NVIDIA Accelerates Presto Analytics Queries on GB200 NVL72 GPU Platform

NVIDIA described in a developer-focused article how GPU acceleration of the open source SQL engine Presto on the GB200 NVL72 platform allows performing analytical queries on very large datasets with low latency.

What is Presto and why it matters

Presto is an open source distributed SQL engine, originally created for fast interactive queries against massive data sets without pre-transferring that data into a separate analytical warehouse. The tool is used when an analyst or data engineer needs to get an answer to an SQL query in seconds, not after long batch processing typical of classic big data tools.

What GPU acceleration provides

According to NVIDIA, when transferring Presto computations to GPU infrastructure GB200 NVL72, the engine demonstrates peak performance precisely on low-latency analytical workloads — that is, in scenarios where response speed matters when working with very large data volumes. This is not about replacing Presto with a new product, but about making the same open source SQL engine use GPUs more efficiently instead of typical CPU clusters. The difference between CPU and GPU execution of analytical queries is especially noticeable in scenarios with high concurrent load — when many analysts simultaneously access the same huge dataset and expect real-time response rather than waiting for a long queue of batch jobs.

It is for exactly these low-latency scenarios that NVIDIA positions the Presto + GB200 NVL72 combination.

• Presto — open source distributed SQL engine for interactive queries on big data
• Acceleration platform — NVIDIA GB200 NVL72, company's GPU infrastructure
• Claimed effect — peak performance precisely on low-latency analytical workloads
• Material published in NVIDIA Developer Blog for developers and data engineers

Why NVIDIA is promoting GPUs for analytics

The material continues NVIDIA's line of promoting GPUs not only for AI model training and inference but also for classical database analytics. The company is offering developers to migrate already existing SQL engines such as Presto to GPU platforms like GB200 NVL72 rather than rely solely on scaling up CPU clusters for interactive analytics. For teams already using GB200 NVL72 for AI workloads, such an approach means the ability to leverage the same infrastructure for analytical SQL queries without deploying a separate CPU line for data.

Such an approach is economically logical for organizations that have already invested in GPU clusters for AI model training and inference: instead of maintaining separate CPU infrastructure exclusively for SQL analytics, they get the ability to distribute analytical and AI workloads across the same pool of GPU servers.

What this means

The development of GPU-accelerated versions of popular open source tools like Presto shows that the competition for data infrastructure extends beyond neural network training: GPUs are increasingly being positioned by NVIDIA as a universal computing platform not just for AI but also for everyday business analytics over very large datasets. For the industry as a whole, this is yet another signal that the boundary between AI infrastructure and data infrastructure continues to blur, and decisions about GPU purchases are increasingly made with an eye not only to model training but also to everyday SQL analytics workloads. For NVIDIA, it is also a way to show that an open source Presto engine with properly configured GPU acceleration is not just a cheap alternative but a working tool for real analytical scenarios.

Who benefits most from this combination

Most of all, teams benefit from such acceleration where analytical queries on large data are already hitting CPU infrastructure latency limits, and the data itself is too large or volatile to pre-prepare it in a separate warehouse for each specific report.