Integrated acceleration and offloading
Applies to: SQL Server 2022 (16.x)
SQL Server 2022 (16.x) provides a framework for offloading specific SQL Server workload compute to hardware devices.
This article explains offloading and acceleration to hardware devices.
SQL Server 2022 (16.x) enables integrated offloading and acceleration with Intel® QuickAssist Technology (QAT). You can offload backup compression with Intel® QAT to reduce CPU usage, backup completion times, and reduce storage consumption. To back up or restore databases with accelerated compression, install Intel® QAT drivers. For specific information and instructions to use with SQL Server, see Configure Intel® QuickAssist Technology (QAT).
Back up and restore databases
One use case for integrated acceleration and offloading is backup compression. Compression speeds up back up and restore operations, and saves space. Compression also consumes CPU resources which may impact application performance. Because Intel® QAT is an integrated acceleration and offloading solution, you can use it to back up and restore databases with compression, and reduce the load compression places on the CPU.
Backup compression is enabled by including the COMPRESSION keyword for BACKUP T-SQL commands, and is now available with an ALGORITHM parameter to choose between MS_XPRESS (default) and QAT_DEFLATE. For information about how to compress backups, see Configure Backup Compression (SQL Server).
Edition specific capabilities
Integrated acceleration and offloading are supported starting with SQL Server 2022 (16.x) on Windows.
Preview software is Evaluation Edition. It supports:
- Hardware offloading - can use a physical device. If no device is available, it falls back to software based compression.
- Software compression.
- Compressed backups to be restored if the drivers are available.
Hardware is not required to successfully restore a previously compressed backup. However, to back up or restore databases with accelerated compression, you must install product drivers.
Integrated acceleration and offloading are not available on Linux or containers.