The advent of Graphics Processing Unit (GPU)-based acceleration for handling data-intensive computing tasks is creating a paradigm shift in daily workflows and the approaches scientists and engineers take to solve complex problems. Ryan presents two application examples in the medical device industry that demonstrate how a new type of supercomputer can improve design processes and accelerate time to market.
In the past, using a supercomputer was the only way to run complex simulations or data processing models. As no alternative was available, those who needed the ultimate in performance to compete in the marketplace shelled out huge sums of money to procure the ultimate in computing devices.
For many organizations, the cost of obtaining this type of computer is a prohibitive barrier to better product design, faster test results, higher-resolution data analysis, and the essential information needed to make appropriate business decisions. Fortunately, organizations now have other choices, including multicore processors, FPGAs, cell processors, and GPUs that deliver compelling performance gains to users across several applications. These alternatives provide increased processing capabilities while offering a more efficient, flexible, accelerated system for the most intensive data computations.
GPUs have been particularly successful as a result of their ability to run large simulations using parallel processing elements and their added advantage of high memory bandwidth. The latest GPUs have been redesigned to tackle computational tasks, offering multifold performance increases depending on the type of application.
Combined with acceleration software, the result is a peripheral device that functions as a desk-side supercomputer capable of supercharging computing applications by more than 35x, turning lengthy projects into real-time processes. As an additional benefit, the device reduces power consumption, ultimately lowering total cost of ownership. A simple desk-side computer workstation can effectively replace traditional clusters of CPUs, allowing complex calculations to be run at one's desk, reducing wait time, and increasing efficiency.
Faster medical imaging
As a new way of approaching data-intensive computations, the potential of GPU acceleration software is expanding rapidly. One of the first companies to experience this transition is Boston Scientific, a medical device company based in Natick, Massachusetts.
Researchers at Boston Scientific are investigating how design parameters of pacemakers and other biomedical implants are affected by exposure to electromagnetic fields from MRIs and other diagnostic imaging tools. The design simulations that are required for this task are compute-intensive and take considerable time to run on standard computer clusters.
To boost performance, Boston Scientific implemented a proprietary simulation system from Acceleware that combines SPEAG's SEMCAD X software and an NVIDIA Tesla GPU. Using this system, engineers at Boston Scientific increased simulation runtime by up to 25x compared to the company's previous CPU-based system.
Additional applications benefiting from this simulation system include cell phone design, seismic data processing, PCB design, photonic/communications devices, drug discovery, oil reservoir simulation, lithography mask design, and biomedical image reconstructions.
In an example of the latter application, scientists at Robarts Imaging Research Laboratories in Ontario, Canada, have utilized AxRecon (Figure 1), a GPU-based desktop supercomputer, to accelerate CT reconstructions. Acceleware and Robarts collaborated to assess the supercomputer on the imaging institute's widely used micro-CT scanning system. The evaluation included refining certain 3D reconstruction algorithms, developing new methods for processing raw data, and finding ways to increase image reconstructions with the NVIDIA GPU's added processing capabilities.
Before implementing the supercomputer, researchers experienced slow image reconstruction times that ranged from 15 minutes up to several hours, depending on volume sizes. This significantly reduced the time researchers could devote to the workflow and hindered their processing.
Using the high-performance computer, testing at Robarts demonstrated a dramatically improved workflow in a preclinical setting. The reduced reconstruction times made the expensive scanners more accessible to researchers by allowing each job to be finished faster than what could be achieved using the institute's previous computing platform. The accelerated system has enabled researchers at the lab to attain 50x speedup, generating a more efficient workflow and allowing more image batches to be completed.
Speeding application development
The power of supercomputers is rapidly extending to new applications and markets. GPU-based acceleration supercomputers offer researchers, engineers, and clinicians the opportunity to improve their design processes and simulations as well as get products to market faster.
Developments in market verticals such as electronics, image reconstruction, and seismic data processing are benefiting from using GPU acceleration to handle data-intensive computations. Users are accomplishing more with their desk-side supercomputers than they ever thought possible, saving time and money while transforming the future of applications.