Number of double precision units and architecture in CUDA Geforce GT 650M 2 GB


I have searched over the web but I haven’t found the number of double precision units of Nvidia Geforce GT 650M 2GB graphics card.

Can anyone tell me how much units does it have? Is the number of double precision units equal to the single precision units or is it lower?

Also can anyone confirm that the architecture of this graphics card is kepler gk107?


The number of double precision units will be much less than the single precision units on that GPU.

If you need 64 bit either the Titan or the Tesla K20 are what you will need. The 650 will perform well with 32 bits floats(that is a gaming GPU which does not need 64 bit units).

Thanks for answering.

There is any document that says which are the exactly number of units?

I need them to write in a scientific paper.

Page 79 of the CUDA C Programming Guide has throughput data:

It’s probably easier to quote throughput numbers because it’s difficult to find definitive specs on how many FP64 units are on a device.

Thanks allanmac! I will probably use those.