Performance

We benchmark Traverse against Neo4j and Memgraph on a real social network dataset: 100,000 users, 1.77 million connections, 33 queries. Everything runs on the same machine, from automated and repeatable benchmark suites.

Up to 376x

Faster deep traversals

4-hop queries vs Neo4j

Up to 136x

Faster aggregations

Full-graph analytics vs Memgraph

7.3x

Geometric mean vs Neo4j

Across all 33 queries, single client

Why It's Fast

Cache-Friendly Storage

Connections between nodes are stored in tightly packed, contiguous memory. When Traverse follows a relationship, the next one is already loaded - no waiting for memory.

Multi-Core Execution

Large queries are automatically split across all available CPU cores. The optimizer decides when parallelism helps and how to divide the work efficiently.

Smart Data Access

The engine reads only the data each query actually needs, in the most efficient format possible. Redundant work is eliminated automatically.

Built in Rust

It hits full speed on the very first query, so there's nothing to warm up. And because it never pauses to tidy up after itself the way other engines do, response times stay steady instead of spiking every so often.

Non-Blocking Transactions

Reads and writes happen concurrently without blocking each other. Your application stays fast even when data is being updated.

Efficient Wire Protocol

Results are streamed directly to the client with minimal overhead. Large result sets don't consume extra memory on the server.

Benchmark Results

Every engine ran the same 33 Cypher queries over the Bolt protocol, one at a time on the same machine. Higher throughput is better; the multiplier is how many times faster Traverse is on that query.

7.3×faster than Neo4j, geomean
3.8×faster than Memgraph, geomean
up to 376×on 4-hop traversals
Category
Parallel clients
TraverseMemgraphNeo4j
Single Vertex Read
Traverse
56.3Kfastest
Memgraph
32.0K1.8×
Neo4j
16.4K3.4×
Single Vertex Property
Traverse
55.9Kfastest
Memgraph
31.7K1.8×
Neo4j
18.9K3.0×
Pattern Short
Traverse
53.5Kfastest
Memgraph
28.8K1.9×
Neo4j
18.5K2.9×
Pattern Cycle
Traverse
27.4Kfastest
Memgraph
18.6K1.5×
Neo4j
10.8K2.5×
Pattern Long
Traverse
47.2Kfastest
Memgraph
25.5K1.9×
Neo4j
18.1K2.6×
All 33 queries · 1 clientTraverse wins 31/31 vs Memgraph · 3.77× geomeanTraverse wins 32/33 vs Neo4j · 7.31× geomean

Bars are scaled to the fastest engine in each row, so slow and fast queries stay readable. The multiplier is how Traverse compares on that query - accent when it is ahead, muted and below 1× when a competitor is faster (and its bar is the longest). The win counts are honest, never 33 of 33. Full hardware and dataset details are in the Test Environment below.

Graph Algorithms

Since 0.7.0, Traverse ships its own algorithm library, callable directly from Cypher. The per-algorithm numbers are published with each release.

36+

Algorithms built in

Centrality, communities, paths, similarity, embeddings

0

Plugins required

APOC and GDS surfaces ship in the binary

Test Environment

Hardware

Intel Core i7-12700KF (12 cores, 20 threads), 64 GB DDR5 memory, Ubuntu 24.04 LTS.

Dataset

Pokec social network: 100,000 users, 1.77 million connections. Each query measured over 10 seconds with warm-up runs. Fresh data for each test.

Compared Against

Memgraph 3.8.1 and Neo4j 2026.01.4, both with default settings. All 33 queries sent via the Bolt protocol for a fair comparison. Numbers recorded July 2026.

Run your own benchmark

The traverse-bench tool downloads the same datasets and runs the same suite, so you can check our numbers on your own hardware.