about CUDA programming

I have set up the enviroment and tested serevral examples for CUDA

programming model.

I would like to konw in my programming kernels, how can I obtain the

information about the relationship between the threads and the runing cores.

especially at the problem that can I konw a specific thread runing at which core

for examples, I have GeForce 9800 GT, and I found that there are 112 processing

coles on the GPU . Are there any way to get the information about the cores.

I have read through the reference manual and programming guide.

But I cant get enough information about the hardware implementation

I do research on muilti-processors topic, so it is important to know the

cooperatopn between cores

Can anyone give me some advice or information?

I would appreciate that :))

You mean “cores” ???

I dont think you can find out which core executed which thread. The manual says that the order in which warps inside a block are executed and the order in which the blocks are executed are non-deterministic. The software should not make any assumptions on that.

yes, i mean cores :P


I met the similar questions,too. I use GeForce 866GT which has 32 streaming cores. I really can’t get the information which cores is runing? How do I know the utility of 32 streaming cores? Thank you.

I dont understand. All cores will be utilized to run your CUDA program.

Thank your reply.

In fact, I want to manually dirtribute the computation to specific cores. Can I do this?


Thank you.

Actually, you can, using the %physid special register. But you’d need to code PTX to use it. There’s some threads about it from a while back, search for it. Its mention has been deleted from the latest PTX manuals.