TETRAbase · The database
Graph connectivitySQL ubiquityUniversal simplicity
01 · A Tuesday morning
TETRAbase speaks PostgreSQL and openCypher over one store. Point your existing tools at it. No indexes or caches to maintain, encrypted on disk, in a fraction of the space. In our benchmarks, 82x faster than PostgreSQL and 66x faster than Neo4j.
Harvest Table is an invented example.
02 · Why it is faster
TETRAbase stores tables and graphs in a single matrix and answers queries by performing operations on that matrix. SQL and openCypher use the same stored data, so there is no need to copy data between them.
A query that follows connections does not chase pointers from record to record. On the standard graph benchmark the lead grew with size: 29.5x at the smallest, 66x at SF10.
Deep queries do not pull the graph into memory. On the same queries TETRAbase peaked at less than half the memory of Neo4j and used 13x less CPU.
03 · The numbers
Graph · By the Numbers, OpenCypher
66x
faster than Neo4j
21 LDBC graph queries at SF10
13x
less CPU than Neo4j
Server CPU for the same 21 queries at SF10
10.3x
more reads a second than Neo4j
8 concurrent LDBC readers at SF1
1.78B
edges on a single machine
Time for the 21 graph queries, at five sizes
SQL · TETRA, By the Numbers
82x
faster than PostgreSQL
251 SQL scenarios, 4 million rows
8.8x
smaller on disk than PostgreSQL
TETRAbase74 MiB
PostgreSQL653 MiB
The same 4 million rows, drawn as areas
DataPower · the open benchmark
520
DataPower at 3 million data elements
Quick test profile, one run per database
DataPower scores the useful work a database does for the hardware and setup it needs. We tested five databases on the same data and operations, each with 2 CPU cores and 8 GiB of memory.
How DataPower works →See the graph benchmark against Neo4j → Download the SQL report, 12 October 2026 (PDF) ↓
Graph figures from By the Numbers, OpenCypher, 1 November 2026; SQL figures from TETRA, By the Numbers, 12 October 2026. One run on a single machine. The competing engines were benchmarked once, and those results are shown here. The full report covers the test conditions, limitations, and cases where TETRAbase performs worse. DataPower: Quick test profile, one run per database and size, 1 November 2026.
Bring a workload. We’ll compare TETRAbase with your current system and share the results with you.