Skip to content
Spark Acceleration Compare

Side by side, read 10 Nov 2026

Apache Auron and NVIDIA cuDF plugin for Apache Spark: the published facts in the same rows

Apache Auron profileNVIDIA cuDF plugin for Apache Spark profile

The same rows for both

Options in alphabetical order. Vendor statements are the vendor's own. Read 10 Nov 2026.
RowApache AuronNVIDIA cuDF plugin for Apache Spark
KindOpen-source Rust engine plugin for SparkGPU plugin, formerly RAPIDS Accelerator
Does wellThe project publishes its TPC-DS 1 TB method, with data generation and runner tools in the auron-project/tpcds-tools repository and the full Spark configuration. Its build script lists Spark 3.0, 3.1, 3.2, 3.3, 3.4, 3.5, 4.0, 4.1 and 4.2 with Scala 2.12 and 2.13.AWS's August 2026 post publishes runtime and cost per run: 281 seconds and $2.06 on GPU versus 1,010 to 1,043 seconds and $2.93 to $3.18 on CPU clusters. Amazon EMR on EKS supports the plugin natively from release 6.9.0, and NVIDIA lists Databricks 14.3 ML LTS and 17.3 ML LTS and Dataproc 2.1 to 2.3 as supported.
LimitedYou build the jar yourself (Rust nightly and a JDK are required), and the getting-started page describes building from source. Support for Amazon EMR, EKS or Databricks: none found on 2026-11-10 in the project's pages. The benchmark's per-query results are published as a chart image only.The plugin needs NVIDIA GPUs of the Volta generation or newer, so the cluster moves to GPU instances such as g7.4xlarge. Only the DataFrame API and a limited set of operations are accelerated; RDDs, GraphX and APIs that take Scala or Java functions are not.
Published speed resultsThe project says it has passed all TPC-DS and TPC-H cases and, compared with Spark 3.5, runs about 2x faster and saves about 50% of cluster resources. The only published detail is the TPC-DS 1 TB benchmark page; no per-query numbers are in text form (results are a chart).AWS and NVIDIA report that TPC-DS at 3 TB (103 queries) on 8-node Amazon EMR on EKS clusters ran in 281 seconds on g7.4xlarge GPU instances (NVIDIA RTX PRO 4500 Blackwell) versus 1,010 to 1,043 seconds on comparable CPU instances, which AWS labels a 3.7x speedup. Median of 5 iterations, Spark 3.5.6, cuDF plugin 26.04.2, us-east-1. NVIDIA's own FAQ gives a general expectation and says there is no formula for a specific query. This is a vendor claim with no named benchmark behind it. NVIDIA's overview page compares the published AWS GPU result with separate NDS (TPC-DS derived) 3 TB runs of open-source Spark 3.5.6 on EC2. It says the GPU run is 8.4x faster than m7gd.8xlarge and 11.0x faster than m9gd.4xlarge, with 77% and 72% lower reported cost per run. NVIDIA itself notes the runs use different hardware and software stacks.
Published cost results and pricesNone found on 10 Nov 2026.AWS reports cost per benchmark run of $2.06 on the GPU cluster versus $2.93 to $3.18 on the CPU clusters, using On-Demand us-east-1 prices as of May 2026. The g7.4xlarge cluster costs $26.35 per hour (8 x $3.042 EC2 plus $2.01 for EMR on EKS) against $10.43 to $10.99 per hour for the CPU clusters. Cost excludes the EKS control-plane fee, EBS, S3 and the driver pod. The software itself is free; you pay for GPU instances.
Code changesAuron is added by putting a jar on the Spark classpath and setting spark.auron.enable, the AuronSparkSessionExtension and the AuronShuffleManager; no job changes. The project describes fine-grained fallback for unsupported operators, and the Spark UI has an Auron tab that records per-operator fallback reasons. You must build the jar yourself (Rust nightly and JDK required); the getting-started page describes building from source.NVIDIA says existing Spark applications run with no code change once the plugin is launched and enabled by a configuration setting. Unsupported operators and UDFs fall back to the CPU, so the job completes but those stages are not accelerated. Only the DataFrame API (Dataset[Row]) and a limited set of SQL and DataFrame operations are accelerated. Custom classes or types with Dataset, APIs that take Row or Scala/Java functions (flatMap, foreach, foreachPartition), GraphX and RDDs are not accelerated. Structured streaming is not a focus.
TuningThe suggested config disables Spark off-heap memory and moves memory into executor overhead (4096 MB in the quick start). The benchmark used executor memory 8g with overhead 16384 for Auron versus 20g with 4096 for plain Spark, plus spark.auron.memoryFraction 0.8, forceShuffledHashJoin and sort-merge-join and UDAF fallback settings. ORC with large string/binary values needs spark.auron.orc.batchSize lowered.The user guide has a tuning guide (executors, tasks per executor, pooled and pinned memory, concurrent tasks per GPU, shuffle partitions, input partition size, columnar batch size), an AutoTuner, a Profiling tool and a Qualification tool, plus a recommended-practices section on handling GPU out-of-memory errors. In the AWS test the GPU config set executor memoryOverhead 30G, pinnedPool 8G, concurrentGpuTasks 3 and RapidsShuffleManager, and AWS advises using the Spark History Server to tune concurrentGpuTasks and pinnedPool.size. AWS recommends a pre-migration assessment with the NVIDIA qualification tool to estimate GPU acceleration potential, and setting spark.rapids.sql.explain=NOT_ON_GPU to see which operations run on CPU.
Spark versions, platforms and table formatsThe build script lists supported Spark versions 3.0, 3.1, 3.2, 3.3, 3.4, 3.5, 4.0, 4.1 and 4.2 with Scala 2.12 and 2.13, and Celeborn 0.5/0.6 and Uniffle 0.10 remote shuffle. The 8.0.0 notes add Spark 4.0 and 4.1 build support. No EMR, EKS or Databricks support is documented. The 8.0.0-incubating release notes say it added and expanded native scan support for Apache Iceberg, Apache Hudi and Apache Paimon, including Copy-on-Write tables and Iceberg changelog scans. The homepage says it supports Hudi and Paimon. 8.0.0 also adds Flink integration. Delta Lake is not mentioned in the fetched pages.Release v26.10.0 supports Apache Spark 3.3.x, 3.4.x, 3.5.0 to 3.5.9, 4.0.0 to 4.0.4, 4.1.1 to 4.1.3 and 4.2.0 (Scala 2.12 and 2.13 depending on version). Supported Databricks runtimes on AWS and Azure are 14.3 ML LTS (Spark 3.5.0) and 17.3 ML LTS (Spark 4.0.0). Dataproc 2.1 to 2.3 and Dataproc Serverless runtimes 1.2, 2.2 and 3.0 are supported. Amazon EMR on EKS natively supports the plugin (release 6.9.0 or later; the '-spark-rapids' release variant ships it pre-installed). The user guide also has a getting-started page for AWS EMR (EC2) and covers Apache Iceberg and Delta Lake support pages, including reading and writing tables.
Where it runsBuild needs Rust nightly and JDK; the README says regularly tested on JDK 8, 11 and 17 and also on JDK 21. The build script supports OS images centos7, ubuntu24, rockylinux8, debian11 and azurelinux3, and macOS builds were added in 8.0.0. The homepage says it supports storage such as HDFS and S3. No cloud-specific instance guidance was found.Needs NVIDIA GPUs: Volta, Turing, Ampere, Ada Lovelace, Hopper and Blackwell generations; tested on V100, T4, A10, A100, L4, H100, B100, RTX PRO 4500 and RTX PRO 6000. Linux with glibc 2.28 or newer; NVIDIA driver R525+ (CUDA 12) or R580+ (CUDA 13). On AWS the benchmark used g7.4xlarge (1 GPU, 16 vCPU, 64 GB RAM); NVIDIA's EMR guide maps m5d.8xlarge to g4dn.8xlarge for T4 GPUs.
How open its benchmarks areThe benchmark page publishes the method end to end: TPC-DS 1 TB data generated with a datagen tool (derived from maropu/spark-tpcds-datagen) and run with a benchmark runner (derived from databricks/spark-sql-perf), both in the auron-project/tpcds-tools repo, with the full Spark config. It ran Spark 3.5.6 against Auron 6.0.0-preview on YARN. Node types and counts are not given (the config shows executor cores 8 and a very large maxExecutors).NVIDIA publishes the NDS benchmark automation (data generation, power and throughput runs) in a public Apache-2.0 repo, with a disclaimer that NDS is derived from TPC-DS and is not comparable to published TPC-DS results. The AWS post publishes the cluster, Spark config, EMR release and pricing method; it does not link scripts.
Change Budget: codeNo job code change stated. Auron is added with a jar on the Spark classpath and three settings (spark.auron.enable, the AuronSparkSessionExtension and the AuronShuffleManager); the project states no job changes.No code change stated. NVIDIA says existing Spark applications run with no code change once the plugin is launched and enabled by a configuration setting. Unsupported operators and UDFs fall back to the CPU.
Change Budget: clusterBuild the jar, set three configs. You must build the jar yourself (Rust nightly and a JDK). The quick start disables Spark off-heap memory and moves memory into executor overhead (4096 MB).GPU instances and plugin settings. The cluster needs NVIDIA GPUs (Volta to Blackwell generations) and a current driver. AWS's test used g7.4xlarge nodes with executor memoryOverhead 30G, pinnedPool 8G, concurrentGpuTasks 3 and RapidsShuffleManager.
Change Budget: cloudNone found on 2026-11-10. No cloud requirement is stated. The homepage says it supports storage such as HDFS and S3, and no cloud-specific instance guidance was found. Documented support for EMR, EKS or Databricks: none found on 2026-11-10.AWS, Azure and Google Cloud documented. NVIDIA lists Databricks runtimes on AWS and Azure and Dataproc on Google Cloud; Amazon EMR on EKS supports the plugin natively from release 6.9.0.
Change Budget: vendorApache-2.0, no contract stated. Auron is Apache-2.0 licensed under the Apache Software Foundation; no vendor or platform contract is described.Free software; GPU instances paid. The software is free (source Apache-2.0); you pay for GPU instances, and AWS's cost per run includes EC2 and EMR on EKS charges.
Ledger entriesAuron TPC-DS 1 TB against Apache Spark 3.5.6 (run by its maker)GPU Spark on Amazon EMR on EKS with G7 instances, TPC-DS 3 TB (run by its maker); NVIDIA comparison with open-source Spark on EC2, NDS 3 TB (run by its maker)

How to read this page

The rows hold what each vendor or a named third party published, with no conclusion drawn. Benchmarks in the two columns were run on different hardware, data and baselines unless the Ledger entry says otherwise, so the numbers are not comparable across columns. The Ledger method explains how entries are recorded.

Sources

  1. Apache Auron (incubating), formerly Blaze: auron.apache.org, read 10 Nov 2026
  2. Apache Auron (incubating), formerly Blaze: auron.apache.org/documents/benchmarks.html, read 10 Nov 2026
  3. Apache Auron (incubating), formerly Blaze: raw.githubusercontent.com/apache/auron/master/auron-build.sh, read 10 Nov 2026
  4. Apache Auron (incubating), formerly Blaze: auron.apache.org/archives/v8.0.0-incubating.html, read 10 Nov 2026
  5. Apache Auron (incubating), formerly Blaze: raw.githubusercontent.com/apache/auron/master/README.md, read 10 Nov 2026
  6. NVIDIA RAPIDS Accelerator for Apache Spark (now NVIDIA cuDF plugin for Apache Spark): aws.amazon.com/blogs/big-data/gpu-accelerated-apache-spark-with-amazon-e, read 10 Nov 2026
  7. NVIDIA RAPIDS Accelerator for Apache Spark (now NVIDIA cuDF plugin for Apache Spark): docs.nvidia.com/spark-rapids/user-guide/latest/faq.html, read 10 Nov 2026
  8. NVIDIA RAPIDS Accelerator for Apache Spark (now NVIDIA cuDF plugin for Apache Spark): docs.nvidia.com/spark-rapids/user-guide/latest/overview.html, read 10 Nov 2026
  9. NVIDIA RAPIDS Accelerator for Apache Spark (now NVIDIA cuDF plugin for Apache Spark): docs.nvidia.com/spark-rapids/user-guide/latest/index.html, read 10 Nov 2026
  10. NVIDIA RAPIDS Accelerator for Apache Spark (now NVIDIA cuDF plugin for Apache Spark): nvidia.github.io/cudf-spark/docs/download.html, read 10 Nov 2026
  11. NVIDIA RAPIDS Accelerator for Apache Spark (now NVIDIA cuDF plugin for Apache Spark): raw.githubusercontent.com/NVIDIA/cudf-spark-benchmarks/dev/nds/README.md, read 10 Nov 2026

Last reviewed 10 Nov 2026.