-1

I'm using Tesla C2050. I want to run my code with "-arch=sm_20" but I get -nan while the calculations are correct using "-arch=sm_13" ?! What should I figure out the problem?

Thanks, BehZad

Behzad Baghapour
  • 169
  • 1
  • 7
  • Could you provide you code that illustrate your problem? – geek May 31 '12 at 06:28
  • Actually the code is very big to demonstrate. Beside, I tested the code in visual studio 2008 and on GTX480 with above option and did not get that mistake, but when I switched to the linux and ran the code on Tesla C2050, I faced the problem. Does it relate to this change or not? – Behzad Baghapour May 31 '12 at 08:07
  • 2
    @BehzadX: How can you possibly expect anyone to know the answer to this sort of question? You have changed lots of things - gpu, host machine, host operating system, compiler, potentially CUDA version. Any one of them could be contributing to this. Are you getting any API errors when the code runs? (do you check them *all*?) Have you tried `cuda-memcheck` or `cuda-gdb` on the linux host? What do they show? Help us help you. – talonmies May 31 '12 at 10:59
  • This question in a nutshell: "I just changed by jacket from a read one to a blue one and now I'm really cold, how can that be..?". And then, in a later comment, "Btw, the red jacket is a down jacket, the blue jacket is a t-shirt and I've moved from Africa to Siberia." – Roger Dahl May 31 '12 at 19:57

1 Answers1

1

Step through each version of your program in the CUDA debugger and note where they differ.

Roger Dahl
  • 15,132
  • 8
  • 62
  • 82