CUDA 4.0 Runtime API is not working while Device API is working

Hi all,

when I run SDK/deviceQueryDrv, my card is seen and everything seems well

but when I run SDK/deviceQuery

[deviceQuery.exe] starting...

c:\ProgramData\NVIDIA Corporation\NVIDIA GPU Computing SDK 4.0\C\bin\win32\Debug\deviceQuery.exe Starting...

CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 3

-> initialization error

[deviceQuery.exe] test results...

FAILED

Press ENTER to exit...

I never had this kind of problem with previous CUDA version, neither with CUDA 4.0 RC.

Does anybody have an idea where it can come from?

Thanks,

– pium

NB: I have installed all new CUDA4.0 stuff (toolkit, sdk and 270.61 driver)

edit: I also have cuda3.2 that is working with the new driver

Oh dear, I hate installing the SDK. It’s so huge. They really should split it.

Just tried, no problem for me. I use the 270.81 driver, though. I’m on 32-bit Windows 7 as well, with an ATI X1600 as the desktop renderer and a GTX 460.

maybe a laptop driver issue?

http://forums.nvidia.com/index.php?showtopic=201355

Exact same problem here! I can not figure out how to run the sample applications.

As for you, the deviceQueryDrv.exe application works fine, but the deviceQuery.exe fails with the same error message:

C:\ProgramData\NVIDIA Corporation\NVIDIA GPU Computing SDK 4.0\C\bin\win64\Release>deviceQuery.exe

[deviceQuery.exe] starting...

deviceQuery.exe Starting...

CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 3

-> initialization error

[deviceQuery.exe] test results...

FAILED

Press ENTER to exit...

All other applications fail as well, such as vectorAdd.exe:

C:\ProgramData\NVIDIA Corporation\NVIDIA GPU Computing SDK 4.0\C\bin\win64\Release>vectorAdd.exe

[vectorAdd.exe] starting...

Vector Addition

d:/bld_sdk10_x64.pl/rel/gpgpu/toolkit/r4.0/sdk/SDK10/Compute/C/src/vectorAdd/vectorAdd.cu(70) : cudaSafeCall() Runtime API error 3: initialization error.

I un-installed and re-installed everything from scratch just to make sure. But it still does not work.

I installed components in that order:

    Cuda Toolkit – cudatoolkit_4.0.17_win_64.msi

    Cuda Tools – cudatools_4.0.17_win_64.msi

    GPU Computing SDK – gpucomputingsdk_4.0.17_win_64.exe

My system specs are:

    Windows 7 64-bit

    Geforce 8600 GTS

    Driver 270.61

Can anyone help?

Ozo.

Edit: Someone in another thread suggested to re-install with your virus scanner disabled, which I did. Same problem.

Exactly the same problem here!

No progam that uses runtime api for cuda 4.0 seems to work, even my own very simple programs, though everything worked fine with cuda 4 RC2.

My system:

Linux *** 2.6.32-28-generic #55-Ubuntu SMP Mon Jan 10 23:42:43 UTC 2011 x86_64 GNU/Linux

============

here what deviceQuery reports:

[deviceQuery] starting…

./deviceQuery Starting…

CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 3

-> initialization error

[deviceQuery] test results…

FAILED

==========

DeviceQueryDrv works fine:

[deviceQueryDrv] starting…

CUDA Device Query (Driver API) statically linked version

There are 3 devices supporting CUDA

Device 0: “Tesla C2070”

CUDA Driver Version: 4.0

CUDA Capability Major/Minor version number: 2.0

[…]

Device PCI Bus ID / PCI location ID: 3 / 0

Compute Mode:

< Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

Device 1: “Tesla C2070”

CUDA Driver Version: 4.0

CUDA Capability Major/Minor version number: 2.0

[…]

Device supports Unified Addressing (UVA): Yes

Compute Mode:

< Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

Device 2: “GeForce GTX 460”

CUDA Driver Version: 4.0

CUDA Capability Major/Minor version number: 2.1

[…]

Device PCI Bus ID / PCI location ID: 7 / 0

Compute Mode:

< Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

[deviceQueryDrv] test results…

PASSED

We resolved the problem by installing the newest driver, 270.41.19

So-called “production” CUDA 4.0.17 does not work under WHQL driver 260.71. It is amazing how irresponsible are Nvidia’s developers and managers.
You must “upgrade” to 270.81 “developer” driver (non-WHQL), supplied with CUDA 4.0 or downgrade to something older.
Guys, how in universe customers of commercial CUDA software (if any) will handle this (and numerous other) problem(s)? Especially, if they have 100+ computers/boards?
My conclusion (after 30+ years of programming): Nvidia’s CUDA is not for commercial software.
It is a toy.

I had similar problem - samples not working with 260.71, but worked with 270.81.

FYI - I just upgraded driver to the newly released 275.33 (WHQL) and it is working. I have not yet tried new version of NSight - really hope that works as well…