SCIENCE

Rogue Wave Software announced the release of TotalView 8.9.2 with ReplayEngine 2.1 and MemoryScape 3.2.2.

These debugging and memory analysis products simplify development for data-intensive applications that are multi-threaded or distributed across a network or cluster. With support for NVIDIACUDA 4.0 and SDK 4.0, TotalView provides debugging at the cutting edge of GPU development.

TotalView 8.9.2 supports developers who are taking advantage of CUDA 4.0 features, such as sharing GPUs across multiple threads, and faster multi-GPU programming with unified virtual addressing. Debugging with TotalView provides intuitive control of GPU device kernel threads and a straightforward graphic display of CUDA exceptions, with a clear representation of GPU device memory types for all CUDA variables.

TotalView has the unique ability to simultaneously debug parallel applications using more than one GPU device per node across an entire cluster. It is designed for developer productivity, simplifying and shortening the process of developing, debugging, and optimizing complex code. TotalView provides a distinctive combination of capabilities for pinpointing and fixing hard-to-reproduce bugs, memory leaks, and performance issues related to parallel development.

"As GPU computing becomes increasingly pervasive in the HPC arena, the availability of TotalView 8.9.2 with CUDA 4.0 GPU debug capability is particularly useful for developing applications that scale to thousands of processes," said Sanford Russell, director of CUDA marketing at NVIDIA. "TotalView is a great complement to the CUDA development platform, particularly in the context of MPI applications running on NVIDIA accelerated clusters."

"We're seeing increased adoption of GPU acceleration across our customer base, so we're proud to release CUDA 4.0 support for TotalView. By supporting the latest release of CUDA, TotalView continues to set the standard in HPC debugging," says Chris Gottbrath, Principal Product Manager at Rogue Wave Software. "We've been hearing from customers working on CUDA code that TotalView allows them to work twice as fast as they could with other techniques."