muhkuh
04-24-08, 01:28 PM
Hello,
we are experiencing a continuous memory leak using drivers ranging from 1.0-9762 to current 173.08. The leak seems to be in glXSwapBuffers.
There are two things that seem to influence the matter:
1. There is some frame cap code that does several calls to usleep and gettimeofday to limit the frame rate to 30 fps. When this code is removed the leak disappears.
2. When VSYNC is disabled the leak disappears too.
I attached a minimal test case that uses glut to create a window and does nothing more than:
clear framebuffer
call usleep(10) several times
swap buffers
When starting this test case with __GL_SYNC_TO_VBLANK=1 and monitoring the kernel module by calling:
vmstat -m | grep nvidia
there is a significant leak (1k every 10s). Setting __GL_SYNC_TO_VBLANK=0 the leak is gone as stated above.
System: FreeBSD 6.2
Intel Celeron (P4) 2GHZ
512 MB Ram
GeForce FX 5200 256 MB
Intel 865G Chipset
The system is used in an embedded environment so the leak is a big problem.
The attached test requires a version of glut to compile.
Thanks in advance.
Markus
Edit:
I made a bug report: NVIDIA Bug ID: 420133
we are experiencing a continuous memory leak using drivers ranging from 1.0-9762 to current 173.08. The leak seems to be in glXSwapBuffers.
There are two things that seem to influence the matter:
1. There is some frame cap code that does several calls to usleep and gettimeofday to limit the frame rate to 30 fps. When this code is removed the leak disappears.
2. When VSYNC is disabled the leak disappears too.
I attached a minimal test case that uses glut to create a window and does nothing more than:
clear framebuffer
call usleep(10) several times
swap buffers
When starting this test case with __GL_SYNC_TO_VBLANK=1 and monitoring the kernel module by calling:
vmstat -m | grep nvidia
there is a significant leak (1k every 10s). Setting __GL_SYNC_TO_VBLANK=0 the leak is gone as stated above.
System: FreeBSD 6.2
Intel Celeron (P4) 2GHZ
512 MB Ram
GeForce FX 5200 256 MB
Intel 865G Chipset
The system is used in an embedded environment so the leak is a big problem.
The attached test requires a version of glut to compile.
Thanks in advance.
Markus
Edit:
I made a bug report: NVIDIA Bug ID: 420133