CUDA in emulation mode on projects Some compiled projects do not run

Accelerated Computing CUDA CUDA Programming and Performance
1

When I try and run some of the Linux C UDA projects they do not work. They all compile
but, when they run well the output is in error.

See clock and Bandwidth work when run, but BlackScholes, eigenvalue, cppIntegration , matrixMul and Montecarlo do not. The
error seems to be common: feature is not yet implemented. Now I am only emulating, I have not installed a NVIDIA cuda enabled graphics card.
It seems that for some programs you can only run native, not in emulation.

Is that correct?

Newport_j

james ~/NVIDIA_CUDA_SDK/bin/linux/emurelease$ ls
asyncAPI bandwidthTest BlackScholes clock cppIntegration deviceQuery dxtc eigenvalues lineOfSight matrixMul MonteCarlo scan template
james ~/NVIDIA_CUDA_SDK/bin/linux/emurelease$ ./asyncAPI
time spent executing by the GPU: 0.00
time spent by CPU in CUDA calls: 146.90
CPU executed 0 iterations while waiting for GPU to finish

Test PASSED

Press ENTER to exit…

james ~/NVIDIA_CUDA_SDK/bin/linux/emurelease$ ./BlackScholes
Initializing data…
…allocating CPU memory for options.
…allocating GPU memory for options.
…generating input data in CPU mem.
…copying input data to GPU mem.
Data init done.
Executing Black-Scholes GPU kernel (1 iterations)…
cutilCheckMsg() CUTIL CUDA error: BlackScholesGPU() execution failed
in file <BlackScholes.cu>, line 195 : feature is not yet implemented.
james ~/NVIDIA_CUDA_SDK/bin/linux/emurelease$ ./cppIntegration
cutilCheckMsg() CUTIL CUDA error: Kernel execution failed in file <cppIntegration.cu>, line 106 : feature is not yet implemented.
james ~/NVIDIA_CUDA_SDK/bin/linux/emurelease$ ./eigenvalues
Matrix size: 2048 x 2048
Precision: 0.000010
Iterations to be timed: 1
Result filename: 'eigenvalues.dat’cppIntegration
Gerschgorin interval: -2.894310 / 2.923304
cutilCheckMsg() CUTIL CUDA error: Kernel launch failed. in file <bisect_large.cu>, line 204 : feature is not yet implemented.
james ~/NVIDIA_CUDA_SDK/bin/linux/emurelease$ ./lineOfSight
Line of sight
cutilCheckMsg() CUTIL CUDA error: Kernel execution failed in file <lineOfSight.cu>, line 234 : feature is not yet implemented.
james ~/NVIDIA_CUDA_SDK/bin/linux/emurelease$ ./matrixMul
cutilCheckMsg() CUTIL CUDA error: Kernel execution failed in file <matrixMul.cu>, line 140 : feature is not yet implemented.
james ~/NVIDIA_CUDA_SDK/bin/linux/emurelease$ ./MonteCarlo
cudaSafeCall() Runtime API error in file <MonteCarlo.cpp>, line 151 : feature is not yet implemented.
james ~/NVIDIA_CUDA_SDK/bin/linux/emurelease$ ./scan
cutilCheckMsg() CUTIL CUDA error: Kernel execution failed in file <scan.cu>, line 138 : feature is not yet implemented.
james ~/NVIDIA_CUDA_SDK/bin/linux/emurelease$

2

Did you installa a CUDA enabled driver?

I had the same problem (emulation with no CUDA capable hardware) and I had to install driver 180.22 to solve.

3

Yes, I believe that is it. I did not install the driver 180.22 because my NVIDIA graphics card is not CUDA enabled. I have one built around 2002. I thought that to install the 180.22 driver you must have a CUDA enabled graphics card (else, during the installation the software will tell you that it does not recognize the currently installed graphics card and thus will ignore it).

Newport_j

4

CUDA enabled hw is not mandatory, but it seems there are problems with the new drivers (both 180.22 and 180.44) and sdk 2.1 (see [topic=“94736”]here[/topic]). Please let me know if it works for you with sdk 2.1.

5

If you put the 180.22 driver into your computer and you have no CUDA enabled graphics card (else why would you be emulating) then all graphics on your system will be messed up. It says during the installation that the 180.22 software does not recgnize your computer’s graphics card do you wish to continue? If you say yes then your non-CUDA ebabled graphics card will work in a very messed up way with an inappropriate CUDA driver. I have my Unbuntu Linux well tuned with the appropriate driver for the card that is in the PC now. As I said it is not CUDA enabled.

I would like to install it just for the other reasons to BlackScholes, and other project programs to work. I do not want its graphics capability. Must I take that along with the other parts?

I am running Ubuntu 8.04.

Newport_j

6

I think driver installation was ok for me. What hw do you have?

After installing this new driver, acceleration is still working on a geforce go 7300 (i.e. glxgears at 900fps). Only some apps (like phun) shift the screen when I close them…

7

Try using the 2.0 toolkit! This worked for me… It seems that 2.1 breaks some emulation features (including CUBLAS, which fails to initialize in emulation mode). By the way, it seems that only switching runtimes is enough: because the 2.0 toolkit doesn’t support VS2008, I use 2.1 (nvcc, etc) for compiling. After that, putting the 2.0 dll-s (or .so files) near the freshly built executable makes them work.