High-performance computing infrastructure is required due to the huge sizes of whole-slide histopathology images and the sophisticated image analysis routines. We are working to develop efficient and effective computational processing methods for image analysis.
GPU stands for Graphics Processing Unit. Initially designed for gaming applications to implement complex rendering operations for real-time effects.
Now, with their improved performance they are being used for a variety of purposes.
We are using GPUs to process high-resolution histopathology images. We implemented the image analysis routines on GPU to classify stromal development for NB image analysis. GPU provided remarkable computational savings with 45 times faster processing. Processing of a relatively small whole-slide image with a resolution of 50Kx50K reduced to 47 sec. on GPU compared to 35 min. on CPU.
Cell processor has been designed at IBM in conjunction with Toshiba and Sony.
The Cell is a 9 core chip with a shared L2 cache.
1 PowerPC (PPC) core.
8 Synergistic Processing Elements (SPE).
High performance I/O to support data-intensive applications.
Instruction-level parallelism (vector operations in pipeline stages)
Thread-level parallelism (via multiple tiles computing at once).
Dynamic scheduling of operations on SPEs.
Tiles are read from storage node(s) and pushed to PS3(s).
6 times faster processing of image tiles compared to CPUs.