Hi,
I am trying to use GPU’s for scientific computing (molecular dynamics) on a code I am developing myself. I have a GTX 470 for developing on and proving things in principle while I try and get the funding together for a Tesla or time on one. I am having some problems with double precision on my machine. The GTX 470 should be capable of double precision calculation but when I try and compile even the simplest of code (copying to the GPU, writing to and copying back) the compiler says:
ptxas /tmp/tmpxft_000016a0_00000000-5_forcekernels.compute_10.ptx, line 85; warning : Double is not supported. Demoting to float
The makefile is really simple and I have pasted it below and the simple code I am testing with is attached.
I am using cuda 3.0 on Ubuntu 10.04. I don’t know why this isn’t working and would very much like to have a double precision version of this working (even if it is limited by the card I have) before moving to a tesla.
If anyone has any suggestions I would really appreciate it.
Thanks
Dean
################################################################################
Build script for project
################################################################################
Executable name
EXECUTABLE := onefile.prog
CUDA source files (compiled with nvcc)
CUFILES := onefile.cu
CUDA dependency files
CU_DEPS :=
C/C++ source files (compiled with gcc/g++)
CCFILES :=
################################################################################
Rules and targets
ROOTDIR = /home/dean/NVIDIA_GPU_Computing_SDK/C/common/
BINDIR = ./bin
include $(ROOTDIR)common.mk
flag to enable double precision
NVCCFLAGS += -arch sm_13
flag to report on use of registers, shared memory, etc
NVCCFLAGS += -Xptxas -v
onefile.cu (2.46 KB)