Hard Crash

[Phage]

Hi All,

I've just upgraded the box to a SSD and CF the 5870. The spec is on the left. The new machine boots fine and happily will chunter away at Unigine and 3dMark Vantage. OC settings are unchanged, and I'm running the 10.10 WHQL drivers.

Soooo, last night it was time to test the new performance. Firing up BC2 was much faster. Playing, there were the usual network dropouts but nothing unusual However, after a little while I started to notice some tearing and then the game crashed to desktop. Real temp showed 56C on the CPU, but I didn't have GPUZ running.

Firing it up again, I left GPUZ running thinking the GPU temp was possbly a problem in CF. The next time it crashed GPUZ showed temp of 76C and CPU was unchanged at 56C. Third time lucky ?

This time it crashed hard. Locked up, all inputs were unresponsive, picture frozen, and I had to hit the reset to get it to stop. However, it didn't come back from the reset. No fans, no image, nada. Just the glow of the mainboard power switch and CMOS reset button.

I unplugged the power, cleared CMOS and tried again. Nothing.
After a few mins the LEDs stopped flashing and it booted (Yay). The machine went into startup repair, not something I'd seen before. The report produced showed no errors, just an unspecified config change. As this was a new install this seems unlikely. It was now quite late, and I left it as being able to boot at least albeit with no OC.

What do you think ?
Temps ? Seems unlikely...
10.10 is a bad driver ? Possibly.
PSU dying ? How can I tell ?

Thanks in advance.

[peterb]

I'd have thought temps personally. Each time you rebooted the GPU was warmer and the crash was quicker and more drastic. After leaving it for a while (when it had a chance to cool) it rebooted normally - repairing damage caused by the unexpected shutdowns. You are also overclocking so system timings are becoming more critical.

The other possibility is the PSU which may be overloaded (and overheating) with similar effects.

Unless you have some sophisticated test equipment, the PSU would be hard to diagnose (really need a data logger monitoring voltage and preferably current) on all rails so you can see what is happening at the time of the crash. Not easy (or cheap) to set up. So it comes down to the old standby of substitution. Can you try the GPU in another machine? Do you have another PSU you can try in that machine?

It might be worth runing memtest86+ too, just to rule out a memory problem.

[Phage]

GPU Temps was my first thought as well. Might try another fan in the front.
No spare PSU, I'm afraid, but the machine has been rock solid the last few months with only one GPU, and boots AOK, so I'm fairly sure the SSD is not relevant.

Which card is GPUZ measuring the temp of ? Is there a way to get GPUZ to log temps of the top card and keep the data through a crash ?

EDIT: Damn Noctuas are out of stock. Can anyone recommend a 120mm case fan that is as quiet ? (no LED)

[jim]

Scythe Slipstream, choose your speed (SL, L, M, H, SH suffixes) and away you go

http://www.silentpcreview.com/article832-page6.html

In my girlfriend's PC I run the M variety, @5V they are completely silent. If you get the slower varieties, they'll be quiet to start with obviously. Depends what you can get hold of conveniently and cheaply! Either way, they're very quiet fans.

[Domestic_Ginger]

Quick test would be to run with the door off to rule out overheating.

[GheeTsar]

I also suspect it is the GPUs overheating. As a test, perhaps try it with the case open and a household desk-fan or similar blowing directly at the GPUs to see if that improves stability?

[Phage]

Ah yes the old side panel and desk fan. I should have thought of that. Thanks. I'll give that a go.

And the Scythe, eh ? lets have a look at Scan...

[Phage]

I can see that some sort of ducting is going to be required to get between the cards...

[peterb]

I can see that some sort of ducting is going to be required to get between the cards...

Can you separate the cards? (If they are adjacent to each other) so that one isn't blowing hot air onto the other? (or a baffle to redirect the air to the main body of the case where it will be extracted by the existing fans?

[Behemoth]

Thats going to be hard inside an Antec 300. It's not a bad case for cooling, but when you load ot up with hardware like that there's very little room left inside !

[Phage]

Aye - hardly any room now. However, I think I can make a duct (Blue Peter anyone ?) from one of the front fans to the small gap between the GPUs. A cardboard tube and funnel should do the trick. The cardboard end will deform to fit between the cards and the funnel should act as venturi. It won't be pretty, but short of water it seems the best option. There do not apear to be any commercial products to suit.

[Phage]

And you thought I was joking.

[Behemoth]

Has it worked ?

[GheeTsar]

Some 1337 Blue Peter skillz there. The cardboard won't get too hot that close will it? Also - is the wind tunnel noisy?

[peterb]

Some creative work with some plastic drainpipe, or tumble drier hose?

[Phage]

That is a plastic funnel with the end cut off, the inside of a tube of Christmas paper, and of course duct tape.
Gave it a quick spin and the noise is neglible. I fired up GPUZ, and had an hour of BC2, Alt-tabbing to GPUZ to see temps. The highest I saw in those glimpses was 77C, and the max recorded by the app was 79.5C. I don't know which of cards is being measured though and... sure enough it crashed to desktop right at the end of the hour. GPUZ said the temp at the crash was 77C. Hardly excessive I thought, and I have had crashes from BC2 before the CF upgrade. The game is fully up-to-date according to the updater.exe, and the OC has been removed. CPU temps are fine at 56C.

So two questions for the Hexus Gurus.
1) Is there a better tool for measuring the GPU temp that will tell what each card is doing ?
2) Any other explanations you can think of ?

I'll try some other games, or the kids will later.

EDIT: according to Ray Bradbury the temp for paper to ignite is 232.8C, I think it's OK. Of course RB wasn't a fireman...