GPU Usage

Question: what utility are you using to view your GPU load?

thought so:

http://boards.cityofheroes.com/showt...evga+precision

http://boards.cityofheroes.com/showt...evga+precision

Evga Precision tool is known to cause errors with the game.

Turn it off.

If you want to monitor your GPU usage please use GPUZ: http://www.techpowerup.com/gpuz/

Okay, now that we've eliminated Evga Precision as a potential problem, onto what could be going on.

What it sounds like is a Pipeline stall: http://en.wikipedia.org/wiki/Pipeline_stall // http://en.wikipedia.org/wiki/Hazard_...rchitecture%29

Now given that you have an Nvidia graphics card my gut reaction would be that the card is overheating and is dropping clocks and/or stalling instructions to take a moment and cool off, so we'll start with a potential problem with overheating.

Download and run the The Unigine Heaven 2.5 benchmark using the OpenGl API.

This will tell us if the problem you are experiencing is game-specific, or if it is API / Driver specific.

If you get the same halts while running Heaven 2.5 in OpenGL mode, then we'll know that the problem is not with City of Heroes, but with either the Nvidia driver in use, or a generalized OpenGL rendering issue.

If you do not get the same halts while running Heaven 2.5 in OpenGL mode then we'll know that the issue is related to the City of Heroes client.


Oh, and just so I'm not dodging your specific question:

Overheating Nvidia graphics cards is fairly common. I'm not going to go into the whole www.nvidiasettlement.com thing again since I've done beat that horse into the ground.

As far as I am aware to your specific issue of the game halting and the GPU stalling, there are no outstanding issues that I'm aware of in either the game client or the Nvidia drivers that would cause this stalling to occur.


As to your last question about what is a good GPU load... that's a really hard question to answer as the answer itself depends on a knowledge of how chips are physically built.

http://boards.cityofheroes.com/showp...31&postcount=6

http://boards.cityofheroes.com/showp...01&postcount=4

With the chip's physical boundaries laid out we then get into actually feeding instructions to the chip. This is where the driver-developers get to have their fun.

One of the job's driver developer's have basically consists of optimizing the compiler to send as many instructions as possible and get as close to the physical cap limits as possible.

If your software product generates a low amount of data that requires rendering instructions, it's very easy for the driver-compiler to get all of the instructions into the processor without having to run at full clock speed.

This is why many modern graphics cards have a 2D rendering mode and a 3D rendering mode. There's no point to the graphics card running at full clocks if all of the functions can be met within the basic clock-speeds.

This is also why GPU-load indicators may or may not be accurate. A GPU-load indicator, in most cases, can only tell you that the processor is active. There is no real way for a Load-indicator to measure how much data is actually being processed.

For example: if you are sitting still and staring at a wall, the texture for the wall, the shaders for any shadows, and the lighting calculations have already been accomplished. All the GPU has to do is just update the frame-buffer with the image of the wall. A GPU-load indicator could indicate that the GPU is very busy as it refreshes the frame-rate as fast as it can... or it could indicate that the GPU is doing nothing at all since no new information is being fed into the GPU and all the GPU is doing is just updating what it has already processed.