Webinar - PyTorch Performance on CDNA2 and RDNA3

Calculating spectra of fluids in irregular geometry is a challenging task that requires the estimation of the eigenmodes of a Laplacian operator. For most situations, obtaining all of the eigenmodes requires too much memory and we are comfortable making the choice to examine only a subset of the kinetic energy spectra; dividing the spectra into modes that we can calculate and a portion that we deem unresolved. To find a subset eigenpairs while making efficient use of memory, we can use a matrix-free implementation of the implictly restarted Lanczos method. This webinar discusses the algorithm and the intended use case for this code to motivate the selection of PyTorch as the implementation framework. Next, we explore how to install ROCm and Pytorch with ROCm backends on CDNA2 and RDNA3 platforms that are both hosted within Fluid Numerics' on-premise cluster resources. The algorithm that was initially implemented using numpy, numba, and scipy is compared with our implementation in PyTorch in terms of runtime and energy consumption.
***
Subscribe: bit.ly/Subscribe_to_AMD
Join the AMD Red Team Community: www.amd.com/en/gaming/community
Join the AMD Red Team Discord Server: / discord
Like us on Facebook: bit.ly/AMD_on_Facebook
Follow us on Twitter: bit.ly/AMD_On_Twitter
Follow us on Twitch: / amd
Follow us on LinkedIn: bit.ly/AMD_on_Linkedin
Follow us on Instagram: bit.ly/AMD_on_Instagram
©2024 Advanced Micro Devices, Inc. AMD, the AMD Arrow Logo, and combinations thereof are trademarks of Advanced Micro Devices, Inc. in the United States and other jurisdictions. Other names are for informational purposes only and may be trademarks of their respective owners.