Is my GTX 680 faulty?

[lucasalles]

Hello folks!

I bought an EVGA GTX 680 for my MAC Pro 2010 (Intel 2x3.46GHz, 64GB DDR3, originally ATI 5770) to be able to render in octane, after effects and premiere, I setup a very simple scene (cube, sphere, cylinder and 3 lights) and everytime I hit render the MAC just reboots straight away, I don't even get a error log or something from octane render in Cinema 4D.

I'm wondering if the GPU is faulty, but to be honest, using the computer normally doesn’t seems to have any problems, browsing internet, any office applications and even some graphic design APP’s such as Indesign or Photoshop, including some gaming (Counter Strike GO, Diablo 3, World of Warcraft) all worked completely fine, the problem comes using the GPU for rendering with Premiere PRO, After Effects, or Cinema 4D using Octane.

All the times I tried any of these with the GTX680 the computer just turned off after a few seconds of render. I also have a PC with EVGA GTX 1080Ti which I tested on the MAC Pro, I had to install the NVIDIA web graphics to get the 1080 running on High-Sierra and tried the same APPs, all worked fine. I did also some tests with the GTX680 and web graphics but the same problem occurs. I did check about the compatibility between the GTX 680 and seems that all app's including octane are fully compatible.

Except from octane which needs CUDA, I tested the same after effects and premiere renders with the ATI 5770 and all also worked fine, that's where my questions come, is the EVGA GTX680 not strong enough to run octane? if so, why would be recommended on the octane and NVIDIA website? Or this may be a faulty GPU as the same crash/re-boot happens when intensely using the GPU?

Sorry if this is not the correct place to post.

Thanks in advance,

lucas

 

[h9826790]

May be your GPU come with a 8pin power input, and it's too rely on it. It can explain the shutdown after fews seconds start rendering.

However, still can't explain the restart. If overload the mini 6pin, the cMP will power down itself. Regardless if you choose the auto restart after power failure option. It will remain power off (I tested it myself). The auto reboot really feels like software issue. However, if you already tried both web driver and Apple driver. Then it should not be the driver's issue (unless it's the CUDA driver's problem).

 

[ActionableMango]

everytime I hit render the MAC just reboots straight away

All the times I tried any of these with the GTX680 the computer just turned off after a few seconds of render.

Every time it reboots, or every time it turns off? This is an important distinction. If it is turning off, then I think you might have a power draw or power supply problem. If it is rebooting, well, that could be a million things.

 

[OS6-OSX]

Except from octane which needs CUDA...

Well here is another angle with it's own set of issues.
Since the PC 1080 ti works and has 3584 CUDA cores vs 1536 CUDA cores for the 680, it is the card to try and use. This is were the issues come into play. You can have the 1080 ti (250w as CUDA only) and connect the monitor to a gtx 120 50w. Will need h9826790 to chime in to say, since 300w is the max should the PCIe load not be exactly 300w. The other issue of course would be the 10w HyperX Predator already in the slots. That would be 310w!
Since the 1080 ti works the question is how much time does it take off the rendering? Is it worth the $ to place the 5770 and or the gtx 680 in an external box connected to the 5,1. Connect the monitor to the 5770 and have the two nVidia's Compute only. Not sure about your apps but Resolve 14 can use multiple CUDA cards.

 

[h9826790]

Well here is another angle with it's own set of issues.
Since the PC 1080 ti works and has 3584 CUDA cores vs 1536 CUDA cores for the 680, it is the card to try and use. This is were the issues come into play. You can have the 1080 ti (250w as CUDA only) and connect the monitor to a gtx 120 50w. Will need h9826790 to chime in to say, since 300w is the max should the PCIe load not be exactly 300w. The other issue of course would be the 10w HyperX Predator already in the slots. That would be 310w!
Since the 1080 ti works the question is how much time does it take off the rendering? Is it worth the $ to place the 5770 and or the gtx 680 in an external box connected to the 5,1. Connect the monitor to the 5770 and have the two nVidia's Compute only. Not sure about your apps but Resolve 14 can use multiple CUDA cards.

300W is the total power limit for all 4 slots. Apple didn't make it clear about how to count it, however, since standard PCIe slot power is 75W indeed. Therefore, I tend to believe the real limit is “75W each slot”, more than “All 300W can goes into a any slot(s)”. Besides, this should NOT including the mini 6pin.

Only reference 1080Ti is 250W card, most of non reference card raised the target TDP to around 300W. However, almost none of them will draw more than 70W from the slot in any case. 250W is the TDP thermal limit, not necessary 100% equals to card’s power draw (e.g. the onboard fan(s) itself can draw extra 10-20W in some case). In fact, a reference 1080Ti comes with 6+8pin power input. That will offset most of the power demanded, and leave only ~50W delivered by the PCIe slot. So, it’s almost impossible to exceed that 300W limit in general.

For any card that comes with a 8pin input, that single 8pin connector can draw up to 150W (with factory setting), however, the cMP only has 6pin output, which is rated up to 75W. Even though actual test shows that a single mini 6pin can deliver up to 120W before it trigger the shutdown protection. It's capability still 20% below a 8pin can draw.

Therefore, it's all about how the card draw the power.

e.g.

A 250W TDP 1080Ti draws
50W from slot
50W from 6pin
150W from 8pin

And the user use a single "mini 6pin -> 6pin" and a single "mini 6pin -> 8pin" cable to power the card. The Mac will shutdown.

The same card, the same power draw. But the user using a dual mini 6pin -> single 8pin -> dual 6+2pin combination to power the card. Effectively share the 50W+150W = 200W to the two mini 6pin. Since each of them now only need to deliver 100W. The Mac will survive.

Or

Another 250W 1080Ti, but draws
75W from the slot
75W from the 6pin
100W from the 8pin

Then no matter how to power card, it will survive.

TDP of course is an issue, however, what usually cost the problem is the power distribution. Not really the overall power draw.

Some EVGA 680 actually set the power target at 250W. If it demand more than 120W from the 8pin, and the user only power that connector by a single mini 6pin. There is a high chance that the Mac with shut itself down (but will NOT auto reboot).

 

[lucasalles]

May be your GPU come with a 8pin power input, and it's too rely on it. It can explain the shutdown after fews seconds start rendering.

However, still can't explain the restart. If overload the mini 6pin, the cMP will power down itself. Regardless if you choose the auto restart after power failure option. It will remain power off (I tested it myself). The auto reboot really feels like software issue. However, if you already tried both web driver and Apple driver. Then it should not be the driver's issue (unless it's the CUDA driver's problem).

Thanks for your comments. I have it installed with a mini 6-6pin and mini 6-8pin, same cables I used for the 1080ti. I was using the cuda driver 8.0.90 because if I install the latest (CUDA 9.0.222 driver for MAC) it keeps telling me that I need to update and CUDA stops working, after seen your comment I decided to try again and update to the latest 9.0.222 and again, it stoped working, now if I try to open CUDAz or Octane it says that can't find any cuda devices.
[doublepost=1510316186][/doublepost]

Every time it reboots, or every time it turns off? This is an important distinction. If it is turning off, then I think you might have a power draw or power supply problem. If it is rebooting, well, that could be a million things.

Thanks mate, my bad, I meant "reboots" and not turns off.
[doublepost=1510318391][/doublepost]

Well here is another angle with it's own set of issues.
Since the PC 1080 ti works and has 3584 CUDA cores vs 1536 CUDA cores for the 680, it is the card to try and use. This is were the issues come into play. You can have the 1080 ti (250w as CUDA only) and connect the monitor to a gtx 120 50w. Will need h9826790 to chime in to say, since 300w is the max should the PCIe load not be exactly 300w. The other issue of course would be the 10w HyperX Predator already in the slots. That would be 310w!
Since the 1080 ti works the question is how much time does it take off the rendering? Is it worth the $ to place the 5770 and or the gtx 680 in an external box connected to the 5,1. Connect the monitor to the 5770 and have the two nVidia's Compute only. Not sure about your apps but Resolve 14 can use multiple CUDA cards.

Thank for your reply mate!

I mostly use the PC for gaming and render when using octane, thats the reason i would prefer to have the 1080 on PC rather then the MAC, another reason is that I find quite annoying change the GPU in case I need boot screen, thats why I decided to get the GTX680 which suppose to work out of the box, to be honest I'm actually using more the PC, even for work now. I have to say i was quite disappointed with the performance of the 1080 on the mac, for me seems not worth the amount of money invested for the performance you can get if compared to the PC.

Regarding power, it seems that I'm a fortunate owner of a good PSU on my mac as I haven't found any issues using the 1080 with the 6to6pin + 6to8pin straight from the logic board.

 

[h9826790]

Thanks for your comments. I have it installed with a mini 6-6pin and mini 6-8pin, same cables I used for the 1080ti. I was using the cuda driver 8.0.90 because if I install the latest (CUDA 9.0.222 driver for MAC) it keeps telling me that I need to update and CUDA stops working, after seen your comment I decided to try again and update to the latest 9.0.222 and again, it stoped working, now if I try to open CUDAz or Octane it says that can't find any cuda devices.
[doublepost=1510316186][/doublepost]

Thanks mate, my bad, I meant "reboots" and not turns off.
[doublepost=1510318391][/doublepost]

Thank for your reply mate!

I mostly use the PC for gaming and render when using octane, thats the reason i would prefer to have the 1080 on PC rather then the MAC, another reason is that I find quite annoying change the GPU in case I need boot screen, thats why I decided to get the GTX680 which suppose to work out of the box, to be honest I'm actually using more the PC, even for work now. I have to say i was quite disappointed with the performance of the 1080 on the mac, for me seems not worth the amount of money invested for the performance you can get if compared to the PC.

Regarding power, it seems that I'm a fortunate owner of a good PSU on my mac as I haven't found any issues using the 1080 with the 6to6pin + 6to8pin straight from the logic board.

It’s not the PSU. That’s a 980W rated PSU, of course can drive the 1080 (Ti? With or without Ti make a big difference in this case).

It’s all about how to distribute the power demand between the mini 6pins.

 

[lucasalles]

It’s not the PSU. That’s a 980W rated PSU, of course can drive the 1080 (Ti? With or without Ti make a big difference in this case).

It’s all about how to distribute the power demand between the mini 6pins.

Yes, it's a EVGA 1080Ti Founders Edition

 

[h9826790]

Yes, it's a EVGA 1080Ti Founders Edition

FE should be fine.

They are all 250W TDP, and the power draw is not that high in general. In fact, I just did the power draw test from my 1080Ti reference card 2 days ago.

Since I did the load balance between the mini 6pins, therefore, the 2 boosters always shows similar readings. Anyway, as you can see, apart from Furmark, the 6+8 pin combined power draw rarely exceed 150W, that means if the card draw power by 1:2 proportion, it will draw 100W from the 8pin, and 50W from the 6pin. Which also means, a single mini 6pin -> 8pin will survive.

However, it you run Furmark with that cable config. There is a high chance the Mac will shut down. Because it will draw 200W in total from the 6+8pin. And the 6pin can only deliver up to 75W (in fact, I suspect it will only draw up to ~65W and draw the remaining 135W from the 8pin). This may trigger the shut down protection.

Anyway, back to the topic. Since you already confirmed it's a reboot, but not shut down. So, it's already rule out your card overloaded the PCIe slot or the mini 6pins.

From the consistence of the issue. I will narrow down the source to either a faulty card, or software.

Since you mentioned about "boot screen", and your card is not the Mac edition card. There is a high chance that you get a badly flashed card. Which can cause all sorts of issues. If you can get a full refund. I will return that card without any hesitation.

 

[lucasalles]

FE should be fine.

They are all 250W TDP, and the power draw is not that high in general. In fact, I just did the power draw test from my 1080Ti reference card 2 days ago.
View attachment 734314 Since I did the load balance between the mini 6pins, therefore, the 2 boosters always shows similar readings. Anyway, as you can see, apart from Furmark, the 6+8 pin combined power draw rarely exceed 150W, that means if the card draw power by 1:2 proportion, it will draw 100W from the 8pin, and 50W from the 6pin. Which also means, a single mini 6pin -> 8pin will survive.

However, it you run Furmark with that cable config. There is a high chance the Mac will shut down. Because it will draw 200W in total from the 6+8pin. And the 6pin can only deliver up to 75W (in fact, I suspect it will only draw up to ~65W and draw the remaining 135W from the 8pin). This may trigger the shut down protection.

Anyway, back to the topic. Since you already confirmed it's a reboot, but not shut down. So, it's already rule out your card overloaded the PCIe slot or the mini 6pins.

From the consistence of the issue. I will narrow down the source to either a faulty card, or software.

Since you mentioned about "boot screen", and your card is not the Mac edition card. There is a high chance that you get a badly flashed card. Which can cause all sorts of issues. If you can get a full refund. I will return that card without any hesitation.

Thank you very much for your help!

That's actually what I'm trying to do however the seller keeps trying to convince me that the GPU is fine and that the problem is related to my software which I'm finding hard to believe as the problem only happened with his card.

BTW, a very newbie question, how can I find out if the card is a MAC edition or a PC with flashed room?
I believe it's a PC flashed, this is the card I bought:

https://www.ebay.co.uk/itm/EVGA-GTX...e=STRK:MEBIDX:IT&_trksid=p2060353.m1438.l2649

Many thanks again!

 

[h9826790]

Thank you very much for your help!

That's actually what I'm trying to do however the seller keeps trying to convince me that the GPU is fine and that the problem is related to my software which I'm finding hard to believe as the problem only happened with his card.

BTW, a very newbie question, how can I find out if the card is a MAC edition or a PC with flashed room?
I believe it's a PC flashed, this is the card I bought:

https://www.ebay.co.uk/itm/EVGA-GTX680-2GB-GPU-for-Mac-Pro-macOS-2008-2009-2010-2012/292265557747?ssPageName=STRK:MEBIDX:IT&_trksid=p2060353.m1438.l2649

Many thanks again!

There is only one 680 Mac Edition card, which has only 2x 6pin input to fit the cMP's output. Therefore, your card cannot be the genuine one.

Anyway, the card's info still on EVGA website, and it looks like this.

https://www.evga.com/articles/00730/

 

[lucasalles]

There is only one 680 Mac Edition card, which has only 2x 6pin input to fit the cMP's output. Therefore, your card cannot be the genuine one.

Anyway, the card's info still on EVGA website, and it looks like this.

https://www.evga.com/articles/00730/

Thank again mate! Definitely sending it back.

thank for your help guys.