// Benchmarks

Quanton performance.

Compare execution times across Spark engines on the same hardware, from analytical queries to lakehouse writes.

// Benchmark proof · TPC-DS 10 TB

Total runtime · 99 queries · TPC-DS 10 TB

11 × m8gd.4xlarge · any open format · same hardware throughout

Quanton Photon Open / accelerated Spark

OSS Spark

12,200s

5.43×

baseline

Comet (Tuned)

9,122s

4.06×

OOMs on q67, q93

Gluten (Tuned)

8,563s

3.81×

OOMs on q67, q93

Databricks Photon 18.2

2,550s

1.13×

closed, paid

Quanton

2,247s

1.00×

BYOC, per-GiB

Quanton completes the suite in 2,247 seconds, compared with 2,550 seconds for Photon and 12,200 seconds for open-source Spark on the same hardware.

2.0×

TPCx-BB · 1 TB

SF1000 Iceberg · 10 r8g.4xl executors · Spark 3.5. Quanton vs Apache Spark.

2.05×

TPC-DI · 1 TB

Python transformations + TPC-H-like SQL. Native rollup avoids 9× row replication.

4.1×

LakeLoader · 1 TB

MERGE INTO / CDC updates. Low-shuffle columnar MERGE scales with rows changed.

~4×

Iceberg / Hudi compaction

Native columnar compaction · ~75% less wall-time vs OSS row-wise rewrite.

Quanton v0.37.0 result with comparison runs from the Inside Quanton — Storage-aware Spark design-partner deck. Independent re-runs welcome — Quanton runs in your own VPC.