Hi guys!

This is not really the usual question, but is it normal to crash the Nvidia card when messing around with directx 11, like sending huge data chunks to it and then rendering it in realtime, because mine does.

So the question is, have you ever crashed your nvidia card, if not, could my hardware be faulty?

Thanks, as usual. (I know, this is a weird question.)

-MIGI0027

Good to know!

I somehow managed to make my GPU/Driver hang for 40-70 seconds (Pc Froze), I was almost certain that a BSOD was to appear, but instead my application was disconnected from the GPU and crashed, and the reason for that was because I was sending a gigantic amount of data to the GPU, over and over.

I had this happen a few times back when the drivers for D3D11 were just starting to be released and I was doing some big compute shader workloads. It was very repeatable, and it makes you inspect what you are doing very carefully :)

So to answer your question, no it isn't just you - you are just using your GPU a little too aggressively.

It is stupidly easy to crash a graphics driver, but if you have TDR enabled then it is difficult to actually kill it if the operating system is not compromised. Simply put, GPU drivers do not offer error recovery features as advanced as, for instance, native code running on a standard processor and operating system. The driver can try and do some logic checking for you before issuing the work to the graphics hardware, by way of return codes - which you should always check, by the way - but once the GPU is busy, the driver has very few options to detect any error conditions which the device cannot resolve on its own - such as infinite loops in shaders - much less tell your application about them.

A lot of that comes from the fact that the current generation of GPU's do not support preemptive tasking, that is, they cannot be gracefully interrupted once a shader or kernel is launched on them. If the driver does not have TDR - which detects when the GPU has stopped responding for X seconds and resets it - then you are stuck with a brick and have to reboot to get the GPU back.

I wouldn't worry about it. Just use the GPU properly and the driver should not crash, if it still does then it is more likely to be a driver bug than faulty hardware. Generally, assuming it is not overheating, hardware is either broken or not. If it worked yesterday, and it passed POST today, then it will work another day.

Yep.

when I was doing my terrain for the first time. that was me sending bad vertex data to the gpu. DX10

DX11 runtime won't handle that one either and you get a nice callstack somewhere deep down in the driver, walk the stack untill you find the faulting drawcall and see how many verts you are telling it to have.

Badly formed drawcalls will end up in a similar state, so always validate that what you are sending to the GPU is what you expect it to be.

THis might become harder to debug depening on which VS version you are using as call stacks can get corrupted when dealing with user->kernel->user->kernel code transitions which are common when doing d3d calls. http://randomascii.wordpress.com/2013/08/19/developers-rejoicewindows-7-stack-corruption-fixed/

Btw another sign of misusing the API or driver is when the frame rate is extremely low for simple scenes and shaders, again look at the calls and see that they are issued correctly, I found that the D3D11.1 runtime isn't that forth coming in telling you whats going on either when you run the debug version. I found a fix for this as soon as you are using the DX11.1 runtime the SDK control panel does nothing any more. You can find how to set the debug runtime here: http://www.altdevblogaday.com/2013/09/30/fixing-the-directx-d3d-debug-layer/