Which is the least-capable CUDA implementation?

For testing purposes, I need to locate the slowest/least capable CUDA implementation that will be found in the field. I need to ensure that my code runs properly (and is actually accelerated over pure CPU) on all CUDA implementations.

Does anyone know?

A GeForce G 105M. It has only 8 SPs (!!!)

The 9100M and 9200M mobile parts only have 8 SP as well. Something like the new Ion platform (which is a 16SP 9400M with an Intel Atom CPU) probably represents a realistic lowest common denominator if you are looking for a “low end” reference platform to work on.

Thanks to both of you. Do you know which one also has the lowest clock rate? Is there a table anywhere of all the CUDA-enabled parts, the # of SP’s and clock rate/memory bus width and clock?

Thanks again…

This Wikipedia page is fairly comprehensive:


Whoa! PERFECT!!!