SDK sample code failures only on samples that launch a kernel

Accelerated Computing CUDA CUDA Programming and Performance
1

hello all, thanks for looking:

fedora core 9, 64bit, c1060 bd and 9800GT bd., latest driver.

for example, the “release” build of simpleMultiGPU when run gives an error:
cutilCheckMsg() CUTIL CUDA error: reduceKernel() execution failed.
in file <simpleMultiGPU.cpp>, line 92 : invalid device function .
Segmentation fault

it’s the line in simpleMultiGPU.cpp (90) that launches the kernel code.
the emurelease simpleMultiGPU works.

this goes for any examples that run a “kernel”, all give some similar error.

is there a linux kernel module to load that i’m just missing to run GPU SDK examples?
(other examples run fine. ie: emurelease)

thanks,
ek

2

what does the devicequery program from the SDK tell you?

3

Device 0: “GeForce 9800 GT”

Major revision number: 1

Minor revision number: 1

Total amount of global memory: 536150016 bytes

Number of multiprocessors: 14

Number of cores: 112

Total amount of constant memory: 65536 bytes

Total amount of shared memory per block: 16384 bytes

Total number of registers available per block: 8192

Warp size: 32

Maximum number of threads per block: 512

Maximum sizes of each dimension of a block: 512 x 512 x 64

Maximum sizes of each dimension of a grid: 65535 x 65535 x 1

Maximum memory pitch: 262144 bytes

Texture alignment: 256 bytes

Clock rate: 1.51 GHz

Concurrent copy and execution: Yes

Device 1: “Tesla C1060”

Major revision number: 1

Minor revision number: 3

Total amount of global memory: 4294705152 bytes

Number of multiprocessors: 30

Number of cores: 240

Total amount of constant memory: 65536 bytes

Total amount of shared memory per block: 16384 bytes

Total number of registers available per block: 16384

Warp size: 32

Maximum number of threads per block: 512

Maximum sizes of each dimension of a block: 512 x 512 x 64

Maximum sizes of each dimension of a grid: 65535 x 65535 x 1

Maximum memory pitch: 262144 bytes

Texture alignment: 256 bytes

Clock rate: 1.30 GHz

Concurrent copy and execution: Yes

Test PASSED

thanks–ek

4

That is strange, I don’t have a clue what is wrong, since the driver is apparently loaded, since the devicequery works. Does the bandwith test also work?

5

yes, on both devices too (–device=0 or --device=1).

this has me stumped too. any other ideas?

also, let me get this straight, in emulation, it runs on the cpu’s not the gpu’s right?

eddie

6

btw: added cudaSetDevice(dev); (dev = 1, for c1060 bd or dev = 0 for 9800gt bd)

to the simpleMultiGPU.cpp and results and error are the same.

7

yep.

I honestly have no clue, and it looks like mfatica is on holidays, so maybe send him a pm? (he knows a lot about linux & cuda)

8

what driver are you using? are you mixing 2.1 toolkit with a 2.0 SDK? what version of libcuda.so is actually installed? what kind of machine is this, and what motherboard is it using?

9

i let the NVIDIA web site choose what got installed (by choosing the linux 64bit/FC9 drivers, etc)

NVIDIA-Linux-x86_64-180.06-pkg2.run

cuda-linux64-rel-nightly-2.1.1635-3065709.run

cuda-sdk-linux-2.10.1126.1520-3141441.run

(which is all correct, i bet…)

as for HW: (see a post from yesterday for an output of “deviceQuery”

9800gt video bd

c1060 telsa bd.

EVGA mb – NOTE: it’s a “nForce 780i SLI” mb w/3ea PCI-E x16 slots

(which make almost EVERYTHING on the MB made by NVIDIA) (should buy NVIDIA stock…)

8 gigs ram, 500g hd, 850w ps, etc. except for cdroms/dvd’s and chassis, it’s all new.

the PN for the MB: 132-CK-NF78 (nForce 780i SLI)

if i forgot anything or you need more info, please let me know. --eddie

10

just wondering - could a segmentation fault be a hardware error?

11

Let me ask a few more questions:

  1. is it possible to write a multicpu (multigpu) software without a kernel?

  2. is the memory on cuda complient boards memory mapped?

(could the segmentation fault be a host system fault?)

thanks, eddie

12

Only for make sure you have install what you said

Know the devices

$lspci | grep VGA

glxinfo | grep rendering

glxinfo | grep NVIDIA

ldconfig -p | grep cuda
13

lspci | grep VGA

03:00.0 VGA compatible controller: nVidia Corporation Unknown device 0614 (rev a2)

glxinfo | grep rendering

direct rendering: Yes

glxinfo | grep NVIDIA

server glx vendor string: NVIDIA Corporation

client glx vendor string: NVIDIA Corporation

OpenGL vendor string: NVIDIA Corporation

OpenGL version string: 2.1.2 NVIDIA 177.82

ldconfig -p | grep cuda

    libicudata.so.38 (libc6,x86-64) => /usr/lib64/libicudata.so.38

    libcufftemu.so.2 (libc6,x86-64) => /usr/local/cuda/lib/libcufftemu.so.2

    libcufftemu.so (libc6,x86-64) => /usr/local/cuda/lib/libcufftemu.so

    libcufft.so.2 (libc6,x86-64) => /usr/local/cuda/lib/libcufft.so.2

    libcufft.so (libc6,x86-64) => /usr/local/cuda/lib/libcufft.so

    libcudart.so.2 (libc6,x86-64) => /usr/local/cuda/lib/libcudart.so.2

    libcudart.so (libc6,x86-64) => /usr/local/cuda/lib/libcudart.so

    libcuda.so.1 (libc6,x86-64) => /usr/lib64/libcuda.so.1

    libcuda.so.1 (libc6) => /usr/lib/libcuda.so.1

    libcuda.so (libc6,x86-64) => /usr/lib64/libcuda.so

    libcuda.so (libc6) => /usr/lib/libcuda.so

    libcublasemu.so.2 (libc6,x86-64) => /usr/local/cuda/lib/libcublasemu.so.2

    libcublasemu.so (libc6,x86-64) => /usr/local/cuda/lib/libcublasemu.so

    libcublas.so.2 (libc6,x86-64) => /usr/local/cuda/lib/libcublas.so.2

    libcublas.so (libc6,x86-64) => /usr/local/cuda/lib/libcublas.so

there is a output of “deviceQuery” listed in one of my replies.

thanks for looking – eddie

14

You stated that you installed 180.06, yet the output you just provided indicates that you have 177.82. Did you switch drivers, or is your driver installation broken (due to a known Ubuntu bug)?

15

that’s interesting, but the install files that i listed the other day are still

on my computer. like i said, let the Nvidia web site do the choosing.

(fc9, 64-bit.) these are the files in my Download directory:

cuda-linux64-rel-nightly-2.1.1635-3065709.run

cuda-sdk-linux-2.10.1126.1520-3141441.run

NVIDIA-Linux-x86_64-180.06-pkg2.run

could there be a problem with the install files?

thanks, eddie

16

The only problem that I see here is that your driver installation is broken. If (re)installing 180.06 does not fix the issue, then please generate and attach an nvidia-bug-report.log.

17

that certainly took care of the “segmentation fault” on kernel SDK samples.

wish i knew how that old driver got on the system (new harddrive).

i’m happy - thanks for the help.

eddie

18

Only guess, perhaps you installed from the ubutu repos, then you only double clicked the installation file of 180, maybe ;).