Re: [HTCondor-users] Bug report: condor_gpu_discovery.exe crashes on Windows 7 & 8.1 32 & 64 Bit in OpenCL detection

[Ernst-Georg Schmid <ernst-georg.schmid@xxxxxxxxx>]

Hi John,

this is what I got:

------------------------------------------------------------------------------------

Lenovo ThinkPad T410
Core i5-M560
No GPGPU
Windows 7SP1 32 Bit
Intel HD driver: 8.15.10.2622
Intel CPU OpenCL runtime: 3.0.1.15216

condor_gpu_discovery -properties -verbose

NAME: "Intel(R) OpenCL"
VERSION: "OpenCL 1.2 "
VENDOR: "Intel(R) Corporation"
PROFILE: "FULL_PROFILE"
DEVICES = 2008376704

Crash

condor_gpu_discovery -properties -verbose -diagnostic

diag: ocl_Init()
ocl platform 0 = 356d28
        NAME: "Intel(R) OpenCL"
        VERSION: "OpenCL 1.2 "
        VENDOR: "Intel(R) Corporation"
        PROFILE: "FULL_PROFILE"
        DEVICES = 2008376704

Crash

condor_gpu_discovery -properties -verbose -opencl
condor_gpu_discovery -properties -verbose -diagnostic -opencl

Same crashes as above with changing  DEVICES = xxx numbers

-------------------------------------------------------------------------------------------------

Fujitsu Celsisus H730
Core i7-4700MQ
Intel HD4600 and Nvidia Quadro K1100M with Optimus switching
Windows 8.1 64 Bit
Nvidia driver: 332.76
Intel HD driver: 10.18.10.3496
Intel CPU OpenCL runtime: 3.0.1.15216

Quadro enabled:

condor_gpu_discovery -properties -verbose

DetectedGPUs="CUDA0"
CUDACapability=3.0
CUDADeviceName="Quadro K1100M"
CUDADriverVersion=6.0
CUDAECCEnabled=false
CUDAGlobalMemoryMb=2048
CUDARuntimeVersion=5.50

OK

condor_gpu_discovery -properties -verbose -diagnostic

diag: simulating cudart using nvcuda
using nvcuda.dll to simulate cudart
DetectedGPUs="CUDA0"
CUDACapability=3.0
CUDADeviceName="Quadro K1100M"
CUDADriverVersion=6.0
CUDAECCEnabled=false
CUDAGlobalMemoryMb=2048
CUDARuntimeVersion=5.50

OK

condor_gpu_discovery -properties -opencl -verbose

NAME: "Intel(R) OpenCL"
VERSION: "OpenCL 1.2 "
VENDOR: "Intel(R) Corporation"
PROFILE: "FULL_PROFILE"
DEVICES = 0

Crash

condor_gpu_discovery -properties -verbose -opencl -diagnostic

diag: simulating cudart using nvcuda
using nvcuda.dll to simulate cudart
diag: ocl_Init()
ocl platform 0 = da2718
        NAME: "Intel(R) OpenCL"
        VERSION: "OpenCL 1.2 "
        VENDOR: "Intel(R) Corporation"
        PROFILE: "FULL_PROFILE"
        DEVICES = 0

Crash

HD4600 enabled:

Error: cuInit returned 100
Error 127: Cant find cudaGetDeviceCount in library: cudart.dll
DetectedGPUs=0

OK

condor_gpu_discovery -properties -verbose -diagnostic

diag: simulating cudart using nvcuda
Error: cuInit returned 100
Error 127: Cant find cudaGetDeviceCount in library: cudart.dll
DetectedGPUs=0

OK

condor_gpu_discovery -properties -opencl -verbose

Error: cuInit returned 100
Error 127: Cant find cudaGetDeviceCount in library: cudart.dll
DetectedGPUs=0

OK

diag: simulating cudart using nvcuda
Error: cuInit returned 100
Error 127: Cant find cudaGetDeviceCount in library: cudart.dll
DetectedGPUs=0

OK

------------------------------------------------------------------------------------------------

Samsung Series 3
Core i7-3630QM
Intel HD4000 and AMD Radeon HD 7670M with Enduro switching
Windows 8.1 64 Bit
AMD driver: 13.152-130830a-163894C-Samsung
Intel HD driver: 10.18.10.3496

Radeon enabled:

condor_gpu_discovery -properties -verbose

NAME: "Intel(R) OpenCL"
VERSION: "OpenCL 1.2 "
VENDOR: "Intel(R) Corporation"
PROFILE: "FULL_PROFILE"
DEVICES = 15790964

Crash

condor_gpu_discovery -properties -verbose -diagnostic

diag: ocl_Init()
ocl platform 0 = e2aee8
        NAME: "Intel(R) OpenCL"
        VERSION: "OpenCL 1.2 "
        VENDOR: "Intel(R) Corporation"
        PROFILE: "FULL_PROFILE"
        DEVICES = 12906740

Crash

condor_gpu_discovery -properties -opencl -verbose

NAME: "Intel(R) OpenCL"
VERSION: "OpenCL 1.2 "
VENDOR: "Intel(R) Corporation"
PROFILE: "FULL_PROFILE"
DEVICES = 9238580

Crash

condor_gpu_discovery -properties -verbose -opencl -diagnostic

diag: ocl_Init()
ocl platform 0 = cdaf10
        NAME: "Intel(R) OpenCL"
        VERSION: "OpenCL 1.2 "
        VENDOR: "Intel(R) Corporation"
        PROFILE: "FULL_PROFILE"
        DEVICES = 10679732

Crash

HD4000 enabled:

Same crashes as above with changing  DEVICES = xxx numbers

---------------------------------------------------------------------------------------------------

As you can see, only CUDA detection seems to be working correctly.

Let me know if I can help you with further information.

best regards,

Ernst-Georg