Hello,
Im kinda noob at this GPGPU (for now :) ) and have few questions:
-
when i run fluidsGL compiled as release i get (on the top of the window) 512x512 and when compiled as emurelease only 64x64 (as i understand code is the same, but why is in emurelease case only 64x64?))
-
why is cpu at 100% when running fluidsGL as release (as i understand it should be computing on graphics card, is this the cause of printing image on screen, or something related to the bandwidth?)?
-
why i get at example “alignedTypes” all TEST FAILED, while in emurelease mode i get TEST PASSED?
-
why i get segmentation failed at some examples running compiled as release (it works fine if i run executable which was compiled as emurelease)?
-
how exactly ‘make’ with emurelease compile programs, it simulate all gf8 functions or … (why is soo much slower)?
Thanks for reply.
PS: I appologize if questions seems stupid :">
As you pointed out later in this post, emurelease is very slow. Would you really want it to run 512x512? :)
When the CPU is waiting for the GPU to finish computing or to finish a DMA operation, it uses 100% of the CPU. This will reportedly be fixed in a future release, though NVIDIA has not indicated exactly when.
emurelease runs on the CPU instead of the GPU. So it is possible to see differences. But I haven’t run into these problems you mentioned. Maybe you can give details of your hardware/software environment.
Yes, it simulates the GPU threads on the CPU. In fact, you can see hundreds of threads being launched on the CPU when a kernel is executed.
Thanks for fast reply.
Well my graphic card is low-end performance, soo it this might be the case.
Geforce 8400 gs 256mb
AMD 64 x2
2gb ram
runing on 64-bit ubuntu gusty gibbon
Ubuntu was never qualified for use with CUDA-1.0. Ubuntu will be qualified for the next CUDA release.