HEAVY.AI, a pioneer in GPU-accelerated analytics, has announced the general availability of its analytics platform integrated with NVIDIA’s advanced GH200 Grace Hopper Superchip. This new release, part of the broader HEAVY.AI 8.2 platform update, is poised to revolutionize big data processing and visualization with unparalleled speed and cost efficiency.
- Key Features of the NVIDIA GH200 Grace Hopper Superchip
- Combines an NVIDIA-designed Arm-based CPU with an ultra-fast NVIDIA Hopper GPU.
- Utilizes NVIDIA NVLink-C2C interconnect for 7X faster data transfer compared to traditional PCIe systems.
- Supports processing and visualization of massive datasets at interactive speeds.
- Performance and Cost Benefits
- 900GB/sec bidirectional bandwidth between CPU and GPU.
- Capability to query and visualize datasets exceeding GPU memory capacity.
- Example: Public demo with over 20 billion records (AIS data) achieved 70% cost savings using NVIDIA GH200 on Vultr Cloud.
- Grace Hopper and Blackwell Architectures
- Introduction of NVIDIA GB200 Grace Blackwell Superchip and GB200 NVL72.
- Liquid-cooled, rack-scale solution with a 72-GPU NVLink domain acting as a single massive GPU.
- Delivers 30X faster real-time trillion-parameter LLM inference.
- Benchmarking Against Competitors
- HeavyDB on NVIDIA GH200 outperformed CPU-based data warehouses by up to 21X in speed and 9X in cost efficiency (TPC-H SQL Data Warehouse benchmark).
- HEAVY.AI’s ongoing efforts to establish GPU-accelerated analytics as the default for big data processing.
- Customer Impact and Testimonials
- Quote from Todd Mostak, CEO of HEAVY.AI: “Customers can now achieve both faster performance and lower cost.”
- Quote from Ivan Goldwasser, NVIDIA: “The Grace Hopper architecture boosts performance and reduces costs for analyzing large datasets.”
HEAVY.AI’s collaboration with NVIDIA sets a new standard in big data analytics, offering unmatched performance and cost benefits. With the integration of NVIDIA Grace Hopper and Blackwell Superchips, organizations can seamlessly scale to larger datasets while reducing operational costs. This innovation underscores the growing dominance of GPU-accelerated analytics as the preferred solution for handling massive datasets.
