Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

CUDA Visual Profiler 'Interactive' X config option?

I get the following warning when launching an application in computeprof on Ubuntu 11.04:

The selected counter(s) "gld instructions 8bit", "gld instructions 16bit", "gld instructions 32bit", "gld instructions 64bit", "gld instructions 128bit", "gst instructions 8bit", "gst instructions 16bit", "gst instructions 32bit", "gst instructions 64bit", "gst instructions 128bit" can cause GPU kernels to run longer than the driver's watchdog timeout limit. In this case the driver will terminate the GPU kernel resulting in an application error and the profiling data will not be available. Setting the X Config option 'Interactive' to false is recommended when these counters are selected.

Where do I set this config option?

I am also not getting any counts for 64b and 128b loads and stores when profiling an app on Ubuntu, whereas I see them when I run the profiler on OS X. Is the 'Interactive' config option also affecting this? Note that the kernels do not appear to be timing out; they complete well before the default timeout of 30 seconds.

This is CUDA 4.0.17 on Ubuntu and 4.0.19 on OS X.

like image 454
geordan Avatar asked Aug 02 '11 02:08

geordan


People also ask

How does nvprof work?

The nvprof profiling tool collects and views profiling data from the command-line. It enables the collection of a timeline of CUDA-related activities on both CPU and GPU , including kernel execution, memory transfers, memory set and CUDA API calls and events or metrics for CUDA kernels.

What is Nvprof?

Nvprof is a command-line light-weight GUI-less profiler available for Linux, Windows, and Mac OS. This tool allows you to collect and view profiling data of CUDA-related activities on both CPU and GPU, including kernel execution, memory transfers, etc.

What is CUDA profiler?

Profiling Overview The Visual Profiler is a graphical profiling tool that displays a timeline of your application's CPU and GPU activity, and that includes an automated analysis engine to identify optimization opportunities.


1 Answers

this option can be set in /etc/X11/xorg.conf. Just add Option "Interactive" "0" to the device section of your GPU. Here is what my device section looks like:

Section "Device"
    Identifier     "Device0"
    Driver         "nvidia"
    VendorName     "NVIDIA Corporation"
    BoardName      "GeForce GTX 460"
    Option         "Interactive" "0"
EndSection

I run the same version of CUDA in ubuntu but I'm using OpenCL. Profiling with one of these instructions causes high CPU usage and never finishes (I have to kill the process) - even when interactive mode is disabled. Anyways, you can give it a try.

Also make sure to verify the compute capability of your GPU and check if the requested count is supported : Visual Profiler User Guide. Although it seems that this is not the cause of you problem since it's working in OS X.

like image 186
user909998 Avatar answered Oct 10 '22 16:10

user909998