I compiled CUDA SDK 2.0 with 2.1 Toolkit, there is no problem running make with no options but make dbg=1 or emu=1 fails.
Device 0 is GeForce 8800GTS 512 @PCIe x4
Device 1 is Tesla c870 @ PCIe x16
System is Fedora 9 x64, gcc 4.3.0(yes, I added the missing headers in SDK), nvidia driver 180.06
Output of make dbg=1
...
make -C projects/reduction/
make[1]: Entering directory `/home/t0bi/NVIDIA_CUDA_SDK/projects/reduction'
reduction_kernel.cu: In function ‘void reduce_sm10(int, int, int, int, T*, T*)’:
reduction_kernel.cu:337: error: ‘reduce0_sm10’ was not declared in this scope
reduction_kernel.cu:340: error: ‘reduce1_sm10’ was not declared in this scope
reduction_kernel.cu:343: error: ‘reduce2_sm10’ was not declared in this scope
reduction_kernel.cu:346: error: ‘reduce3_sm10’ was not declared in this scope
make[1]: *** [obj/debug/reduction_kernel_sm10.cu_o] Error 255
make[1]: Leaving directory `/home/t0bi/NVIDIA_CUDA_SDK/projects/reduction'
make: *** [projects/reduction/Makefile.ph_build] Error 2
Output of make emu=1
...
make -C projects/matrixMulDrv/
make[1]: Entering directory `/home/t0bi/NVIDIA_CUDA_SDK/projects/matrixMulDrv'
nvcc fatal : Emulation mode is not compatible with the phase specified
make[1]: *** [data/matrixMul_kernel.cubin] Error 255
make[1]: Leaving directory `/home/t0bi/NVIDIA_CUDA_SDK/projects/matrixMulDrv'
make: *** [projects/matrixMulDrv/Makefile.ph_build] Error 2
Next problem is OpenGL with Tesla c870. When I try to run app from SDK on c870 that use OpenGL, seems to freeze(nbody, particles) or raises segmentation fault (simpleGL) or prints in console (fluidsGL) :
$ ./fluidsGL -device=1
Using device 1: Tesla C870
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.1/cufft/src/execute.cu, line 1070
cufft: ERROR: CUFFT_EXEC_FAILED
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.1/cufft/src/execute.cu, line 316
cufft: ERROR: CUFFT_EXEC_FAILED
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.1/cufft/src/cufft.cu, line 151
cufft: ERROR: CUFFT_EXEC_FAILED
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.1/cufft/src/execute.cu, line 1070
cufft: ERROR: CUFFT_EXEC_FAILED
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.1/cufft/src/execute.cu, line 316
cufft: ERROR: CUFFT_EXEC_FAILED
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.1/cufft/src/cufft.cu, line 151
cufft: ERROR: CUFFT_EXEC_FAILED
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.1/cufft/src/execute.cu, line 1070
cufft: ERROR: CUFFT_EXEC_FAILED
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.1/cufft/src/execute.cu, line 316
cufft: ERROR: CUFFT_EXEC_FAILED
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.1/cufft/src/cufft.cu, line 151
cufft: ERROR: CUFFT_EXEC_FAILED
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.1/cufft/src/execute.cu, line 1070
cufft: ERROR: CUFFT_EXEC_FAILED
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.1/cufft/src/execute.cu, line 316
cufft: ERROR: CUFFT_EXEC_FAILED
cufft: ERROR: /root/cuda-stuff/sw/rel/gpgpu/toolkit/r2.1/cufft/src/cufft.cu, line 151
cufft: ERROR: CUFFT_EXEC_FAILED
There is no problem running on Device 0 (Geforce 8800 GTS 512), and even runs very,very fast comparing to 2.0. See the attached file, but don’t belive the fps and GFLOPS ;-) but e g nbody has about 60 fps instead 30 I used to have on 8800GTS. Particles demo runs about 200 fps, after increasing iterations count to 10 fps drops to 40. Does new driver really give so big speedup ?
When will be SDK 2.1 beta released ?
