Biomedical Imaging to Revolutionize Patient Outcomes
Taking the leap of faith to run code optimized for graphics processing unit (GPU) architectures can provide significant performance benefits for imaging applications, but it can also introduce risks including reduced overall system reliability, upgrade suitability and tedious system manageability.
Performance at Scale
Migrating to GPU architectures can improve performance by 10 to 20 times for medical imaging applications, but there are potentially many bottlenecks in getting there. Pioneers running medical imaging applications on GPU clusters often face daunting challenges, including:
- System throttling – Running GPUs at high utilization for extended periods will often create thermal issue, causing systems to throttle and slow down.
- System instability – Thermal and other performance issues will often impact reliability, causing the system to stutter and ultimately fail, a troubling scenario in a healthcare environment.
Imaging for Digital Pathology Workflows
Cray's CS-Storm cluster supercomputer is well-suited to accelerating digital pathology workflows at scale. Leveraging 40 years of HPC expertise, Cray built the CS-Storm system with balanced I/O, memory and CPU/GPU capabilities to handle today's and tomorrow's most demanding digital pathology workflows. According to NVIDIA, the CS-Storm system is the only cluster capable of running both CPUs and GPUs at maximum sustained utilization, which means throttling or temperature issues won't interrupt your workflow. You'll get the reliable performance you demand in a healthcare environment, instead of the unpredictable performance other GPU clusters provide.
Accelerating Cryo-EM Workflows
Cryo-electron (cryo-EM) microscopy is rapidly becoming a key technology for 3D molecular structure determination. Cray's CS-Storm cluster is ideally suited for cryo-EM workflows in single-particle imaging, where iterative mathematical 3D reconstruction processes are applied to large numbers of 2D particle images to produce quality 3D models.
Improving Pathology Productivity with Machine Learning
As the volume of digital images explodes from myriad sources, it becomes challenging for healthcare professionals to sort through the uninteresting images or sections of images to find that needle in a haystack. Machine learning may be beneficial in this area, not by replacing the function of the healthcare professional, but by improving the signal-to-noise ratio.