Optimizing GPU Memory Allocation and Movement using SYCL

Rakshith Krishnappa, Developer Evangelist, AI and oneAPI
Intel Logo Blue

Discover how to use device memory allocation and data movement to achieve top performance from Intel® GPUs.

Building high-performant software applications requires coding to take maximum advantage of the target hardware platform. With the availability of hardware acceleration capabilities—in this case, provided by the latest generation GPUs from Intel—developers can optimize the performance of a software application by managing the allocation of memory for the device and controlling data movements between host and device through coding techniques.

This workshop is led by Rakshith Krishnappa, Developer Evangelist for HPC and oneAPI. The session advances understanding of the methods by which Intel® GPU performance can be optimized within a SYCL* application. Using the Intel® Developer Cloud with access to Intel® Data Center GPU Max series, this live virtual workshop delivers explanations and hands-on experience with SYCL programming techniques to demonstrate optimization methods.

The session covers:

  • Understanding how the use of Buffers and Unified Shared memory models affects performance.
  • Determining the best way to move memory between the host and device.
  • Examining various methods for allocating memory on host and device, while minimizing moving memory back and forth between the two.
  • Learning how to overlap date transfers effectively from the host to device.

An Intel® Developer Cloud account is necessary for engaging in hands-on sessions. Space is limited, so be sure to sign up promptly. Need help? Check out the detailed signup instructions here.