berita perusahaan tentang Starburst supports GPUs for faster distributed data analysis

Starburst achieves 3 to 4x faster AI analysis query responses by adopting GPUs instead of CPUs.

The firm develops and uses open-source distributed SQL engine Trino to query and analyze disparate data sources, previously relying on x86 CPUs for query processing. Its software can run data queries across spreadsheets, databases, cloud warehouses and lakehouses. In May, Dell updated its AI Factory product portfolio. One key update: the platform includes a Starburst-powered Data Analytics Engine with GPU-accelerated SQL analytics, delivering up to 6x faster query performance over non-GPU systems on Nvidia Blackwell GPUs, with support planned for future Vera hardware.

Starburst CEO Justin Borgman shared in a briefing that Starburst software now supports Nvidia’s Vera CPU and Rubin GPU. When asked about Starburst’s market stance relative to Databricks and Snowflake, Borgman commented: “It is classic co-opetition. We cooperate from a customer perspective and interoperate with those platforms, querying their data. If you are a large bank running Databricks, Snowflake, Teradata, IBM and Oracle, Starburst complements all these systems. This is core to our strategy, giving customers flexibility to preserve their existing tech investments. On the competitive side, both platforms aim to serve as the single unified data source.”

“We compete for certain workloads, yet our architecture differs. For large enterprises, we deliver unique value they cannot match: data federation and hybrid deployment. We support on-premises operation, while those two platforms run only on cloud.”

He continued: “The Nvidia partnership becomes compelling because we now support GPUs alongside CPUs. Nvidia released a new CPU this year, though GPUs remain its core business. Enterprises can leverage existing GPUs, including idle capacity, to run analytics workloads and gain greater flexibility.”

“We started this work at the beginning of the year and have just released our first version, built using cuDF, an open-source NVIDIA CUDA-X structured data processing toolkit. Built on optimized CUDA primitives, cuDF leverages GPU parallelism and memory bandwidth to accelerate data processing and analytics pipelines.”

Borgman stated: “We see strong performance, roughly three to four times faster than CPUs. GPUs carry higher costs, which must be weighed in cost-performance planning. But early use cases show large banks have purchased GPUs that do not run at full utilization constantly. They already incurred this capital expenditure; running batch analytics during off-hours can lift GPU utilization from 60% up to 90%.”

He outlined two operational approaches: “You can schedule jobs for known idle windows such as 9 PM to midnight. We also support real-time GPU utilization monitoring to dynamically schedule jobs when spare resources become available.”

Nvidia benefits as this expands its total addressable market.

When asked what sparked this initiative, Borgman said: “Serendipity played a big role. Early this year, Nvidia contacted us to benchmark its new Vera CPU, marking its entry into the CPU market. Our tests showed Starburst ran significantly faster on Vera compared with Intel processors. Nvidia featured our benchmark results in Jensen’s keynote presentation, which kickstarted our collaboration. Then Nvidia suggested we evaluate their cuDF framework to enable Starburst on GPUs.”

The resulting performance gain ranges from 3x to 7x depending on query complexity, with the largest, most complex queries hitting the 7x speedup.

Borgman believes this changes Starburst’s positioning against Databricks and Snowflake: “This sets us apart. Databricks and Snowflake are cloud-native services that absorb compute costs within their pricing model, squeezing their profit margins. We let customers deploy on their own infrastructure. These cloud vendors must continuously seek the lowest-cost compute, which does not align well with Nvidia’s higher-priced hardware. We believe we are the only enterprise data platform well-suited to partner with Nvidia in this space.”

Beijing Qianxing Jietong Technology Co., Ltd.
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!