Customers of the platform see significant performance and cost benefits using NVIDIA accelerated computing
HEAVY.AI, the trailblazer in GPU-accelerated analytics, has officially launched the HEAVY.AI analytics platform, now optimized for the cutting-edge NVIDIA GH200 Grace Hopper Superchip. This release, part of HEAVY.AI’s larger 8.2 platform update, integrates an advanced NVIDIA-designed Arm-based CPU with the blazing-fast NVIDIA Hopper GPU, delivering breakthrough performance for data analytics.
With the addition of Grace Hopper Superchip support, HEAVY.AI users will experience unparalleled processing speeds and reduced operational costs. The key advantage lies in the ultra-fast NVIDIA NVLink-C2C interconnect that links the CPU and GPU, offering 900GB/sec of bidirectional bandwidth. This results in data transfer speeds up to seven times faster than traditional PCIe systems. As a result, HEAVY.AI users can now effortlessly query and visualize massive datasets that far exceed the GPU’s memory capacity—at interactive speeds.
In addition to the Grace Hopper architecture, users will also gain access to the NVIDIA GB200 Grace Blackwell Superchip and the GB200 NVL72 liquid-cooled, rack-scale solution. This solution includes a 72-GPU NVLink domain that operates as a single massive GPU, enabling real-time trillion-parameter LLM inference with 30X faster performance.
The HEAVY.AI platform’s integration with Grace Hopper and Blackwell architectures also delivers significant cost savings. With this release, users can scale to larger datasets using fewer GPU resources. To demonstrate these savings, HEAVY.AI has introduced a public demo featuring over 20 billion records of ship locations (AIS data) along the US coast over the last seven years. This demo runs seamlessly on a single NVIDIA GH200 Superchip on Vultr Cloud—offering nearly 70% hardware cost savings compared to previous configurations that would have required at least four NVIDIA A100 GPUs with 320GB of combined GPU VRAM.
Also Read: Marvell Unveils Innovative Co-Packaged Optics Architecture for Tailored AI Accelerators
Todd Mostak, Co-Founder and CEO of HEAVY.AI, commented, “The NVIDIA Grace Hopper Superchip changes the game in terms of being able to provide best-in-class performance over our customers’ largest datasets. Now, customers no longer have to choose between faster performance and lower cost; with HEAVY.AI running on NVIDIA Grace systems, they can have both.”
Ivan Goldwasser, Director of Data Center CPUs at NVIDIA, added, “Customers worldwide are looking to boost performance and reduce cost when analyzing large datasets. By accelerating HEAVY.AI’s analytics platform with NVIDIA Grace Hopper and Grace Blackwell Superchips, customers can speed up high-performance data processing and visualization for big data analytics.”
In addition to cost savings, HEAVY.AI recently unveiled benchmark results showcasing the performance advantages of its HeavyDB GPU-accelerated database. In a comparison with three leading CPU-based data warehouses using the TPC-H SQL Data Warehouse benchmark, HeavyDB on the NVIDIA GH200 Grace Hopper system delivered up to 21X faster performance on average, while being up to 9X more affordable to operate per hour. Full details of the benchmark can be found [here].
The launch of HEAVY.AI‘s analytics platform on NVIDIA Grace architecture is part of a broader initiative to further cement GPU-accelerated analytics as the go-to solution for analyzing large datasets. Todd Mostak emphasized, “Our overarching goal is to make the performance and cost advantages of GPU-accelerated analytics so compelling that it becomes the default means of analyzing large datasets.”
