Quadro Plex IV on CentOS 5.4 CUDA does not recognize it.

CUDA does not recognize my Quadro Plex (found from running the devicequery program)

Also, running /sbin/lspci gives the following as part of it’s output:

06:00.0 VGA compatible controller: nVidia Corporation G80 [Quadro FX 5600] (rev a2)

80:00.0 RAM memory: nVidia Corporation MCP55 Memory Controller (rev a2)

80:01.0 RAM memory: nVidia Corporation MCP55 LPC Bridge (rev a3)

80:01.1 SMBus: nVidia Corporation MCP55 SMBus (rev a3)

80:0a.0 PCI bridge: nVidia Corporation MCP55 PCI Express bridge (rev a3)

80:0d.0 PCI bridge: nVidia Corporation MCP55 PCI Express bridge (rev a3)

80:0f.0 PCI bridge: nVidia Corporation MCP55 PCI Express bridge (rev a3)

83:00.0 PCI bridge: nVidia Corporation Unknown device 01b3 (rev a3)

84:00.0 PCI bridge: nVidia Corporation Unknown device 01b3 (rev a3)

84:01.0 PCI bridge: nVidia Corporation Unknown device 01b3 (rev a3)

86:00.0 PCI bridge: nVidia Corporation Unknown device 01b3 (rev a3)

87:00.0 PCI bridge: nVidia Corporation Unknown device 01b3 (rev a3)

87:01.0 PCI bridge: nVidia Corporation Unknown device 01b3 (rev a3)

88:00.0 VGA compatible controller: nVidia Corporation G80 [Quadro FX 5600] (rev a2)

89:00.0 VGA compatible controller: nVidia Corporation G80 [Quadro FX 5600] (rev a2)

As far as I know, the newest nVidia driver has been installed for the quadro plex. (190.53?)

Any thoughts? I really need to get this solved.

The “Unknown device” entries don’t mean anything other than the version of pcitutils you have installed hasn’t got the minor numbers for the Quadtro plex to the database file it uses. It doesn’t mean the devices can’t/won’t work or that there is a driver problem. So you can safely ignore that.

As for the actual problem itself, unfortunately I have no experience with that hardware, so I can’t help you. It probably isn’t all that different from a Telsa S1070, so you might well be able to use the current NVIDIA driver. I have used that version of Centos with other CUDA capable hardware without issue, so it should be OK (although now we don’t run the standard 2.6.18 kernel on our production cluster, but that is for a reason unrelated to CUDA).

The cards in the Quadro Plex are identical to those in the PCI slots of the machine itself. Could lspci not recognize these either? I’ll try to get the admin to update lspci and see if there’s a difference in what it shows.