Programming High-Performance Heterogeneous Computing Systems with the Radeon Open Compute Platform
Session ChairWu Feng
Event Type
Emerging Technologies
TimeWednesday, November 16th10am - 6pm
Location155-B
DescriptionThe exigent demands of emerging applications to maximize performance in stringent power and thermal constraints have made heterogeneous computing ubiquitous. However, the current solutions that enable heterogeneous computing manifest several inefficiencies like (i) the lack of proper RDMA support (ii) dependence on a single programming language and (iii) an inadequate software stack. We have developed the Radeon Open Compute Platform (ROCm) which improves upon the current state of the art for GPU computing.
Some of the salient features of ROCm include enhanced multi-GPU computing with peer-sync RDMA support, a rich runtime with features like the platform atomics and user-mode queues. ROCm also provides support for several programming languages like C++, OpenCL, and Python via the Heterogeneous Compute Compiler (HCC). The basis of HCC lies in C++, the de-facto for performant programming. HCC eases the development process and provides full control of the platform to the programmer including direct-to-ISA compilation.
Some of the salient features of ROCm include enhanced multi-GPU computing with peer-sync RDMA support, a rich runtime with features like the platform atomics and user-mode queues. ROCm also provides support for several programming languages like C++, OpenCL, and Python via the Heterogeneous Compute Compiler (HCC). The basis of HCC lies in C++, the de-facto for performant programming. HCC eases the development process and provides full control of the platform to the programmer including direct-to-ISA compilation.
