Developers see the addition of CUDA 7 support in version 5.0.1 of their tools - continuing Allinea Software's track record of rapidly supporting the latest CUDA releases. Allinea Forge, which includes the Allinea DDT debugger and Allinea MAP profiler, is used widely on clusters, supercomputers and workstations for developing high performance hybrid CPU/GPU code.
Support for debugging C++11 CUDA and GPU core files that are now created by the GPU in CUDA 7 is added into the debugger. Meanwhile, GPU and global memory utilization metrics are added to the profiler, which enables the performance of hybrid applications to be optimized - even where applications are running across large clusters.
System owners and administrators are also set to benefit from the announcement that GPU metrics have been added to Allinea Performance Reports. The tool transparently analyzes applications running on a system to measure and analyze achieved efficiency - enabling rapid triage and targeting of applications and systems that are missing out on performance.
"Moving from single-node to multiple nodes is a significant challenge for all developers, especially when making use of multiple GPUs. To achieve real speedup you need to see where and why code spends time in MPI, I/O, GPU and OpenMP code, to measure GPU utilization without affecting performance", stated Mark O'Connor, VP Product Management for Allinea Software. "Our users can see how each line of source code spends its time, identify bottlenecks and dive right down into individual kernels with the debugger to understand and fix errors on the path to production-ready CUDA and OpenACC code."