VECTORIOYou want to rerun the numbers yourself.
Numbers you can check.
Every number we publish comes with the machine, the data, the layout and the run policy — so you can reproduce it.
Measured on one Apple M3 Max-class notebook, 16 cores, 64 GB, CPU only; 1 cold + 5 warm runs; logical decoded GB/s. Method and raw table →
Analytical paths. One notebook.
Apple M3 Max class, 16 cores, 64 GB, CPU only. Compressed columnar storage, dictionary encoding, declared sort layout. 1 cold + 5 warm runs; best and median kept; throughput as logical decoded-equivalent GB/s.
| Shape | Rows | Layout | Best | Median | Throughput |
|---|---|---|---|---|---|
| byte_key × SUM(Int32) | 1 B | sorted dictionary key | 32.0 ms | 32.1 ms | 116 GB/s |
| byte_key × SUM(Int64) | 1 B | sorted dictionary key | 69.1 ms | 69.4 ms | 118 GB/s |
| byte_key × SUM+MIN+MAX+COUNT(Float) | 1 B | sorted dictionary key | 67.2 ms | 69.7 ms | 112 GB/s |
| same shape, unsorted | 1 B | unsorted | 71.2 ms | 78.4 ms | 52 GB/s |
| town × SUM(price) | 100 M | sorted string dictionary | 3.6 ms | 3.6 ms | 108 GB/s |
| property_type × MIN/MAX/AVG/COUNT | 100 M | sorted string dictionary | 3.4 ms | 3.4 ms | 111 GB/s |
| Q1-style linestatus reduction | 100 M | dictionary key + predicate | 16.0 ms | 16.3 ms | 99 GB/s |
Internal measurements of selected execution paths under a declared layout. Sorted layouts are part of the benchmark; ad-hoc unsorted queries run at 45–80 GB/s.
A whole pipeline. 24.4 seconds.
103 million input rows through a 21-stage workflow — PII protection, deduplication, enrichment, SQL aggregation, quality checks and several outputs — on one notebook: about 4.2 million rows per second end to end, including orchestration and checkpointing.
Full stage report on request.
10,000 nodes. Real time.
| Beethoven workflow | Nodes | Active EQs |
|---|---|---|
| 10-node reference chain | 10 | 1 |
| 1,000 nodes, 1 active EQ | 1,000 | 1 |
| 1,000 nodes, 100 active EQs | 1,000 | 100 |
| 1,000 nodes, 333 active EQs | 1,000 | 333 |
| 10,000-node chain | 10,000 | configurable |
Demo: measured warm, 48 kHz, one process, ARM64 SIMD on Apple silicon. The 10,000-node variant played the full excerpt with zero buffer underruns.
Where it runs. x64 and ARM64.
| Platform | x64 | ARM64 | Notes |
|---|---|---|---|
| Linux | ✓ | ✓ | Servers and edge; Raspberry Pi-class boards in beta |
| Windows | ✓ | ✓ | Engine and apps |
| macOS | ✓ | ✓ | Engine and apps; benchmarks above |
| Cluster | beta | 4–7 peers demonstrated; the formal scaling benchmark will be published here | |
Reproduce it on your hardware.
The evidence pack holds the benchmark brief, the data generators and the run policy. Request the evidence pack
Bring us your hardest data problem. Let's solve it together.
A short pilot on your data and your hardware. We agree on what success looks like before we start.