You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A view of a device could select a subset of CPU cores (and memory?) that are used for execution.
The current Device concept always models the DeviceView concept by always using all the cores.
With this it would be possible to partition a 8 core CPU into e.g. 6 cores for one DeviceView and 2 cores for another DeviceView. Kernels executed in the corresponding streams then will only use the number of cores they are allowed to. This would allow to prevent independent streams from mutually influencing each other. Furthermore this would allow to guarantee selected compute resources for time critical tasks.
The text was updated successfully, but these errors were encountered:
A view of a device could select a subset of CPU cores (and memory?) that are used for execution.
The current Device concept always models the DeviceView concept by always using all the cores.
With this it would be possible to partition a 8 core CPU into e.g. 6 cores for one DeviceView and 2 cores for another DeviceView. Kernels executed in the corresponding streams then will only use the number of cores they are allowed to. This would allow to prevent independent streams from mutually influencing each other. Furthermore this would allow to guarantee selected compute resources for time critical tasks.
The text was updated successfully, but these errors were encountered: