Initially released two years ago was an Intel QuickAssist/QAT adaptation for Zstd compression. The v0.1 release happened just over two years ago followed by a v0.2 release one year later… Since then, it had been all quiet. Arriving today to surprise is the QAT Zstd 1.0 plug-in release.
QAT Zstd 1.0 is now available for those interested in Zstd compression offloading to QuickAssist Technology accelerators. At least going off the data shared by Intel back in 2023, there are some nice performance wins compared to just leveraging Zstd on the CPU:
This plug-in works by tapping into the Zstd API with its ability to handle custom sequence producers. This Zstd acceleration works on QAT found with 4th Gen Xeon Scalable and newer.
The QAT Zstd 1.0 plug-in release now has NUMA-aware memory allocations and other code optimizations. Plus there is support for Zstd 1.5.7 that was released back in February. Plus various bug fixes and static analysis fixes as part of OpenScanhub and Coverity scans.
Not the most exciting update for breaking the big “v1.0” milestone but the NUMA node aware allocations and Zstd 1.5.7 support are useful albeit overdue. Those interested in this QAT Zstd plug-in can find it via GitHub.
