The PostgreSQL Engine on FPGA
We ported the core PostgreSQL engine directly onto FPGA via our Kubernetes platform. The goal: accelerate the heaviest queries by running critical operations at the hardware level.
What we did
- Ported the SQL engine onto Alveo U250 FPGA via Azure NP10s
- CI/CD pipeline to compile, test and deploy the PostgreSQL kernel in under 10 minutes
- Benchmark on massive OLAP workloads — aggregations, complex joins, table scans
Results
Against the CPU, the results are not comparable. The FPGA executes the same queries with latency that even high-end CPUs cannot approach.
The CPU didn’t lose. It’s just not in the same category.
Why it matters
Databases are the heart of every enterprise application. Accelerating PostgreSQL at the hardware level, without changing a single line of SQL, delivers an immediate gain to millions of existing applications.
Technologies
- AMD Alveo U250 FPGA
- Microsoft Azure NP10s VMs
- Kubernetes (AKS) with FPGA operator
- PostgreSQL 16
- GitLab CI/CD pipeline