Bodo Blog

Building a Native GPU Iceberg Writer for Apache Iceberg

Here is a look at the internals of how we built our native GPU Iceberg writer. By building Iceberg write support directly into Bodo’s native execution engine, the result is not simply a faster Parquet write path—but a GPU-native Iceberg sink that preserves the throughput of the broader distributed pipeline. 

Bodo vs Dask-CuDF: TPC-H on Distributed GPU clusters

Benchmarking all 22 TPC-H queries on a multi-node GPU cluster and show an overall speedup of more than 3× over Dask-CuDF across the full benchmark suite. These results highlight the advantages of an MPI-based SPMD execution model for communication-intensive analytical workloads.

Scaling GPU DataFrames: Why Bodo’s SPMD Architecture Outperforms Task-Based Engines

By combining a database-grade optimizer, streaming execution, and an efficient HPC-style distributed runtime, Bodo optimizes and executes entire workloads across nodes and devices, maximizing performance while minimizing overhead.

Building PyDough: Moving Correctness Upstream with a Knowledge Graph

In PyDough, accuracy does not depend primarily on prompt tuning or semantic-layer completeness. For relational reasoning to be reliable—especially in the context of large language models—it must sit on top of an explicit structural model of meaning. In PyDough, that model is a knowledge graph.

Pandas 3’s Native Integration with Bodo JIT and Iceberg

Pandas 3 takes a major step forward in performance and scalability by introducing native Bodo JIT integration for accelerating UDFs, and native Apache Iceberg support for reliable, scalable data management. We walk through both integrations and show how to get started.

Bodo DataFrames vs PySpark and Dask on TPC-H Benchmarks

We compare Bodo DataFrames, Dask, and PySpark on a cluster of four Amazon EC2 instances (128 physical cores).

Building PyDough: Why We Built a New Analytics Language

As analytics becomes more conversational, iterative, and collaborative, that gap matters more than ever. PyDough is our attempt to close it—not by generating better SQL, but by giving people a language that’s closer to how they actually think about their data. In this post, we walk through the specific limitations of SQL that led us to building PyDough, and show how PyDough approaches analytics from a different starting point.

2025 Wrapped: Bodo’s Year in Review

Bodo Native Integration in PyIceberg 0.10: Bringing Scalability to PyIceberg with Pandas APIs

The PyIceberg 0.10 release introduces a new first-class integration: Table.to_bodo(). With this method, PyIceberg now plugs directly into Bodo’s high performance distributed DataFrame engine, all without leaving the Pandas API.

From Iceberg to LoRa: A Unified LLM Fine-Tuning Pipeline with Bodo

We show how to use Bodo DataFrames to load and preprocess data directly from an Apache Iceberg table and feed it seamlessly into Bodo AI Toolkit to fine-tune a Llama 3.1 8B model using LoRa. We'll be training a chatbot on its own "liked" feedback to improve its responses.

PyTorch Conference 2025: Recap and Key Takeaways

Last week, some of us had the chance to attend PyTorch Conference 2025. We wanted to share some of our key takeaways from the conference; from distributed training to efficient inference, as well as highlight one aspect that is absolutely crucial but received less attention: data infrastructure.

Introducing PyDough-CE: A simpler, safer path for natural language analytics

Today we’re releasing PyDough Community Edition (PyDough-CE), an open source project that finally makes natural language analytics both practical and safe.