Published in News

Intel calls it a Knight on “HPC chip of the future"

by on28 May 2024


It’s been a hard days knight

Chipzilla has withdrawn support for its Xeon Phi Knights Mill and Knights Landing accelerators from the LLVM/Clang 19 compiler as a sign that its MIC architecture is toast.

Intel once thought that MIC architecture would be the future of high-performance computing (HPC) particularly when it was the centrepiece of the Aurora supercomputer. However, with that supercomputer now up for sale it looks like MIC will go with it.

The Knights Mill processors experienced numerous setbacks and failed to achieve performance objectives, leading to the abandonment of the first iteration of the Aurora supercomputer after several years of postponements.

The Department of Energy (DoE) subsequently revised Aurora's design to incorporate Intel's Sapphire Rapids and Ponte Vecchio compute GPUs, yet due to issues with these CPUs and GPUs, Aurora encountered further delays and performance challenges.

These hardware complications, cooling system failures, and other stability concerns continue to hinder Aurora from realising its full capabilities. Nevertheless, the system is now poised for complete deployment within this year, nine years since its initial announcement.

Intel officially declared the end of life for these products around August 2017, never making complete clang/llvm's support for these products.

Chipzilla recently announced that the GCC compiler ceased support for Intel's Xeon Phi accelerators, possibly due to its discontinuation of hardware support. The phasing out commenced earlier this year with LLVM/Clang 18, and the full removal is scheduled for the LLVM 19 release in September.

Phoronix thinks that the withdrawal of support for Xeon Phi began with LLVM/Clang 18, which was designated deprecated. This is in line with GCC's strategy, which deprecated Xeon Phi in version 14 and eliminated it in version 15. Intel's decision to remove support in LLVM/Clang 19 mirrors a wider trend across compilers as the firm concentrates on precise AI and HPC.

Additional factors include the prior discontinuations of support in ICC and ICX compilers, which also generated errors when interfacing with these accelerators. Intel has highlighted that this removal will diminish maintenance efforts, streamlining the development and support processes for present and forthcoming compiler versions.

This signifies the conclusion of an extensive journey for the Larrabee-derived Xeon Phi products, which Intel officially declared not to be in 2019.

 

Last modified on 28 May 2024
Rate this item
(0 votes)

Read more about: