Quantum computer systems and classical HPC are historically “disparate methods [that] function in isolation,” IBM researchers clarify in a new paper. This may be “cumbersome,” as a result of customers should manually orchestrate workflows, coordinate scheduling, and switch information between methods, thus hindering productiveness and “severely” limiting algorithmic exploration.
However a hybrid method can simplify the method of making use of quantum computing to issues in areas like chemistry, supplies science, and optimization, and “resolve issues that have been beforehand out of attain,” IBM says.
The researchers describe quantum-centric supercomputing (QCSC) as evolving via three distinct phases: quantum methods as specialised compute offload engines inside HPC environments; quantum and basic HPC methods coupled via superior middleware; and absolutely co-designed HPC and quantum methods for hybrid workflows.
The primary part focuses on establishing “foundational integration throughout a number of dimensions,” the researchers clarify. The second part focuses on lowering latency, creating a number of subtle suggestions mechanisms, and supporting complicated hybrid algorithms. The third part represents the “fruits” of the mixing via “absolutely co-designed heterogeneous methods the place quantum and classical sources are architected as unified platforms from the bottom up.”
The latter mirrors the trajectory of GPUs in HPC methods, the researchers notice; early GPUs sometimes functioned as exterior accelerators connected to host processors. However then interconnects have been established between GPUs and CPUs, and from GPU to GPU, to offer a lot greater bandwidth and decrease latency.
“Equally, quantum methods will transition from standalone models to totally built-in elements inside co-designed quantum-HPC platforms,” the researchers contend.
