SURREALDB 3.0
A new foundation for performance
Graph queries up to 24x faster. Query planning 4,600x smarter. Vector search ~8x faster. Built on a rearchitected execution engine designed for production-grade workloads.
SurrealDB 3.0 introduces a new streaming execution engine with a standard AST → LogicalPlan → ExecutionPlan pipeline, delivering dramatic performance improvements across every workload.
Performance at a glance
Key improvements in SurrealDB 3.0 vs. SurrealDB 2.0 across all storage engines.
4–24x
Faster graph queries
Across all traversal depths and patterns
O(n) to O(1)
Faster WHERE id = queries
Smarter planner queries now sub-millisecond
3–7x
Faster table scans
With LIMIT, START, and combined operations
~3x
Faster ORDER BY
Across all storage engines
~8x
Faster vector search
HNSW indexed queries: ~36s → ~4.5s
~36%
Faster GROUP BY
Multi-aggregation and dedup workloads
Why these benchmarks matter
Enterprises like Tencent and NVIDIA run SurrealDB in production across high-throughput, business-critical workloads. As adoption grows, demonstrating predictable, production-grade performance matters more than ever. SurrealDB 3.0 is the next major step - sharing both the results and the architectural changes behind them.
Benchmarking a multi-model database
SurrealDB unifies relational, document, graph, vector, geospatial, and full-text workloads in one engine. Fair benchmarking is complex - databases vary in durability guarantees and configuration trade-offs. We've kept comparisons balanced and transparent.
Customers like Tencent reduced nine separate tools down to one with SurrealDB.
The new execution engine
SurrealDB 3.0 rearchitects query execution around a standard AST → LogicalPlan → ExecutionPlan pipeline with optimisations at each stage. The engine is fully streaming internally - end-to-end client streaming is coming in a future minor release.
Our open-source benchmarking tool
We built crud-bench in Rust to evaluate SurrealDB across its full workload spectrum - CRUD, graph traversals, vector search, and more - consistently across embedded, networked, and remote deployments.
Benchmark hardware
All benchmarks were run on the same hardware for consistency.
| Component | Specification |
|---|---|
| CPU | AMD Ryzen Threadripper 9970X |
| RAM | Samsung DDR5 4800MHz 64GB RDIMM |
| SSD | Lexar EQ790 4TB NVMe |
| Motherboard | ASUS PRO WS TRX50-SAGE WIFI A |
| CPU Cooler | Arctic Freezer 4U-M |
| Power Supply | CORSAIR RM1000x |
| GPU | PowerColor AMD Radeon RX 7600 Fighter 8GB |
| Case | In-Win IW-R400-01N 4U |
Benchmark results
SurrealDB 3.0 vs. SurrealDB 2.0 across three storage engines. Lower latency and higher throughput is better.
Create
2.0 Mean
1.94 ms
3.0 Mean
0.72 ms
Latency
+62.9%
Throughput
+168.7%
Update
2.0 Mean
2.58 ms
3.0 Mean
0.73 ms
Latency
+71.7%
Throughput
+249.6%
WHERE id = record:42
2.0 Mean
3,936 ms
3.0 Mean
0.68 ms
Latency
+100.0%
Throughput
+436,303%
LIMIT (all fields)
2.0 Mean
17.45 ms
3.0 Mean
2.46 ms
Latency
+85.9%
Throughput
+550.9%
ORDER BY
2.0 Mean
17,271 ms
3.0 Mean
5,275 ms
Latency
+69.5%
Throughput
+232.3%
ORDER BY (multi)
2.0 Mean
17,388 ms
3.0 Mean
5,193 ms
Latency
+70.1%
Throughput
+235.6%
Graph out depth 3
2.0 Mean
18.02 ms
3.0 Mean
3.02 ms
Latency
+83.2%
Throughput
+449.0%
Graph multi out
2.0 Mean
148.68 ms
3.0 Mean
23.32 ms
Latency
+84.3%
Throughput
+504.6%
Graph depth 2 + LIMIT
2.0 Mean
78.83 ms
3.0 Mean
9.41 ms
Latency
+88.1%
Throughput
+696.2%
GROUP BY (multi agg)
2.0 Mean
9,738 ms
3.0 Mean
6,203 ms
Latency
+36.3%
Throughput
+53.7%
GROUP BY (dedup agg)
2.0 Mean
9,735 ms
3.0 Mean
6,278 ms
Latency
+35.5%
Throughput
+54.0%
HNSW vector search
2.0 Mean
38,582 ms
3.0 Mean
4,847 ms
Latency
+87.4%
Throughput
+671.2%
SELECT FETCH
2.0 Mean
39.07 ms
3.0 Mean
7.58 ms
Latency
+80.6%
Throughput
+401.8%
Explore the data
Filter by engine and category. Hover for values and drag to zoom.
What's next
Performance improvements are rolling out across coming releases, with broader cross-database benchmarks to follow.
Write support
Expanding the new executor to all statement types.
E2E streaming
From internal execution all the way to the client.
Indexing
Better planning, reduced scan overhead, new index types.
Pipeline
Continued optimisation of planning and execution.
Related resources
Dive deeper into SurrealDB 3.0 and our performance journey.
