Difference between revisions of "Teleconference 2016-05-11"
| (One intermediate revision by the same user not shown) | |||
| Line 4: | Line 4: | ||
Chris met with Pat McCormick. Is fully on board with VTK-m/Legion collaborating. Next step is running what we have at scale. There is also some potential of running some CPU and GPU running simultaneously. The underlying code should be able to handle this, but there needs to be some (fairly trivial) changes to the filter mechanism and polices to allow this. | Chris met with Pat McCormick. Is fully on board with VTK-m/Legion collaborating. Next step is running what we have at scale. There is also some potential of running some CPU and GPU running simultaneously. The underlying code should be able to handle this, but there needs to be some (fairly trivial) changes to the filter mechanism and polices to allow this. | ||
| + | |||
| + | The XVis part of the VTK-m/Legion integration is the measuring of the effectiveness. We also want to compare Legion's execution to the standard VTK pipeline model. This of course requires integration of the contouring algorithm with VTK, which is on its way. | ||
| + | |||
| + | Glut code is checked into examples of his branch. Still some issues with the rendering branch on dashboards. Some issues with Mesa. | ||
| + | |||
| + | Updated graphics cards on two dashboards. Solved memory issues on one dashboard. The problem is that we found issues with CUDA 7.5 on latest version of hardware. Coordinating with Tom Fogal on getting a workaround for VTK-m. Dashboards are stabilizing. Need to enable Mesa on one or more. | ||
Latest revision as of 12:30, 11 May 2016
Attendees: Ken Moreland (SNL), Matt Larsen (LLNL), Chris Sewell (LANL), Jeremy Meredith (ORNL), Berk Geveci (Kitware), Robert Maynard (Kitware)
Matt working on integrating VTK-m with VisIt's build system.
Chris met with Pat McCormick. Is fully on board with VTK-m/Legion collaborating. Next step is running what we have at scale. There is also some potential of running some CPU and GPU running simultaneously. The underlying code should be able to handle this, but there needs to be some (fairly trivial) changes to the filter mechanism and polices to allow this.
The XVis part of the VTK-m/Legion integration is the measuring of the effectiveness. We also want to compare Legion's execution to the standard VTK pipeline model. This of course requires integration of the contouring algorithm with VTK, which is on its way.
Glut code is checked into examples of his branch. Still some issues with the rendering branch on dashboards. Some issues with Mesa.
Updated graphics cards on two dashboards. Solved memory issues on one dashboard. The problem is that we found issues with CUDA 7.5 on latest version of hardware. Coordinating with Tom Fogal on getting a workaround for VTK-m. Dashboards are stabilizing. Need to enable Mesa on one or more.
