Jun 13, 2024

Databricks and NVIDIA Target New Analytics Use Cases for GPUs

Voltron Data

Databricks and NVIDIA

The data center is being re-imagined with GPUs at the center. Thousands of CPU servers are being replaced with just dozens of GPU servers. So what does this mean? Beyond increasing workload performance and saving money, accelerating analytics on GPUs has the power to drastically reduce an enterprise’s data center footprint - saving on space, energy, and cooling costs.

This is why we are excited to see Databricks and NVIDIA partner to accelerate data processing on GPUs. Together, they will work to GPU-accelerate the analytical compute layer of Databricks’ Photon product.

Voltron Data has been working on this challenge for more than a decade. GPUs have been a staple in compute infrastructure well before the GenAI boom, primarily for AI training and inference. However, one often-overlooked category of GPU workloads is data processing, particularly ETL.

Many ETL workloads are naturally parallelizable, which makes them ideal for GPUs. For example, if you want to process 1 billion rows, you can distribute them to a thousand GPU cores. GPUs handle data workloads too big for CPUs at a fraction of the time, energy, and cost. More specifically, moving ETL from CPU to GPU can speed up your workflow up to 60x while reducing costs by 50x, as shown in our Spark (on CPU) vs. Theseus (on GPU) benchmark.

Lower is better! GPUs help reduce node count nd cost while increasing speed. Learn more here: https://voltrondata.com/benchmarks/theseus

Voltron Data’s query engine, Theseus, is built for the largest data problems: petabyte-scale ETL, SQL, and Dataframes. It’s not for the masses; it’s for big data problems and enterprises that need to process the toughest and most time-consuming queries.

It’s why Voltron Data sees success with government agencies, hedge funds, and cybersecurity companies. They share two critical pain points:

  1. Analytics at scale are too slow and costly.
  2. There’s a growing need to reduce data center footprint and energy consumption.

Theseus is built from the ground up to take full advantage of accelerated data system performance: from GPU compute and high memory bandwidth, to advanced networking and storage. It’s the Ferrari of query engines – really fast, elegantly designed, and ultra-power efficient for those who need it.

The work to be done with Photon will expose more customers to what is possible with Theseus and analytics on GPUs at scale. Voltron Data is the leading designer and builder of accelerated data systems. Together, let’s accelerate data processing and work to get the most out of your GPU investment. Reach out.