Vray RT killed my motherboard!

[Swahn_Kung]

Okay, don't take the title of this thread to seriously. I don't blame Vray RT or Chaosgroup for this, but I would like to share my experience I just had testing Vray RT on my brand new computer, and hear eventual thought on this.

I was testing GPU rendering (CUDA) in Vray RT when my motherboard completely broke, and I'm curious how this might have happened. Perhaps it was just a coincidence the hardware breakdown happened in the middle of GPU rendering using Vray RT, but there was absolutely nothing going on besides this rendering when the hardware crash occurred. Perhaps it was a bad motherboard specimen, difficult to say. The computer was professionally assembled, and all other components are still working after testing. Only the motherboard died.

Perhaps it's not possible to pinpoint exactly what happened, but anyway, let's start with the relevant components involved:

PSU: EVGA Supernova G2 750W
• Motherboard: X99e-ITX/AC
• GPU: Gigabyte Geforce GTX 1080
• CPU: Intel Core i7-6800K 3,4hgz
• RAM: 1 x Kingston 16GB 2133M DDR4 Fury Black
• Case: Kab Corsair 250D
• Cooling: Corsair Hydro H80, ShadowWings 120mm PWM BL026, (and some additional fans I believe).
• No overclocking or hardware tweaking has been made on the computer.

The thing is that I have been trying out the latest games on the computer with ultra settings. Have run perfectly smooth and stable during many days of extensive gaming without generating significant heat inside the case. No signs of elements that might cause instability. The cooling have handled the load perfectly, and should be able to handle heavy load without problem.

But it only took around one hour of simple GPU rendering in Vray RT before the computer completely shut down, apparently killing the motherboard. I acknowledge that it might just have been a coincidence.

However, I find it reasonable that a GPU rendering process such as Vray RT pushes the performance of the GPU in ways normal game rendering does not. Game graphics are designed to require as little GPU computation as possible, while Vray RT (as I imagine it) is designed to utilize as much performance power as possible for rendering images.

Is it possible that the load of the GTX 1080 can become too much for the motherboard to handle, something that only becomes apparent when the GPU is pushed to its limits? Could my motherboard be too weak to handle such load? Or should there be fail safe limits that kicks in that prevents damage to the card?

On ASRock's product page they brag about this motherboard being "built for content creators", so this motherboard appears to be suitable for heavy graphics related load.

Can it somehow be related to the PSU, yet it affected the motherboard in the end? 750W should be enough. Gigabyte recommends a minimum of 500W for any system using this particular GPU. ASRock's page lists suitable Quadro cards on the product page, they don't differ much from a 1080 card which is tested to consume around 300W at maximum performance according to Tom's Hardware.

But I am no hardware wiz so that's why I'm asking all these questions, in case someone knows.

Again, I don't hold Vray or Chaosgroup responsible in any way. I'm more interested if it's possible to figure out if there is a logical reason for my motherboard's death, or if this was just an obnoxious coincidence from a bad motherboard specimen.

Thanks for any input!

[benb32]

I seriously doubt Vray RT or indeed any software was responsible for this...

There are features in your hardware config that protect the components. These are based on the thermal properties (i.e. temperature and also current draw with your CPU) and throttle the components when the limits are passed. One thing I would say is your power supply is quite close to the minimum you should have - on paper it can supply enough power, but when you are in the top reaches of it's power design it will be much less efficient. Anyway, if the power supply was responsible you would most likely find that you were having the GPU crash or the computer restarting way before anything got cooked.

Games vary a lot in how and what components they tax. Some are more CPU bound others push the GPU much more... but certainly no more than any other CUDA or GPU app.

As it's a new system put it down to the motherboard being faulty and a good thing the fault demonstrated itself in such a conclusively way. I realise it's a pity that it has happened, but you will get a refund or new motherboard on warranty.

All the best Ben

[Swahn_Kung]

Thank you for the feedback.
Yes, it seems reasonable that the motherboard might have been a bad specimen. Talking with the guys doing the service on my computer told me that it is surprising how frequently they replace faulty motherboards. So you recommend a more powerful PSU? Do you think 850W is enough, or do I need 1000W?

Anyway, I guess the only way to be sure it was only a bad motherboard is to do stress the computer again when I get it back, and cross my fingers this crash won't happen again.

[benb32]

850W is good if you're planning on keeping the same configuration bar maybe adding some more RAM. Go for 1000W if you might consider going for a second GPU. In my experience, spend as much as you can afford on a good quality one

[Swahn_Kung]

Yes, my original intention was to put two 16GB memory cards into the computer, but the place I bought it forgot to add that extra. But I have already planned the extra memory card as part of the now ongoing service of the computer. No plans on adding extra GPUs though. Don't think two would even fit inside that 250D case, but I'm quite happy with the speed with just one card.

From the little testing I was able to perform, one GTX 1080 card is ridiculously fast. I think it's at least four times the speed of my previous Quadro in terms of FLOPS, and ten times the CUDA cores if I remember correctly. And a 1080 costs only 16% of the price you have to pay for a Quadro. So if speed is the only thing you are after - highly recommend a 1080!

[savage309]

It should not be possible for any software application to cause a hardware damage. There are a lot of software and hardware protections in case of something went wrong.
Usually RT GPU uses less power than games, because of the specialized game hardware that we don't utilize at all. There are some "power virus" benchmarks, like FurMark that are specially designed to use as much power from your GPU as possible, and even with it there are no reports for permanent damage, neither on the GPU or any of the other components.
So I think it is most likely a coincidence ... such things happen. I am also not a hardware guy, but for example, I have being multiple 200W+ TDP GPUs with 750W power supply and didn't had troubles so far.

If it is a new motherboard, I guess the warranty kick in and you should get a replacement.

Best,
Blago.

[benb32]

The EVGA is a good power supply and is gold rated so should be quite efficient up to around 600 - 650 W. When it will produce more heat and fan noise. It is also helpful to see the biggest demand on the PSU is providing a large current to the GPU. To be clearer your current PSU will be more than sufficient even if you were to upgrade to a more powerful pascal card when they come out in the future. On a side note, having one card for the viewport and a second for RT makes things much more fluid.

Having a slightly over rated PSU does give scope for over clocking - by all accounts the 1080 overclock very well at the small price of demanding more power. Your CPU boosts to 3.6GHz on one thread but doubt after the initial loading of RT it shouldn't be too stressed keeping the GPU busy - as Blago has mentioned on another thread. A different scenario is a CPU render with progressive, while running the denoiser on the GPU. On the back of an envelope calculation gives around 250 W on overclocked 1080 GPU, around 200 W for CPU and motherboard combined with mild overclock, 32GB ddr4 and ssd together fitting into 20 W.

There was a time a couple of chip generations ago where AMD and nVidia detected whether furmark was running to purposely down clock their cards as they were worried about the temp and current draws - from memory, a few people manged to cook their cards by moding the GPU BIOS and over volting. Do take this example as an extream as the furmark benchmark is wholly unrepresentative of any real world test.