What if the fusion energy that powers the sun and the stars could solve the world’s energy problems? Scientists and engineers all over the world are developing technology to replicate this process on Earth, with the aim of creating a new source of highly sustainable energy.
Fusion is routinely referred to as the ultimate power source. It’s a process that fuses together light nuclei at very high temperatures and pressures to create an incredibly hot, fluid-like state of matter called a “plasma.” If confined well enough, the process will release enormous quantities of energy, “nearly four million times more energy for every kilogram of fuel than burning coal, oil, or gas,” according to the UK Atomic Energy Authority (UKAEA).
But there’s a problem.
“The challenge we have,” says Dr. Rob Akers, Head of Advanced Computing, UKAEA, “is that there isn’t enough time for using test-based design to work out what [a fusion] power plant needs to look like. Therefore, we’ve got to design it ‘in-silico,’ that means in the virtual world, using huge amounts of supercomputing and artificial intelligence.”
Future-Proofing Code with oneAPI
UKAEA aims to deliver a “digital twin” of the world’s first fusion powerplant and is working in collaboration with the Cambridge Open Zettascale Lab1 and Intel. This involves developing the next-generation engineering tools and processes needed to design, certify, construct, and regulate a fusion powerplant, built upon advanced solutions such as Intel® oneAPI Tools and DAOS high-performance storage.
“Our simulations are going to rely upon some of the world’s largest supercomputers and huge amounts of data science,” says Dr. Rob Akers. “We’ll be very much relying upon the confluence of artificial intelligence, machine learning, and high-performance computing that is now becoming a reality.”
While many of the applications needed for the fusion project run on scalable Intel x86 hardware, future HPC workloads will be performed using a variety of different architectures. These will include 4th Gen Intel® Xeon® processors, the latest GPUs, Field Programmable Gate Arrays (FPGAs) and other new solutions that appear. So, being able to program once for all of these is key.
“Our solution to this code ‘portability’ problem is oneAPI,” explains Dr. Paul Calleja, Director of Research Computing at the University of Cambridge. “[It’s] an open and cross-architecture programming model that frees developers to use a single code base across multiple architectures. The result is accelerated compute without vendor lock-in. Intel oneAPI Tools and optimized AI frameworks already deliver leading performance on popular deep learning and molecular dynamics benchmarks—our challenge is to significantly broaden this application set to include engineering, fusion materials, and plasma simulation.”
Accelerating Storage with DAOS
The Cambridge Open Zettascale Lab is a oneAPI Center of Excellence and places a focus upon porting potential exascale code such as CASTEP, FEniCS, and AREPO to oneAPI. But seeking to future-proof code with oneAPI is only one part of the exascale computing challenge.
“When we’re talking about fusion,” Dr. Calleja says, “we’re talking about some very large simulations, deployed across thousands of GPU nodes. A machine like that, running a single plasma turbulence simulation, can produce a huge amount of output in a very short time. For example, in the US, a single simulation of the XGC turbulence code, without compression, can generate a whopping 200 Petabytes of data.2
“Dumping data for post-simulation analysis can be a rate limiting step that will slow your simulation down. So, you need a fast file system to be able to accept the simulation output. DAOS is designed to do exactly that; to accept large streaming I/O in an HDF file format from simulation to solid state storage. That’s because it’s designed from the ground up to be very efficient for solid state file systems.”
Distributed Asynchronous Object Storage (DAOS) is the foundation of the Intel exascale storage stack. DAOS is an open-source, software-defined, scale-out object store that provides high bandwidth, low latency, and high I/O operations per second (IOPS) storage containers to HPC applications. Together with Intel Xeon processors, it enables next-generation, data-centric workflows that combine simulation, data analytics, and AI.
Using these advanced solutions, UKAEA, the Cambridge Open Zettascale Lab and Intel are collaborating to accelerate and de-risk the UK’s roadmap to commercial fusion power. Intel® hardware and software, along with oneAPI and DAOS, are playing a crucial role in this endeavor, driving scientific success today, as well as shaping the HPC landscape of tomorrow.
“If we can efficiently bottle a star [using a fusion power plant],” says Dr. Rob Akers, “we will have produced a technology that we believe is going to be absolutely pivotal to creating a technologically advanced net zero world… And that is incredibly exciting.”
Read “How to bottle a star: Revolutionizing Fusion Powerplant Design with Intel® Technology” ›