|
7# voodoo12345
natively support c++ = cuda compiler can compile c++ program. In gt200, it can only compile c program. That is the only difference. Optimization is not there initially. But as an open source community, anyone can contribute to an optimized extension. So one day, there might be a huge open library that provides optimized api. Currently, video transcoding, adobe ps4, flash, matlab, etc, are such examples. The selling point of gpu in the future would like this, if one bot a SOC (junior cpu + senior gpu), one could speed up one of his existing/developing C++ programs by calling a function in an c++ cuda extension. If that gives you even 2x speedup, that sounds valuable. Let alone, future operation systems running faster in SOC than intel only. |
|