Fermi Warp Sheduling

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello, I write GPGPU reviews for my master’s diploma. And I have questions about dual warp sheduler on Fermi.

[list=1]

[*]Why double-precision instructions can’t be dual-issued with another instructions, such as transcendent functions, or load/store ones?

[*]Are double-precision instructions executed connecting two cores (like on AMD GPUs)?

If possible, please tell about sm2.0 and sm2.1 respectively.

Thanks, and sorry for my poor english External Image

2

Hi all, I’ve found answwers for those questions in article Inside Fermi: Nvidia's HPC Push
External Image