Modify ↓
Opened 13 years ago
Closed 12 years ago
#180 closed task (fixed)
Compiling and building OpenCL programs takes too much time on compute2
| Reported by: | Matthias Vogelgesang | Owned by: | Suren A. Chilingaryan |
|---|---|---|---|
| Priority: | major | Milestone: | |
| Component: | Infrastructure | Version: | |
| Keywords: | Cc: |
Description
The AMD compiler is very slow compiling OpenCL code, especially if using multiple GPUs. Creating a FFT plan on all four devices for a round-trip transform, takes about 34 seconds, the actual computation just a fraction of a second.
Caching the binaries in the framework would not be a big deal but for third-party libraries (e.g. Apple FFT) this becomes cumbersome.
Attachments (0)
Note: See
TracTickets for help on using
tickets.
![(please configure the [header_logo] section in trac.ini)](/ufo/chrome/site/your_project_logo.png)
Low quality hardware solved the low quality software problem for now. None of the cards still have cooler running. ipepdvcompute2 now runs Intel and Nvidia.