19.11.2014 Views

Tutorial: Introduction to CUDA Fortran | GTC 2013

Tutorial: Introduction to CUDA Fortran | GTC 2013

Tutorial: Introduction to CUDA Fortran | GTC 2013

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Unified Virtual Addressing<br />

• CPU and GPUs use a unified virtual address space<br />

– Device can be determined from address of data<br />

– Requires: <strong>CUDA</strong> Toolkit >= 4.0, compute capability >= 2.0, 64-bit OS<br />

• Enables peer-<strong>to</strong>-peer computing<br />

GPU<br />

GPU<br />

GPU<br />

GPU<br />

– Direct Transfers: copy between<br />

GPUs not staged through host<br />

Memory<br />

Memory<br />

Memory<br />

Memory<br />

– Direct Access: a kernel on one GPU<br />

can access memory on another<br />

PCI<br />

PCI<br />

CPU<br />

CPU<br />

Memory<br />

Memory

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!