CUDA not detecting all GPU's

Accelerated Computing CUDA CUDA Programming and Performance
1

I have two 9800gx2 cards (4gpu’s total) in a 64 bit system. On redhat enterprise linux with driver version 169 (64bit) successfully run the multigpu sample application and it detects 4 gpu’s. here is what /proc/driver/nvidia/cards shows:
[root@puppy cards]# ls
0 1 2 3
[root@puppy cards]# cat 0
Model: G92-450
IRQ: 58
Video BIOS: 62.92.39.00.09
Card Type: PCI-E
DMA Size: 40 bits
DMA Mask: 0xffffffffff
Bus Location: 0c.00.0
[root@puppy cards]# cat 1
Model: G92-450
IRQ: 66
Video BIOS: 62.92.39.00.10
Card Type: PCI-E
DMA Size: 40 bits
DMA Mask: 0xffffffffff
Bus Location: 0d.00.0
[root@puppy cards]# cat 2
Model: G92-450
IRQ: 74
Video BIOS: 62.92.39.00.09
Card Type: PCI-E
DMA Size: 40 bits
DMA Mask: 0xffffffffff
Bus Location: 83.00.0
[root@puppy cards]# cat 3
Model: G92-450
IRQ: 82
Video BIOS: 62.92.39.00.10
Card Type: PCI-E
DMA Size: 40 bits
DMA Mask: 0xffffffffff
Bus Location: 84.00.0

I switched over to kbuntu 9.04 64 bit and installed the latest 190 nvidia driver and ran the multigpu sample app and detect only 2 gpu’s. I’ve also tried fedora 10 in the same configuration and get the exact same results. I’ve also tried the latest vanilla kernel 2.6.30.7 on fedora and redhat enterprise and still the same results with the 190 driver. Here is the /proc/driver/nvidia/cards:
[root@puppy cards]# ls
0 1
[root@puppy cards]# cat 0
Model: GeForce 9800 GX2
IRQ: 19
Video BIOS: 62.92.39.00.09
Card Type: PCI-E
DMA Size: 40 bits
DMA Mask: 0xffffffffff
Bus Location: 0c.00.0
[root@puppy cards]# cat 1
Model: GeForce 9800 GX2
IRQ: 18
Video BIOS: 62.92.39.00.10
Card Type: PCI-E
DMA Size: 40 bits
DMA Mask: 0xffffffffff
Bus Location: 0d.00.0

Thanks

2

Problem solved. I had to disable “ACPI Bus Segmentation” in the BIOS for some reason on my HP xw9300 workstation. Works now with latest nvidia drivers and newest kernels.