Not well with flushed GTX680

[kunia]

I've just crashed...second time since flashing in os x
frist time happened while playing wow after about 2 hours of playing
this time, in about 5 minutes.
I'm looking at the console msgs and I cant really see the event...

strange....its not a complete crash, the audio and mouse till work, but i can't open/close anything. Have to hard reboot.

Yesterday under windows I did a full furmark brun in test, and played some bf3, that didn't crash.

and btw, i cant do the open cl oceanwaeve test ever since i put back the gt 120.

perhaps that's the problem. any ideas?

 

[kunia]

May 4 12:18:14 Adrians-Mac-Pro kernel[0]: NVDA(Private): Channel exception! exception type = 0x20 = PBDMA Error
May 4 12:18:32 --- last message repeated 4 times ---
May 4 12:18:32 Adrians-Mac-Pro kernel[0]: NVDA(OpenGL): Channel timeout!
May 4 12:18:34 Adrians-Mac-Pro kernel[0]: NVDA(Private): Channel exception! exception type = 0x20 = PBDMA Error
May 4 12:18:59 localhost bootlog[0]: BOOT_TIME 1367684339 0

heres what crashed me

 

[kunia]

I've googled this around and looks like some sort of driver issue.
Ive just installed 10.8.3 combo (gotta feeling that nvidia drivers i download yesterday are screwing **** up) and installed it...hope it works

 

[kunia]

still crushing...any ideas?
Cuda driver verions: 5.0.45
GPU driver version: 8.10.44 304.10.65f03

 

[cal6n]

Reset PRAM?

 

[kunia]

dunno whats going on....now its back to x8 land width

 

[cal6n]

Is the GT120 still in there? And, if so, Why do you need it?

It's possible that it's causing a driver conflict.

 

[kunia]

I took it out after the first crash..
and the card is still crashing.

the card is again at x8 speed and half the bandwidth (same under windows, x8)
installed the original rom in windows, took out cuda drivers/reinstalled 10.8.3 combo, reinstalled cuda, reset pram and nada

still at x8, half the bandwidth, and now when I shut down the system there is a 'dos' like gray box at top right corner (on black screen)

this is just getting annoying...im about to return it the card and go back to 5870.,..at least that worked

 

[cal6n]

I can't help but wonder if you've got a hardware fault on your motherboard. Are you in warranty? If so, reinstall the 5870 and get Apple to take a look at it.

 

[kunia]

I think Im on warranty till november
so yea...might be...I'll see once I get the 5870 reinstalled..(its at work on my older mac pro

I got the 5770 at the moment...and its running at x16, 5gb link speed

the gtx was running great for half the day....x16, then it crushed in wow....since then only problems

 

[xcodeSyn]

I got the 5770 at the moment...and its running at x16, 5gb link speed

the gtx was running great for half the day....x16, then it crushed in wow....since then only problems

The following quote is from what MVC wrote about your x8 problem previously:

This can also be the result of running a card that draws too much power, it will burn out one trace and cut speed in half.

Assuming one of your power traces was defective, that might explain why your GTX 680 lost 8-lane over time for two cards in a row. You could try running your 5770 by alternating the 6-pin power trace connector, and see if there is any significant difference in performance. It's less likely that both GTX 680 cards are defective than a possible power trace issue. Anyway, it's better to let Apple techs figure out the problems.

 

[kunia]

The following quote is from what MVC wrote about your x8 problem previously:


Assuming one of your power traces was defective, that might explain why your GTX 680 lost 8-lane over time for two cards in a row. You could try running your 5770 by alternating the 6-pin power trace connector, and see if there is any significant difference in performance. It's less likely that both GTX 680 cards are defective than a possible power trace issue. Anyway, it's better to let Apple techs figure out the problems.

I've returned the second card today...they store did a test and the card was fried...so yes...something's wrong with the Mac;(
Strange thing is, all the other cars are running at x16...gt120 and 5770. My old 5870 is at work, and ill pop it back in on Monday.
Would newer generation cards need more out of the pcie slot than an older card?
What do you mean by power trace? The power cables or the actual pins on the pcie slot?
Would this be a slot/logic board issue or power supply problem...the second card didn't work on a test pc machine...fan was running, but there was no video coming out of the dvi ports

Btw, the first card worked at x8 speed in both slots 1 and 2....the second one was at x16 on slot 1 till it crashed and went back to 8...never tested it on slot 2

 

[cal6n]

You can measure the power being drawn through the 4 PCI slots and the 2 PCI leads using Hardware Monitor. There's a free version but I think it only does temperatures, unfortunately.

One thing to be aware of. While slots 3 and 4 are labeled correctly as "line", the labels for the slots 1 and 2 are "boost line" and the leads 1 and 2 are "line". I suspect they've been mis-labeled. I contacted the developer about this but the labels are based on Apple's tech docs and so we have to live with it.

I've attached a history window configured to show a breakdown of power draw on my system. I only have the one PCIe card, so I haven't enabled any of the unused sensors. The high power section was the Valley benchmark on the extreme preset. You can see that the card is drawing around 170W, with around 40 - 45W being provided by the slot itself and 60 - 65W coming from each 6-pin lead.

 

[Tesselator]

My guess is inadequate power too. But that's only a wild guess.

 

[kunia]

So either the PSU is dying, or there's not enough power going to the PCIE slot?

 

[xcodeSyn]

I've returned the second card today...they store did a test and the card was fried...so yes...something's wrong with the Mac;(
Strange thing is, all the other cars are running at x16...gt120 and 5770. My old 5870 is at work, and ill pop it back in on Monday.

The GTX 680 has a TDP of 195W, and the Radeon 5870's TDP is 188W. If your MP has a power problem that could fry the GTX 680, then the 5870 could also be in danger if you push it too hard.

So either the PSU is dying, or there's not enough power going to the PCIE slot?

You probably summarized the possible causes. Here's a picture of the MP's PSU label, there are 6 +12V rails. One of them that supplies graphics power could be failing or the power traces on the motherboard could be damaged. It's time to let the professionals check it out before doing more damages.

 

[kunia]

Since the first card was running at x8 in both 1st and 2nd slot (and it didn't crash - I played BF3 under windows for about 3 Hours and 1.5h of wow under os x), and the second card was running at x16 in the first one for 5 hours till it crashed and went back to x8 (didn't test it in slot 2), would that mean that it's pretty much power supply, not the logic board thats dying? And as I was saying, the second card got fried.

I somewhat doubt that I have two PCIe slots that are dead...or could that be a possibility as well?

Im about to hit the apple store, I just want to know what to tell them

 

[cal6n]

My guess is inadequate power too. But that's only a wild guess.

That's why I suggested measuring the power draw using Hardware Monitor. It would help to see where the power is being drawn from. It would probably not be a good idea to load the card until it's confirmed that the slot and both boost lines are fully functional, though.

So either the PSU is dying, or there's not enough power going to the PCIE slot?

I suspect that one of your boost lines may be shot. That would force the card to draw excessive power from the remaining boost line and fry that circuit as well.

 

[kunia]

That's why I suggested measuring the power draw using Hardware Monitor. It would help to see where the power is being drawn from. It would probably not be a good idea to load the card until it's confirmed that the slot and both boost lines are fully functional, though.

I suspect that one of your boost lines may be shot. That would force the card to draw excessive power from the remaining boost line and fry that circuit as well.

I'll do that once I come back...plus I dont have my old 5870 on hand, so I wont be able to test both of them at the same time.
I dont have the gtx680 anymore...

as a side note, my friend had the same stuff happening on his iMac...strange artifacts when doing gpu intensive stuff, than a freeze. He could still hear the audio (as I did). It turned out that his video card and his logic board was fried.
On top of that the console didn't really tell me much, and at least one didn't even register the crash (or I'm blind)

So when I'll be running the monitor, what should I pay attention to in power settings?

 

[cal6n]

Once you've installed the application, there's a comprehensive user guide under the Help menu > Hardware Monitor Help.

You should pay particular attention to the "Displaying history data" section.

 

[Tesselator]

That's why I...

Hey, I knew that...

 

[kunia]

You can measure the power being drawn through the 4 PCI slots and the 2 PCI leads using Hardware Monitor. There's a free version but I think it only does temperatures, unfortunately.

One thing to be aware of. While slots 3 and 4 are labeled correctly as "line", the labels for the slots 1 and 2 are "boost line" and the leads 1 and 2 are "line". I suspect they've been mis-labeled. I contacted the developer about this but the labels are based on Apple's tech docs and so we have to live with it.

I've attached a history window configured to show a breakdown of power draw on my system. I only have the one PCIe card, so I haven't enabled any of the unused sensors. The high power section was the Valley benchmark on the extreme preset. You can see that the card is drawing around 170W, with around 40 - 45W being provided by the slot itself and 60 - 65W coming from each 6-pin lead.

So what is what? PCIe Slot 1 12V Boost Line - is that the 6 pin connector? and PCIe Slot 1 12 Line pcie slot isleft? which one is the second cable..
sorry, kinda confused about this....i dont know that much about computers

 

[cal6n]

That's how they should be labeled but they're not. Well, on my Mac Pro 4,1 that's been flashed to 5,1 they are like this:

"PCIe Slot 1 12V Boost Line" refers to PCIe slot 1
"PCIe Slot 1 12V Line" refers to 6-pin connector 1

"PCIe Slot 2 12V Boost Line" refers to PCIe slot 2
"PCIe Slot 2 12V Line" refers to 6-pin connector 2

"PCIe Slot 3 12V Line" refers to PCIe slot 3

"PCIe Slot 4 12V Line" refers to PCIe slot 4
​

 

[kunia]

That's how they should be labeled but they're not. Well, on my Mac Pro 4,1 that's been flashed to 5,1 they are like this:

"PCIe Slot 1 12V Boost Line" refers to PCIe slot 1
"PCIe Slot 1 12V Line" refers to 6-pin connector 1

"PCIe Slot 2 12V Boost Line" refers to PCIe slot 2
"PCIe Slot 2 12V Line" refers to 6-pin connector 2

"PCIe Slot 3 12V Line" refers to PCIe slot 3

"PCIe Slot 4 12V Line" refers to PCIe slot 4
​

Thanks!
I have a usb/firewire expansion board in slot 4, and that draws no power...12v line is at 0.0w
is that normal?

 

[cal6n]

I don't know as I don't have any cards except for my graphics card.

However, I'd expect it to draw something, even when idle.

Have you tried your 5770 using each of the 6-pin connectors in turn? Also, have you tried it in both slots 1 & 2?