Head to head · Figures checked September 2026
Databricks Photon vs RAPIDS Accelerator
ETL Compare staff · Figures checked September 2026 · Sourced from vendor docs, project pages and public benchmarks · Published 29 September 2026
In brief
RAPIDS Accelerator has the higher ETL fit score on our published weights (3.7 against 3.4 out of 5). Databricks Photon scores higher on tuning and operating burden; RAPIDS Accelerator scores higher on adoption effort, stage coverage, platform and instance portability, cost model transparency and published evidence. They tie on maturity and community. Photon only runs inside Databricks, so for most teams the platform they use decides this comparison before the scores do.
On our weights RAPIDS Accelerator ranks 3rd of 7 and Databricks Photon ranks 6th of 7. Which one fits depends on the criteria below and on the platform you run.
Databricks Photon is a Databricks-native vectorized C++ query engine inside the Databricks Runtime. RAPIDS Accelerator for Apache Spark (now NVIDIA cuDF for Apache Spark) is an open-source NVIDIA plugin that runs supported Spark SQL and DataFrame operations on GPUs. Neither requires changes to Spark SQL or DataFrame code, according to its own documentation.
3.4 / 5
ETL fit score (editorial assessment, 0-5)ETL fit score (editorial assessment, 0-5)
Rank 6 of 7
Best for teams already on Databricks
3.7 / 5
ETL fit score (editorial assessment, 0-5)ETL fit score (editorial assessment, 0-5)
Rank 3 of 7
Best for teams that already run GPU capacity
How do Databricks Photon and RAPIDS Accelerator score on each criterion?
| Criterion | Photon | RAPIDS Accelerator | Higher score |
|---|---|---|---|
| Adoption effort 20% | 2.6 | 3.4 | RAPIDS Accelerator |
| Stage coverage 20% | 3.8 | 4.2 | RAPIDS Accelerator |
| Tuning and operating burden 15% | 4.6 | 3.0 | Photon |
| Platform and instance portability 10% | 1.6 | 3.2 | RAPIDS Accelerator |
| Cost model transparency 10% | 3.2 | 4.2 | RAPIDS Accelerator |
| Published evidence 15% | 3.2 | 3.6 | RAPIDS Accelerator |
| Maturity and community 10% | 4.6 | 4.6 | Tie |
| ETL fit score | 3.4 | 3.7 | RAPIDS Accelerator |
Why each score
- Adoption effort
- Photon (2.6): One checkbox (or on by default on serverless) with no code changes if you already run on Databricks; for a team on EMR or self-managed Spark it means moving to Databricks first.
- RAPIDS Accelerator (3.4): No code changes, but the job has to move to NVIDIA GPU instances and the cluster needs GPU-specific configuration.
- Stage coverage
- Photon (3.8): Documents native scans (Parquet, Delta, CSV, JSON), filters, joins, aggregates, windows and writes, with fallback to the Spark runtime; no UDF, RDD or stateful streaming support.
- RAPIDS Accelerator (4.2): Documents GPU execution for group by, joins, sorts and windows, Parquet and ORC writing, CSV reading and a RAPIDS Shuffle Manager; unsupported operations fall back to CPU.
- Tuning and operating burden
- Photon (4.6): Managed by Databricks inside its runtime, so there is no plugin for your team to install or version.
- RAPIDS Accelerator (3.0): Qualification and Profiling tools help, but GPU sizing and plugin configuration are your team's job.
- Platform and instance portability
- Photon (1.6): Available only inside Databricks (on AWS, Azure and Google Cloud).
- RAPIDS Accelerator (3.2): The widest platform list in this set (EMR, Databricks, Dataproc, GKE, Azure Synapse, Kubernetes, on-premises, OCI), held back because every one of them needs GPU instances.
- Cost model transparency
- Photon (3.2): Databricks publishes DBU rates, but Photon compute consumes DBUs at different rates, so the net effect has to be modelled per job.
- RAPIDS Accelerator (4.2): The plugin is Apache 2.0 with no fee; the cost question becomes GPU instance price against runtime saved, which you can model from public cloud prices.
- Published evidence
- Photon (3.2): Databricks documents where Photon does not help (queries under two seconds, UDFs); performance comparisons outside Databricks are not possible because it only runs there.
- RAPIDS Accelerator (3.6): The Qualification Tool estimates fit from your own event logs; headline benchmark figures were not reviewed for this edition.
- Maturity and community
- Photon (4.6): The default engine on Databricks serverless and SQL warehouses, backed by Databricks.
- RAPIDS Accelerator (4.6): A long-running NVIDIA project with more than 9,000 commits on main, now published as NVIDIA cuDF for Apache Spark.
What do Databricks Photon and RAPIDS Accelerator cost, as published?
| Photon | RAPIDS Accelerator | |
|---|---|---|
| License | Commercial (part of Databricks) | Apache License 2.0 |
| Published price | Databricks DBU pricing; Photon changes DBU consumption | No plugin fee; GPU instance pricing applies |
| Instances | Supported Databricks instance types, including Graviton | NVIDIA GPU instances (Volta or later) |
| Runs on | Databricks only (AWS, Azure, Google Cloud) | Amazon EMR, Databricks, Dataproc, GKE, Azure Synapse, Kubernetes, on-premises, OCI |
| Code changes | None on Databricks; platform move otherwise | None; plugin replaces internal physical plan parts |
Prices and terms as published on the pages we reviewed, 27 September 2026. Neither vendor's figure is a quote.
Sources: Photon documentation, spark-rapids overview, RAPIDS Accelerator user guide, RAPIDS Accelerator FAQ, NVIDIA/spark-rapids on GitHub · Fetched 27 Sep 2026
Where are they documented to run?
| EMR | Databricks | Google Cloud (Dataproc) | AWS Glue | Self-managed Spark / Kubernetes | |
|---|---|---|---|---|---|
| Photon | Not available | Only platform | Not available | Not available | Not available |
| RAPIDS Accelerator | Documented | Documented | Documented | Not documented | Documented |
Documented means the vendor or project lists the platform on the pages we reviewed. Self-install means you can usually add an open-source plugin to a platform that lets you set Spark configuration and classpath, but the project does not publish a guide for that platform. None of the vendor pages we reviewed list AWS Glue.
Which job stages does each address?
| Accelerator | Read | Transform | Shuffle | Spill | Write |
|---|---|---|---|---|---|
| RAPIDS Accelerator | Documented | Documented | Documented | Not stated | Documented |
| Photon | Documented | Documented | Not stated | Not stated | Documented |
- Documented: the vendor's own documentation says it addresses this stage
- Partial: indirect or experimental, or covered only by an end-to-end claim
- Not stated: not found on the pages we reviewed
This shows what each vendor says, not what we measured. Sources are listed on each review.
Choose Databricks Photon if
- Your jobs already run on Databricks and you want no plugin to install or version
- You want the engine that is the default on Databricks serverless and SQL warehouses
- You run on Databricks
Choose RAPIDS Accelerator if
- You want a long-running project with more than 9,000 commits on main
- You want GPU execution documented for joins, sorts, aggregations, window functions, Parquet and ORC writing and shuffle
- You run on Amazon EMR, Databricks, Google Cloud (Dataproc) and self-managed Spark or Kubernetes
Frequently asked questions
Which has the higher ETL fit score, Databricks Photon or RAPIDS Accelerator?
RAPIDS Accelerator, with 3.7 against 3.4 out of 5 on our published weights. Databricks Photon scores higher on tuning and operating burden and RAPIDS Accelerator on adoption effort, stage coverage, platform and instance portability, cost model transparency and published evidence. They tie on maturity and community. The score measures fit for speeding up existing Spark ETL, not raw speed.
Do Databricks Photon and RAPIDS Accelerator run on the same platforms?
Both are documented for Databricks. RAPIDS Accelerator is also documented for Amazon EMR, Google Cloud (Dataproc) and self-managed Spark or Kubernetes.
What do Databricks Photon and RAPIDS Accelerator cost?
Databricks Photon: Databricks DBU pricing; Photon changes DBU consumption (Commercial (part of Databricks)). RAPIDS Accelerator: No plugin fee; GPU instance pricing applies (Apache License 2.0). For a like-for-like comparison, work out cost per run on one of your own jobs: see Spark cost per job, explained.