I just got a new Dell R815 server (4x Opteron 6174s) running RHEL 5.6 and want to attach half of my Tesla S1070 to it. I took one of the two host interface cards from the other computer that the S1070 is connected to and put it in the R815, and connected the cable from the S1070 to the R815. The R815 does not recognize the two GPUs that are connected to it. /sbin/lspci shows the adapter card but not the GPUs:
23:00.0 PCI bridge: nVidia Corporation NF200 PCIe2.0 switch for Quadro Plex S4 / Tesla S870 / Tesla S1070 / Tesla S2050 (rev a2)
24:00.0 PCI bridge: nVidia Corporation NF200 PCIe2.0 switch for Quadro Plex S4 / Tesla S870 / Tesla S1070 / Tesla S2050 (rev a2)
24:01.0 PCI bridge: nVidia Corporation NF200 PCIe2.0 switch for Quadro Plex S4 / Tesla S870 / Tesla S1070 / Tesla S2050 (rev a2)
24:02.0 PCI bridge: nVidia Corporation NF200 PCIe2.0 switch for Quadro Plex S4 / Tesla S870 / Tesla S1070 / Tesla S2050 (rev a2)
24:03.0 PCI bridge: nVidia Corporation NF200 PCIe2.0 switch for Quadro Plex S4 / Tesla S870 / Tesla S1070 / Tesla S2050 (rev a2)
The other computer shows ten instances of the PCI bridge and two 3D controllers from nVidia.
Is the problem:
(a) a BIOS problem with the R815 (i.e. call Dell);
(b) the R815 has x16 PCIe v2.0 physical slots but only x8 electrical slots;
(c) something else entirely?
Nvidia lists the S1070 as having been tested with the R410, R610, R710 and R900.