Double precision for mobile Nvidia Mobile GPUs

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi everyone;

I am buying a new laptop and I want to run CUDA on it. But I am confused about double precision performance of Nvidia Mobile GPUs.

I prepared following list:

  • 480M,
  • 485M,
  • 560M,
  • 580M,
  • 2000M

The gaming performance of them is not important for me. The importance is CUDA computing, especially double precision performance.

So could you inform me about double precision performance of them?

Thanks…

2

Check out these two tables on Wikipedia.

3

Thank you tera, you are very helpful…

But I have already read these tables you referred. In these tables, there is no any information about double precision calculation, only single precision for 400M and 500M.

And information in the site is confusing, at least for me.

Following information is written.
“Fermi implements the new fused multiply-add (FMA) instruction for both 32-bit single-precision and 64-bit double-precision floating point numbers (GT200 supported FMA only in double precision)”.

But according to the Fermi table, GF100 and GF110 (M2050 and M2070)are supported double precision. I guess at least 480M, 485M and 580M have double precision.

Please any information

4

All of the Fermi-class GPUs support double precision. Double precision throughput is 1/8th of the single precision throughput for Geforce cards based on GF100 or GF110, or 1/12th for all others.

I suppose that double precision throughput for mobile Quadros based on GF100/GF110 is 1/2 of the single precision throughput (as for the desktop cards), although I’m not 100% sure about that.

5

Thank you so much tera, with the help of your reply I can buy 480M (G100 Fermi) GPU.

The ideal GPU is 5000M/5010M but these are so expensive for me.

Thanks again…

Thank you so much tera, with the help of your reply I can buy 480M (G100 Fermi) GPU.

The ideal GPU is 5000M/5010M but these are so expensive for me.

Thanks again…