The Daily Bulletin

The first steps in optimizing code are to gain an understanding of its performance characteristics and then identify parts of the code that present opportunities for improvement. This 90-minute class—at 10 a.m. Thursday, May 28—will introduce techniques and tools for analyzing code performance, pinpointing hotspots and bottlenecks, and deciding where to focus your optimization efforts.

Topics will include timing, profiling, tracing, and code instrumentation to look at performance from a single core to multiple nodes, using tools available on Yellowstone.

You can register to attend in person—at the VisLab in NCAR’s Mesa Lab in Boulder—or via webcast by selecting a link below: 

The Caldera cluster has been expanded with the addition of 14 non-GPU nodes, giving users additional resources for analysis and visualization.

Caldera initially comprised 16 nodes, each of which has two NVIDIA GPUs. The 14 new nodes were added when CISL repurposed the Pronghorn evaluation cluster.

Except for lacking GPUs, the added nodes are the same as the other Caldera nodes, each with two 8-core, 2.6-GHz Intel Xeon E5-2670 (Sandy Bridge) processors and 62 GB of usable memory. Because they share the same node and processor architecture as Yellowstone, Caldera nodes can also be useful for compiling codes for Yellowstone or testing Yellowstone codes on a smaller scale.

To submit jobs that do not require GPUs, specify the “caldera” queue in your job script. Jobs that need exclusive use of nodes for GPGPU or visualization applications should specify the “gpgpu” queue.

See the Geyser and Caldera documentation for additional information or email for assistance in deciding which resource is most appropriate for your work.

An XSEDE Performance Tuning Summer School is scheduled for August 17-21 at CU Boulder and several other locations for those interested in hands-on training in tuning HPC codes. Lectures on performance modeling, prediction and verification are complemented by labs so participants can immediately put their newly acquired knowledge to work. 

See the XSEDE Course Calendar for details regarding the Boulder session and the course listing for other host sites.

A new job submission queue called “intviz” on the Geyser cluster has been created for highly interactive visualization tasks that require frequent keyboard or mouse input and immediate responsiveness. The queue has a four-hour wallclock limit and directs interactive job requests to shared nodes on Geyser.

Appropriate uses include working with the graphical interfaces for tools such as MATLAB and VAPOR to visualize data. The queue is monitored to ensure that it is used as intended.

Users should submit longer, less interactive visualization and analysis jobs to the “geyser” or “caldera” queues, which both have a wallclock limit of 24 hours.

See Selecting a queue for more information.