Organizations Now Able to Analyze Orders of Magnitude Larger Graphs 10-1000x Faster Than Possible with Conventional Computers
NEW YORK, Nov. 09, 2021 (GLOBE NEWSWIRE) — Lucata, provider of a next generation server platform for accelerating and scaling graph analytics, AI and machine learning (ML), today announced it has launched the Lucata Pathfinder server and a customized version of GraphBLAS for the Lucata platform. Performance benchmarks demonstrate the unmatched performance and scalability improvements of the Lucata platform, which enables users to run faster analytics on larger graphs than is possible with conventional computing technologies. The Lucata platform affordably fills the gap between the performance and scalability of conventional servers and the capabilities of supercomputers for Big Data graph analytics. A single rack of Pathfinder chassis provides the same full Breadth-First Search (BFS) performance as over 1,000 Intel® Xeon® processors, while using 1/10th the power of a comparable Xeon-based system.
This leap in price-performance enables more organizations to implement a broad range of high-value use cases at a speed and scale that was previously too expensive, too slow, or simply impossible to consider with current server technology. Financial fraud detection, cybersecurity threat detection, logistics and supply chain management, computational pharmaceutical modeling, retail recommendation engines, and more can all experience step function improvements in graph analytics performance and scale.
“Innovation is always fueled by democratizing access to the latest high-end technology. By making it possible to cost-effectively analyze massive graph databases, Lucata will spur significant innovation in multiple industries,” said Marty Deneroff, Lucata COO. “The Pathfinder benchmarks demonstrate the orders-of-magnitude increase in performance and scalability made possible today by our patented Migrating Thread technology.”
The Lucata server leverages patented Migrating Thread technology that moves compute threads to the data rather than data to the processor. This results in orders-of-magnitude less network traffic and 10x to 60x higher processor utilization rates. Because much less data is being moved and the processors are used much more efficiently than with conventional computing, Lucata delivers 10-1000x the performance of conventional servers for Big Data processing, while using 1/10th the power and occupying a fraction of the floor space, with the ability to process up to 8 PB datasets in a single, unified memory space.
Lucata’s performance benchmarks demonstrate the capabilities of the Lucata platform. The benchmarks include:
Full BFS performance of over 1 GTEPS on a scale 30 Graph500 dataset
Over 1.33 GUPS (Giga Updates Per Second) per chassis following the RandomAccess test from the HPC Challenge
The Lucata version of GraphBLAS simplifies programming for Lucata Pathfinder. It increases developer productivity and provides a fast onramp to porting existing graph applications to the platform. Further improvements can be gained by using LAGraph for even higher developer productivity. The Lucata version of GraphBLAS enables developers to use the standard GraphBLAS libraries to call Lucata algorithms. The Lucata algorithms are optimized for multi-threaded performance with the patented Lucata Migratory Threads technology, delivering massive performance and scalability gains compared to conventional servers. Developers can also call the Lucata-native algorithms directly although the Lucata customized version of GraphBLAS.
“We envision a future when we offer the capabilities of Lucata as an ASIC. The move from FPGAs to ASICs will provide a further 10x improvement in performance and a 16x improvement in memory scalability while slashing production costs,” said …….