Title: Double Precision CUDA GTX 200 OpenGL GLSL Post by: mrrgu on January 05, 2010, 08:35:30 PM Hi
I wanted to buy a GTX 295 card for 64 bit float operations.. But as far as I understand I need to stick with Cuda for this ? or maybe OpenCL? Thing is my CUDA programs keep crashing! and I can not find the cause..sometimes they run fine and then I zoom it crashes..do not think it is temperature..have tested two different computers. So what to do..wanted to go back to OpenGL and write GLSL shaders...but then there is no support for double precision even if the card support it.. So I looked into the possibility of using "double single" float math..but then how to make fast sin() cos(), log(),sqrt()... It does not seem worth the trouble.. Is it easy to use GDB to debug Cuda ? Title: Re: Double Precision CUDA GTX 200 OpenGL GLSL Post by: gaston3d on January 05, 2010, 09:32:48 PM I recommend you elaborate on your gpu programming and wait for http://en.wikipedia.org/wiki/GeForce_300_Series (http://en.wikipedia.org/wiki/GeForce_300_Series) O0 Title: Re: Double Precision CUDA GTX 200 OpenGL GLSL Post by: mrrgu on January 05, 2010, 09:43:07 PM Yes :) That is worth waiting for.. I hope they will update OpenGL and GLSL to use doubles though!!!
I recommend you elaborate on your gpu programming and wait for http://en.wikipedia.org/wiki/GeForce_300_Series (http://en.wikipedia.org/wiki/GeForce_300_Series) O0 Title: Re: Double Precision CUDA GTX 200 OpenGL GLSL Post by: Snakehand on January 06, 2010, 08:39:16 AM ATI supposedly has limited double support on the newest cards with OpenCL, but only for basic arithmetic. So you will have to do series expansion and Newton-Raphson where appropriate. Brook+ also has some double support. |