LuxMark 2.0 question.

[B.Y.O.B.]

AFAIK LuxMark gathers these information from the driver and other sources, so I think eiter the source is wrong or on the transmission to LuxMark something fails... but a developer could give you a more certain answer about that.

[Dade]

I am curious; why does LuxMark report the wrong clock speeds on both GPUs and CPU?

The information come directly from the OpenCL driver: it looks like the driver reports the current clock, not the stock clock or max. allowed clock. So if nothing is running on the CPU/GPU and the device is down-clocked to save energy (as about every modern CPU/GPU does); the OpenCL driver ends to report the clock in power saving mode instead of the max. clock.

I consider this behaviour somewhat a bug of the driver or, at least, an unexpected behaviour. A possible solution would to ask for the current device clocks near the end of the benchmark in order to report a more accurate information (i.e. the clock under heavy load).

[moure]

The information come directly from the OpenCL driver: it looks like the driver reports the current clock, not the stock clock or max. allowed clock. So if nothing is running on the CPU/GPU and the device is down-clocked to save energy (as about every modern CPU/GPU does); the OpenCL driver ends to report the clock in power saving mode instead of the max. clock.

Dade, yeah thats excactly the "issue". Noticed it too (as im using a gpu monitor gadget)

[Arctucas]

I am curious; why does LuxMark report the wrong clock speeds on both GPUs and CPU?

The information come directly from the OpenCL driver: it looks like the driver reports the current clock, not the stock clock or max. allowed clock. So if nothing is running on the CPU/GPU and the device is down-clocked to save energy (as about every modern CPU/GPU does); the OpenCL driver ends to report the clock in power saving mode instead of the max. clock.

I consider this behaviour somewhat a bug of the driver or, at least, an unexpected behaviour. A possible solution would to ask for the current device clocks near the end of the benchmark in order to report a more accurate information (i.e. the clock under heavy load).

If I had any CPU power-saving features enabled, I would accept that conclusion. However, my CPU runs at 4273MHz always, yet LuxMark reports it as the stock speed of 3.07GHz.

The GPUs will downclock when idle, but again, that is not the case for the primary card when running the benchmark, even when only the CPU is being benchmarked.

EDIT: I managed to get the correct GPU speeds reported by re-installing the Forceware.

Here is a CPU-only run:

http://www.luxrender.net/luxmark/detail/result/368

[Arctucas]

OK, I have another question.

When I run the benchmark for a Mode and Scene, then when it finishes, I select either another mode with the same scene, or a different scene with the same mode, I get:

OpenCL ERROR: dCreateContext (-5) in red letters?

[Dade]

OK, I have another question.

When I run the benchmark for a Mode and Scene, then when it finishes, I select either another mode with the same scene, or a different scene with the same mode, I get:

OpenCL ERROR: dCreateContext (-5) in red letters?

"-5" means CL_OUT_OF_RESOURCES. Are you running out of address space/ram ?

[Arctucas]

OK, I have another question.

When I run the benchmark for a Mode and Scene, then when it finishes, I select either another mode with the same scene, or a different scene with the same mode, I get:

OpenCL ERROR: dCreateContext (-5) in red letters?

"-5" means CL_OUT_OF_RESOURCES. Are you running out of address space/ram ?

I do not believe so, what would you recommend doing to check?


EDIT:

I just tried running scene 'Room' with mode 'GPUs only' and the window remained black, although I could see the countdown running.

The test completed with a score of zero, and it showed the wrong clock speeds for my cards again.

[Arctucas]

No one else is having similar issues that I can compare notes with?

I would really like to figure out why I am having these issues, and why some of my scores are very low.

I am concerned I may have a problem with my GPUs, and I want to know if I might need to RMA them or if there is something else I need to address.

Any advice or help would be appreciated.

Thanks

[Dade]

No one else is having similar issues that I can compare notes with?

I would really like to figure out why I am having these issues, and why some of my scores are very low.

I am concerned I may have a problem with my GPUs, and I want to know if I might need to RMA them or if there is something else I need to address.

Any advice or help would be appreciated.

The description of your problems is a bit vague; I suggest you to focus on one problem at time and to look in 2 directions:

1) Do you have less than 8GB ram ?

2) the 460GTX has some peculiar aspect over all NVIDIA GPU. I'm not sure but it may be the only NVIDIA GPU to not use a standard warp size of 32 (i.e I have read it was using a size of 48). This may be the source of some problem with LuxMark default workgroup size (i.e. 64).

[Arctucas]

Yes, I have 6GB of RAM.

According to CUDA-Z, my GTX460s have a warp size of 32.

I apologize for not understanding all this technical stuff, I am only trying to figure out what is wrong.

I appreciate any help you might provide.

Regarding what you posted earlier, about running out of address space/RAM, would you care to explain that and how I might verify that is the issue?

Thanks