With the help of compression algorithms running on AI hardware, scientists can now speed up simulations of complex systems.
KAUST computer scientists have developed an approach to reduce data movement that saves on memory footprint, data transfer, and algorithmic complexity. This image was generated with the assistance of AI. Image Credit: King Abdullah University of Science & Technology.
High-performance computing (HPC) has become a vital tool for processing huge datasets and simulating some of the most complicated systems in nature.
However, researchers face difficulties in developing more intensive models because Moore’s Law – which states that computational power doubles every two years – is slowing, and memory bandwidth cannot keep up.
A team led by computer scientist Hatem Ltaief is attempting to find a solution to this problem by employing hardware designed for artificial intelligence (AI) to help scientists make their code more efficient.
According to the team. they have successfully made simulations up to 150 times faster in the diverse fields of climate modeling, astronomy, seismic imaging, and wireless communications.1
Earlier, Ltaief and co-workers showed that several researchers were riding the wave of hardware development and “over-solving” their models, thereby performing plenty of unnecessary calculations.
With the increasing energy cost of data movement and hardware limitations in terms of energy efficiency, we need algorithmic innovations to rescue the scientific community, which is in panic mode. Reducing data movement becomes like reducing fuel consumption for airlines —a must. What if we could solve a huge memory footprint problem by only operating on the most significant information, and yet still achieve the required accuracy?
Hatem Ltaief, Computer Scientist, King Abdullah University of Science & Technology
Around five years ago, Ltaief and his co-workers began to research ways to decrease data movement. Their method involves reorganizing the HPC workloads so that they can run on AI-focused Intelligence Processing Units (IPUs) created by Graphcore, a company offering priceless technical support.
Importantly, the team organizes the code into matrices – single mathematical objects that work efficiently with numerical libraries that are optimized for the IPUs.
We can perform compression on the matrix operator that describes the physics of the problem, while maintaining a satisfactory accuracy level as if no compression was done. We can still manipulate the resulting compressed data structures by executing linear algebra matrix operations and leverage the high bandwidth of the IPUs.
Hatem Ltaief, Computer Scientist, King Abdullah University of Science & Technology
This method has been proven to save on data transfer, memory footprint, and algorithmic complexity. Already, it is increasing the speed at which researchers can handle issues like adapting astronomical telescopes to post-processing field data in seismic imaging or real-time changes in the atmosphere.
I am lucky to work with scientists that embrace hardware technologies and understand how impactful multidisciplinary work can be.
Hatem Ltaief, Computer Scientist, King Abdullah University of Science & Technology
“Linear algebra operations are the bottleneck of many applications. You can expect to see more headlines about compression technology as it permeates new applications and hardware,” states Ltaief’s colleague David Keyes.
Journal Reference:
Ltaief, H., et al. (2023) Steering Customized AI Architectures for HPC Scientific Applications. International Supercomputing Conference, Hamburg, Germany. doi.org/10.1007/978-3-031-32041-5_7.