Seti@Home optimized science apps and information
Optimized Seti@Home apps => Windows => GPU crunching => Topic started by: Raistmer on 18 Feb 2011, 03:18:30 am
-
Maybe any suggestions, please?
http://setiathome.berkeley.edu/forum_thread.php?id=63230
-
Maybe any suggestions, please?
http://setiathome.berkeley.edu/forum_thread.php?id=63230
You mentioned oil droplets, possibly from the cooling. An aftermarket cooler can be obtained with new heatpipes etc if that's the case, or water cooling, so that part is replaceable most likely.
Have a close look with a magnifier around the solid capacitors in the power supply area for contamination of that substance, it might be those instead of the cooling & the droplets travelld due to air movement. If they look a bit raised with a weird substance, then they are fried. Replaceing those is very difficult to do safely on multilayer boards properly, & not worth the effort IMO, (but can be done). There is no guarantee other parts weren't taken out with the buck converter failure though (if that is what happened).
Last possibility, off the top of my head, is some fault under the GPU itself with the BGA solder balls cracking, or other components with defective solder joints on the board. There is a last ditch chance you can try , by removing all plastic parts and sticking in the oven to reflow the joints... :-\ (Note: can't do that if there are components on both sides ... they'll fall off ::) )
[Edit:] By the way, In a slightly similar situation, had received a dodgily soldered circuit board prototype the other day, with leadless chips poorly soldered on. It took me an hour of fighting with the guy that brought the board to let me fix it, and 30 seconds to fix it, by applying liquid flux to the top of the chip & applying a fencepost soldering iron to reflow the joints ... works great now & the method I used is pretty similar to how the chips are fitted in the factory.
The tools & skills needed to repair surface mount aren't that high tech, and the original manufacture techniques aren't that special either.
-
What puzzles me - there is no dark areas on board while I definitely heard spark sound (so there was short circuit somewhere) and feelt smell of burning.
Weak joints hardly an option IMHO - card worked few years already and this night there was no mechancal movement.... will look for capacitors and other areas...
-
will look for capacitors and other areas...
Look for fingerprints while you are there ... those are evil :o if there for a long duration (but no failure until 'dendritic salt growths' have time to form & short the tracks, you can only see those with a microscope, but the fingerprints are a giveaway. )
-
I think those oily spots on the memory chips is ominous. thats probably what blew. Just for kicks try and smell them. See if thats what was burning
-
I think those oily spots on the memory chips is ominous. thats probably what blew. Just for kicks try and smell them. See if thats what was burning
That's my thought too,
Claggy
-
I would check the mem chips as well, there shouldn't be oil on them anyway, but it could me moisture from thermal paste, as some pastes give of moisture.
One other thing to check is the motherboard.
I have had a what i thought was a dead GFX-card, turn out to be a defective motherboard.
It was back in the AGP/PCI/ISA days, and the board would boot an old TSeng ET4000 card on ISA, but not anything on AGP (bought a cheapo temporary AGP card as replacement).
New motherboard, and the GFX card came back to life.
Perhaps you have similar problem, that doesn't prevent internal GFX card from working.
A defective supply for the PCI-E slots could be the sinner.
In the motherboard I had before the current one, the GFX card would not work in the top PCI-E slot (after leaky watercooling....), but worked fine in the second PCI-E slot. The board eventually died, I guess because of damage from the water/coolant, but thats another story.
-
Fear not. We are going to get you a new card one way or the other. I know full well how important it is to be connected to what you are programming.
Steve
-
Thanks all for suggestions. Unfortunately cards is dead. I cleaned it, reinstalled - new shortcircuits...
For now I installed secondary GPU, will decompose this one again to see where shortcircuits were...
-
will decompose this one again to see where shortcircuits were...
If nothing still immediately obvious, but there is dirt/dust/oil, there is a thorough cleaning cycle I can describe for a last ditch attempt, though the problem IMO would be connections inside the chips come adrift, or cracked balls under the GPU (thernal aging).
For my 2 cents worth, your time is more valuable to ATi card users right now than to rescuing some 2 generation old card. Hopefully the dudes putting together a kitty toward a newer one will insiist on a Youtube video of violent destruction of the dead one...and ship you a newer one ... Then you'd have your remaining older one & a current generation one to work with, which I can vouch for helps a lot.
Jason
-
For my 2 cents worth, your time is more valuable to ATi card users right now than to rescuing some 2 generation old card. Hopefully the dudes putting together a kitty toward a newer one will insiist on a Youtube video of violent destruction of the dead one...and ship you a newer one ... Then you'd have your remaining older one & a current generation one to work with, which I can vouch for helps a lot.
Jason
I second that thought - time is far more important, and valuable, than hardware disection.
If you have the time, use it more productively by thinking about what class of card would be most useful for current and (near) future development testing. The pace of change is too fast to ask for "the best". Work out what would be most useful for now, to you, and give the fundraisers a clear steer. Otherwise you'll end up with a boxful of gamers' review cast-offs - they may help, or they may be a waste of silicon.
-
AFAIK 6xxx is actually 2 families now, 68xx - Barts and 69xx - Cayman.
Barts quite similar to Hd5xxx while Cayman is different beast. Looks like Cayman is new ATi architecture VLIW4 with what they will go. Future development should target it IMHO. For today's mainstraem smth from HD5xxx family would be enough though they missed all FERMI-like features that will allow few kernels to run simultaneously.
So I think "low-end" Cayman (if low-end is applicable here at all ;D ) more preferable than high-end GPU from prev generations, but it all depends on what will be achievable anyway.
-
Agree with Jason and Richard i do, broken card not worth spending much time diagnosing failure,
look to future, make sure you get card that suits your development purposes,
at least you're still got one example of old generation card,
Claggy
-
the 6950 as I suggested on the seti forums should be adequate. Its the same card as the 6970 with part of it streaming processes hobbled.
-
I will kick anything needed to get the best available. This is an opportunity that can't be passed up. If Jason's 480 melts down, then we'll get him a 590. For what you all have done for us, we can deliver back!
Steve
-
Looks like an early Christmas this year! :D
-
I will kick anything needed to get the best available. This is an opportunity that can't be passed up. If Jason's 480 melts down, then we'll get him a 590. For what you all have done for us, we can deliver back!
Steve
Thanks for so wonderful intentions :)
In my case I think HD6950 would be more than enough. It shares same architecture with HD6970, but has lesser (AFAIK) power requirements.
I not sure that top GPU would be happy with my current PSU and have warm enough (sometimes too warm actually ;) ) room to make it even more warm ;D
-
Looks like an early Christmas this year! :D
Or like old X-mas strikes back ;D
-
This sounds like what happens to me. Now we need a bigger PSU, and an air conditioner..... How ever we can help you Raistmer, we will find a way. :)
Steve
-
This sounds like what happens to me. Now we need a bigger PSU, and an air conditioner..... How ever we can help you Raistmer, we will find a way. :)
Steve
:D :D :D
-
the 6950 should have the same 8 pin and 6 pin power plugs. the 4XXX series only required the 1. check your psu
-
HD4870 requires 2 6-pins plugs, look at photos I posted, there were 2 connectors.
I have 450W PSU
-
try new PSU Superflower Crystal Plus, I have it in my V8-Xeon price ~100 Euro, it supports 3 GTX580
www.super-flower.de
www.super-flower.com.tw
heinz
-
Jason needed a new PSU when we got him the GTX 480. I don't mind chipping in for a bigger PSU at all. I am not rich, but I do care, and I've had a bunch of beer, so I can donate without worry.
Steve
-
Jason needed a new PSU when we got him the GTX 480. I don't mind chipping in for a bigger PSU at all. I am not rich, but I do care, and I've had a bunch of beer, so I can donate without worry.
Steve
Yeah, with throwing in my own funds as well, I was able to step up to the GTX 480, though originally only had planned on a GTX 470. The PSU I had was a fairly decent 450Watt, but with all the other load on it I wasn't going to risk anything, So I borrowed a bit extra off the old lady to get a Seasonic gold series X-750 (http://www.seasonicusa.com/X.htm).
The PSU has been running continuously, quietly & cool, ever since with no sign of stress at all, whereas the 450W supply I had is running a GTX 260 in the other room, and does appear to buckle under the strain from time to time.
As a result of my good experiences with the X-750, when I need to buy PSUs again in the future I'll always look for the highest quality, most efficient unit I can find in whatever power rating I need + some headroom ... I'll never buy an 'ordinary' PSU again.
Jason
-
Has anyone told Mike we need a new power supply too so he can keep his fund drive going a bit longer?
-
For the extreme stress we all put on our systems, regular off the shelf components just don't do well. When I built my rig, headroom was the key. I ended up with a BFG 1250 watt PSU with 3 28 amp rails. I have not regretted that for a microsecond. Even with everything heavilly over volted and over clocked, I have plent of headroom available. I even went the extra distance and wired up a 30 amp twist lock circuit directly from my mains. As for RAC, I got about 65.000 with just two GPU's. I already sent Mike another donation to help with a PSU. I really hope we can come up with enough.
Steve
-
HD4870 requires 2 6-pins plugs, look at photos I posted, there were 2 connectors.
I have 450W PSU
you'll definitely want at least a 700W PSU now.