Can't access Tesla C1060 on windows server 2008 unless I also have an nvidia video card - any id

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello,

I have a server box that I want to augment with the Tesla C1060 processor.

When I just use the built-in VGA adapter (AST2050), I can’t run any of the SDK samples as they say there is no CUDA capable card installed. A look at device manager shows the C1060 using the driver that is part of the 2.3 kit is recognized by the OS.

When I plug in an Nvidia display adapter in a spare slot (older-not CUDA capable but uses the same installed driver), I can run all the samples just fine as the C1060 is now accessible.

Does anyone know how I can configure Windows Server 2008 and/or Nvidia driver/CUDA toolkit SW to allow me access to the C1060 when I am only using the built-in VGA adapter?

I’d hate to have to include an extra adapter for no good reason plus lose PCIe slots I need for network adapters.

Please advise!

Best regards,

Jeff Adams

2

Jeff, the Windows display driver model only supports one driver to be loaded at any one time – so when you’re using the normal VGA slot and using that driver, you can’t use CUDA. By plugging in the extra adapter, you’re loading the nVidia driver, which makes both cards work.

I understand that this may be fixed in Windows Server 2008 R2, but I’m not sure (I know they made some driver model changes, but I don’t know if they made it into the RTM or not).

3

I have one of the new SuperMicro 1U Tesla-based Computing Servers. I am having the same issue with Windows 7.

Is there any way around this?

In the worst case scenario, I can always go compute on Linux and continue hating Windows.

4

[url=“High Performance Supercomputing | NVIDIA Data Center GPUs”]Page Not Found | NVIDIA

As you can see in the 4 Tesla C1060 Configuration box:

On-board graphics (works with Linux, Windows requires NVIDIA GPU in one of the PCI-e slots)

It pays to read the system requirements pages before buying a high end computer part.

You can go back to hating Windows now.

5

Can you tell me the vendor and device ID of the display chip in the Supermicro 1U? I haven’t been able to get my hands on one, and I’ve been trying to figure out if the display card has XPDM or WDDM drivers.

6

Running 4 S1070 HIC controllers @ PCI x16.2 on a Tyan S7025, vchip is Aspeed AST2050

I found a PCI (32bit) 8400GS and slipped it in, but now I’m wondering if my MOBO can handle that, sometimes they will deactivate or modify the PCI electrical

bet there is something in the BIOS to set this manually

7

so I disabled the onboard VGA and was checking out the PCI/PnP options

Clear NVRAM (NO)
Plug & Play O/S (NO)
PCI Latency Timer (64)
Allocate IRQ to PCI VGA (YES)

Should any of these be changed?

8

interesting, took out the 8400 hooked up the HICs, booted with onboard vga, but with Plug’N Play enabled (default off)

Windows has found and is installing the drivers on its own at bootup!

I’ll let you know how it turns out.

Cheers,

Josh

9

Hi Josh

Did putting the 8400GS on the PCI (32bit) work?

Sergio

10

I suggest using the Tesla driver for Windows 2008/2008R2, which allows you to run CUDA applications without an NVidia GPU for display.

11

Will this apply to Windows 7 x64 as well? I have tried the latest driver for Windows 7 but still cannot get the C1060 to work correctly without a Nvidia adapter.

12

That driver does apply to C1060 as well (or at least I think it does).