Intel recently launched its advanced hardware accelerator, Heracles. The device, which is the most powerful device on Earth purposefully engineered for FHE computing tasks, was announced at the IEEE International Solid-State Circuits Conference (ISSCC) in San Francisco. This new breakthrough technology represents a huge advancement in the speed and efficiency of processing encrypted data. It accomplishes this at speeds that are an amazing 5,000 times faster than a top Intel server CPU. This announcement represents a huge step forward in the secure computation field. Such developments are proving to be ever more vital in our data-driven, pandemic-stricken society.
Heracles is billed as having 64 compute cores, laid out in an eight-by-eight grid, or tile-pairs. The accelerated hardware includes best-in-class, 3-nanometer FinFET technology with high-bandwidth memory powering its AI processes. With its potential to increase the speed and quality of important FHE calculations, Heracles has the promise to improve many industries where increased data security is needed.
Impressive Performance Metrics
Heracles’ performance metrics are striking. Image courtesy of IBM It operates at a frequency of 1.2 gigahertz and requires just 39 microseconds to perform the required mathematical transformation at the core of FHE. It is because an Intel Xeon CPU clocked at 3.5 GHz would be significantly slower to get work done. It barely manages a 2,355-fold speedup in processing time.
In a real-life test, Heracles accomplished the astounding task of confirming 100 million voter ballots in just 23 minutes. A standard CPU would take a jaw-dropping 17 days to do that work. This major performance increase demonstrates Heracles’ potential to transform operations for several industries. This is especially disastrous for those sectors of the economy which are most dependent on high-volume data validation and processing.
“We have proven and delivered everything that we promised.” – Ro Cammarota
Advanced Architectural Design
Heracles‘ architecture is built from the ground up to maximize data movement and operation efficiency. The system’s on-chip 2D mesh network is a highlight. This architecture connects the tile-pairs together with broad, 512-byte buses, facilitating incredibly fast communication between chip cores. Further, the liquid-cooled package contains two 24-gigabyte high-bandwidth memory chips that accelerate the processor’s performance.
Heracles further uses three parallel pipelines of commands to ensure top efficiency in operations. One stream takes care of transferring data to and from the processor. A second stream handles internal data flow where the third stream takes care of the arithmetic operations required to perform FHE operations. This advanced architecture provides the capability for fluid and efficient data movement and compute, solving one of the core problems in high-performance computing.
“It’s all about balancing the movement of data with the crunching of numbers.” – Sanu Mathew
Implications for Future Computing
Heracles’ launch is a monumental step in creating the first practical, easy-to-use FHE computing. Organizations are looking for safer ways to handle sensitive data. Technologies such as Heracles are poised to help achieve this ambitious goal in a big way. Applications go much further than basic data encryption, as they cover highly advanced machine learning operations, such as powering neural networks and semantic search.
Kurt Rohloff emphasized the importance of hardware advancements for emerging applications: “Where you start to need hardware is emerging applications around deeper machine-learning oriented operations like neural net, LLMs, or semantic search.” With capabilities as extensive as those of Heracles, it’s sure to be a force propelling further innovation across these fields.
“When Intel starts talking about scale, that usually carries quite a bit of weight.” – Kurt Rohloff