functions very well on all GPUs out there right now. The very same code runs on every GPU. so i can't say it is optimized for specific GPUs.

EDIT : mind you, shaders in games uses more different functions than i do. So 1 is unlikely. That's usually why they generate less stress on the card. but what if tyhey find a graphical use to the one i came up with ? They'd run into the same issue. And there... crash.

It's hard to switch from work to Shader talking

[largon]

My reference HD4890 crashes at stock 850MHz/975MHz, with the recommended settins (1680x1050, fullscreen, shader complexity=3).
It's stable at stock core speed 850MHz if memory too is at 850MHz.

*continues testing*

[Biker]

Guys, trying to ignore the thread crappers and some others engaged in a rather dull 7 page long pissing contest, all the OP wanted was to highlight a problem he had encountered whilst testing the functionality of a new feature of his stress testing software which he posted here in the hope of getting more confirmation (or not) that the problem exists.

The obvious conclusion thus far is that some it that some 4870 / 4890's are power starved by their insufficient VRM's and / or there might be some kind of mechanism that kicks in at a certain high current draw threshold when they are highly stressed in this way.

A useful step forward at this point would be to collate the results that have been reported thus far and compile a list of which precise cards and manufacturers + editions are affected....

[Tetedeiench]

My reference HD4890 crashes at stock 850MHz/975MHz, with the recommended settins (1680x1050, fullscreen, shader complexity=3).
It's stable at stock core speed 850MHz if memory too is at 850MHz.

*continues testing*

Thanks for your test. if you keep at 850/850 and push the VGPU up... does it crash ?

[largon]

Stock 1.3125v doesn't crash at 850/850. Bump to 1.35v made it crash.

.
.
.

Anyone tried this with nV cards?
I'm interested what FPS do GTX200 series cards manage at the recommended settings and 1680x1050? I suspect HD4800s push a lot higher frame rates than GTX200 serie cards, similar to Furmark where HD4800s dominate...

[Tetedeiench]

Stock 1.3125v doesn't crash at 850/850. Bump to 1.35v made it crash.

.
.
.

Anyone tried this with nV cards?
I'm interested what FPS do GTX200 series cards manage at the recommended settings and 1680x1050? I suspect HD4800s push a lot higher frame rates than GTX200 serie cards, similar to Furmark where HD4800's dominate...

This is the very same behaviour that what we noticed on the french board

I am developing on a GTX285 @home. I can give you that info. But with the time difference (i'm living in France), so it will be in about 10 hours. i'll do that in Shader complexity 3, same res as you (i do have a 22" LCD).

Many thanks for your confirmation though. That's starting to exclude the isolated problem

That's the kind of help i was waiting for

[cx-ray]

Just reporting that the XFX 4850 I tested shuts down too in both OCCT and Furmark (no idea why but I can consistently reproduce it). It functions without trouble however with Freestone Video Card Stability Test and with games.

[Macadamia]

Just reporting that the XFX 4850 I tested shuts down too in both OCCT and Furmark (no idea why but I can consistently reproduce it). It functions without trouble however with Freestone Video Card Stability Test and with games.

Hmm, this one's expected. The XFX boards are really, really cheapo.

[STaRGaZeR]

My 4870 (reference design, Sapphire) craps out even at 750/800. 750/700 is fine. Will test more later, including voltage and GPU clocks. Testing with 1680x1050@120Hz.

Oh really? He proved two cards? How many other members on this board just posted they encountered no such problem?

All sorts of cards will have varying tolerances, and unless the OP has more than a SMALL SAMPLE SIZE, his "conclusion" is irrelevant

Unless you've got a hundred cards that can all REPRODUCE this same issue, you're playing with small sample sizes

Have you read the thread? Your comment suggest a no. Read it again.

People with burned cards caused by the load of FurMark were also a "small sample size". Not even a hundred cards. It was sooo small that even ATI itself changed the drivers for it. Go figure.

[largon]

Hehe,
OCCT GPU test draws 360W from the wall whereas Furmark stays at 310W...


edit:
Crysis Warhead doesn't go above 270W wall draw.

[Biker]

Hehe,
OCCT GPU test draws 360W from the wall whereas Furmark stays at 310W...

OCCT GPU test = not good for the environment

[fellix_bg]

I had an HD4870 1GB from GB (reference design) which couldn't stand to pass FurMark even at stock speeds at mediocre temp's, and this was under benchmark mode, not the burn-in test!

[kromosto]

nice work Tetedeiench