Back in April PostgreSQL added AVX-512 support for CRC32 computations. At the time the gains for CRC32 computations with this popular open-source database server were reported to be 50% to 3x faster for x86_64 CPUs able to leverage AVX-512. That AVX-512 support is found with PostgreSQL 18.0 that released in September and now Intel is praising this addition to PostgreSQL for which their developers also had a part in along with AWS and others.
Intel published a blog post today outlining the AVX-512 acceleration now found in PostgreSQL for CRC32C checksums. The addition is praised as a “clear and substantial performance improvement for a wide range of realistic buffer sizes (from very small to moderately large). This patch is a state-of-the-art example of utilizing modern SIMD instructions to accelerate critical database primitives, with a focus on portability, correctness, and runtime efficiency.“
The Intel-published results are indeed very positive for PostgreSQL with their tests on Xeon processors:
Another great showing for AVX-512 on real-world software with modern CPUs just like with AVX-512 acceleration for JSON parsing and much more over the years.
This AVX-512 support also works on AMD Zen 4 and newer processors — EPYC servers as well as client Ryzen processors thanks to their widespread AVX-512 support / unified ISA across product lines.
Those wanting to dive into all of Intel’s data and commentary around AVX-512 accelerated checksums in PostgreSQL can find today’s post at community.intel.com.
