Register size in telsa and geforce 200 GTX

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I wanted techspecs for geforce and tesla devices, Can anyone post a link from where
i can get these.

How much register size does a tesla have in a SM? What about the 200 series?

Thanks in advance.

2

The Programming Guide, Appendix A, lists specifications on registers and compute capabilities of the different devices. (To answer your question, both the Tesla C1060 and the GeForce GTX 200 series have 16384 registers per multiprocessor.)

However, to see a comprehensive list of clock rates, multiprocessor counts, and memory bandwidth, I usually go to the wikipedia page:

http://en.wikipedia.org/wiki/Comparison_of…orce_200_Series

3

Ok, I was looking at this particular PDF.

On slide 27, it is mentioned that 10-series (Tesla) has 64KB per SM.

http://www.stanford.edu/dept/ICME/docs/sem…-2008-10-27.pdf

That’s huge, I could go for a tesla instead of 200!!!

Any idea?? Wanted tech docs to just check up such facts.

thanks in advance,

4

Note the units. Each register on all CUDA devices is 32-bits. (When you work with 64-bit types, like doubles, it puts two registers together.) If you take 16384 registers * 4 bytes per register = 64kB of register storage per SM. That’s the same on GeForce and Tesla devices with compute capability >= 1.2.