emulation builds

Accelerated Computing CUDA CUDA Programming and Performance
1

I’m wondering how optimized the emulation builds are that allow an app to run without the GPU?

For instance are they going to be as fast as optimized code using Intel’s IPP libraries or are they just serial implementations?

If they were optimized (like or using IPP), then I could just use one API and not two.

Any ideas?

2

They are amazingly slow External Image – only AFAIK intended for certain aspects of debugging, useful as such, but SLOW.