View Single Post
Old 09-18-12, 01:16 AM   #81
rockob
Registered User
 
Join Date: Nov 2008
Posts: 95
Default Re: Random crashes, NVRM Xid messages

Here's an odd observation: if I run the CoDMW3 'Hit and Run' mission from a btrfs partition (with kernel 3.5.3 or 3.6-rc6, Xserver 1.13), the nvidia card crashes *every* time within ten seconds, often in less than 5 seconds. (The temperature of the card is usually under 55C at this point, so there's definitely no overheating issue.) But if I run it from an ext4 or ntfs partition, it runs for longer, sometimes as long as the entire mission - although it is still quite likely to fail at some random point within a few minutes.

Since the nvidia driver apparently only uses the drive for GL shader caching, and this by default goes in ~/.nv (which is always on an ext4 drive on my PC), I wonder if the nvidia driver is not handling interrupts correctly and btrfs is more interrupt intensive than ext4, increasing the chances of nvidia crashing? I can't see that GL caching would be relevant, and indeed if I disable it by exporting __GL_SHADER_DISK_CACHE=0, nvidia still crashes.
rockob is offline   Reply With Quote