Is there a way to determine the speed of the PCI Express connection to a specific card? I have three cards plugged in:
two Nvidia GTX 480's (one at x16 & and one at x8) one Nvidia GTX 460 running at x8
Is there some way, either by a function call in C or an option to lspci
that I can determine the bus speed of the graphics cards? When I only use one of the cards for my CUDA program, I'd like to use the one which is running at x16.
Thanks!
Note: lspci -vvv
dumps out
For the two GTX 480s. I don't see any differences that pertain to bus speed.
03:00.0 VGA compatible controller: nVidia Corporation Device 06c0 (rev a3)
Subsystem: eVga.com. Corp. Device 1480
Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
Latency: 0
Interrupt: pin A routed to IRQ 16
Region 0: Memory at d4000000 (32-bit, non-prefetchable) [size=32M]
Region 1: Memory at b0000000 (64-bit, prefetchable) [size=128M]
Region 3: Memory at bc000000 (64-bit, prefetchable) [size=64M]
Region 5: I/O ports at df00 [disabled] [size=128]
[virtual] Expansion ROM at b8000000 [disabled] [size=512K]
Capabilities: <access denied>
Kernel driver in use: nvidia
Kernel modules: nvidia, nvidiafb, nouveau
03:00.1 Audio device: nVidia Corporation Device 0be5 (rev a1)
Subsystem: eVga.com. Corp. Device 1480
Control: I/O- Mem- BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
Interrupt: pin B routed to IRQ 5
Region 0: [virtual] Memory at d7ffc000 (32-bit, non-prefetchable) [disabled] [size=16K]
Capabilities: <access denied>
04:00.0 VGA compatible controller: nVidia Corporation Device 06c0 (rev a3)
Subsystem: eVga.com. Corp. Device 1480
Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
Latency: 0
Interrupt: pin A routed to IRQ 16
Region 0: Memory at dc000000 (32-bit, non-prefetchable) [size=32M]
Region 1: Memory at c0000000 (64-bit, prefetchable) [size=128M]
Region 3: Memory at cc000000 (64-bit, prefetchable) [size=64M]
Region 5: I/O ports at cf00 [size=128]
[virtual] Expansion ROM at c8000000 [disabled] [size=512K]
Capabilities: <access denied>
Kernel driver in use: nvidia
Kernel modules: nvidia, nvidiafb, nouveau
04:00.1 Audio device: nVidia Corporation Device 0be5 (rev a1)
Subsystem: eVga.com. Corp. Device 1480
Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
Latency: 0, Cache Line Size: 64 bytes
Interrupt: pin B routed to IRQ 5
Region 0: Memory at dfffc000 (32-bit, non-prefetchable) [size=16K]
Capabilities: <access denied>
And the only differences I see relate specifically to the memory mapping:
myComputer:~> diff card1 card2
3c3
< Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
---
> Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
7,11c7,11
< Region 0: Memory at d4000000 (32-bit, non-prefetchable) [size=32M]
< Region 1: Memory at b0000000 (64-bit, prefetchable) [size=128M]
< Region 3: Memory at bc000000 (64-bit, prefetchable) [size=64M]
< Region 5: I/O ports at df00 [disabled] [size=128]
< [virtual] Expansion ROM at b8000000 [disabled] [size=512K]
---
> Region 0: Memory at dc000000 (32-bit, non-prefetchable) [size=32M]
> Region 1: Memory at c0000000 (64-bit, prefetchable) [size=128M]
> Region 3: Memory at cc000000 (64-bit, prefetchable) [size=64M]
> Region 5: I/O ports at cf00 [size=128]
> [virtual] Expansion ROM at c8000000 [disabled] [size=512K]
18c18
< Control: I/O- Mem- BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
---
> Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
19a20
> Latency: 0, Cache Line Size: 64 bytes
21c22
< Region 0: [virtual] Memory at d7ffc000 (32-bit, non-prefetchable) [disabled] [size=16K]
---
> Region 0: Memory at dfffc000 (32-bit, non-prefetchable) [size=16K]
I tried this:
cat /var/log/Xorg.0.log | grep "NVIDIA"
It will show you the cards and in what PCI Express slot they are and what is the width of that Slot. For example:
20.502] (II) NVIDIA(0): NVIDIA GPU GeForce 9500 GT (G96) at PCI:1:0:0 (GPU-0)
[ 20.502] (--) NVIDIA(0): Memory: 1048576 kBytes
[ 20.502] (--) NVIDIA(0): VideoBIOS: 62.94.1f.00.8a
[ 20.502] (II) NVIDIA(0): Detected PCI Express Link width: 16X
[ 20.502] (--) NVIDIA(0): Interlaced video modes are supported on this GPU
[ 20.502] (--) NVIDIA(0): Connected display device(s) on GeForce 9500 GT at PCI:1:0:0
[ 20.502] (--) NVIDIA(0): Samsung SyncMaster (CRT-1)
[ 20.502] (--) NVIDIA(0): Samsung SyncMaster (CRT-1): 400.0 MHz maximum pixel clock
This is a very old question, but the way to do this properly is run
For each pci-e device, it lists LnkSta, which tells you the speed as _ GT/s, Width x1. It also lists LnkCap, the link capacity.
2.5 GT/s = PCI-e gen 1, 5 GT/s = PCI-e gen 2, 8 GT/s = PCI-e gen 3, and the width is the number of lanes.