Yet another "How many threads run concurrently ?" question.

Hey guys,

I know it has been questioned and answered like a million times. But from all the posts I have read, I am not sure for what is wrong and what is right. So If someone could help me, it would be great.

I have a Tesla C1060 with 30 Multiprocessors and equivalently (30 *8 streaming processors each) 240 cuda cores. But my question now, is How many threads can I have, running concurrently?

From all I have read, I understand that in every clock cycle, I will have 308 = 240 threads running concurrently, and in every 4 clock cycles, 2404 = 960 threads will have complete the execution of 1 instruction.

Is this correct ? Or have I totally misunderstood ?

Thank you in advance for any help.

Hey guys,

I know it has been questioned and answered like a million times. But from all the posts I have read, I am not sure for what is wrong and what is right. So If someone could help me, it would be great.

I have a Tesla C1060 with 30 Multiprocessors and equivalently (30 *8 streaming processors each) 240 cuda cores. But my question now, is How many threads can I have, running concurrently?

From all I have read, I understand that in every clock cycle, I will have 308 = 240 threads running concurrently, and in every 4 clock cycles, 2404 = 960 threads will have complete the execution of 1 instruction.

Is this correct ? Or have I totally misunderstood ?

Thank you in advance for any help.

what’s your concurrently mean? if you mean threads runing at the same time, your analysis is right! but i think concurrent running threads mean active threads, so that number is much greater than 960

what’s your concurrently mean? if you mean threads runing at the same time, your analysis is right! but i think concurrent running threads mean active threads, so that number is much greater than 960

That is pretty much the best case scenario for the GT200. Some instructions take more than one cycle to retire, and there are some special cases where the hardware can dual issue instructions, but otherwise your description should be correct.

That is pretty much the best case scenario for the GT200. Some instructions take more than one cycle to retire, and there are some special cases where the hardware can dual issue instructions, but otherwise your description should be correct.

Thanks both for your answers.

@yyfn Yes I mean threads running at the same time. What do you mean when you say active threads? And how can I calculate (theorytically) those ?

Thanks both for your answers.

@yyfn Yes I mean threads running at the same time. What do you mean when you say active threads? And how can I calculate (theorytically) those ?

He means there are threads that are not done executing, e.g. waiting on some IO.

He means there are threads that are not done executing, e.g. waiting on some IO.

shake