in AMD APP SDK v2.4 release notes there is written that ara improved the PCIe transfer speed and kernel launch times.
So I installed the v2.4 and i see that the launch time is improved from 2502 microsecond to 415 microsecond but at the same time I see that the kernel execution time is worsen from 597 microsecond to 2753 microsecond. How can you see the addition of the two time (launch and execution time) is similar even if there is little worsening. The same thing happen for memory write and read time.
So I don't see an improvement of kernel launch time but a rename of the time.
however this thing happen only with ati catalyst 11.4 driver or the following, with driver 11.3 the launch time anc execution time is equal to SDK v2.3.
Are you using the AMD APP Profiler or your own counters to measure the time.
Can you check your application with Cat 11.6 and report.
Please post a testcase to reproduce the issue if possible.Also mention your system information:CPU,GPU,SDK,Driver,OS
I saw same problem when comparing nvidia sdk particles (256 k particles) when using 11.3 (good results) and 11.4 and newer (really bad results)
System: radeon 5870, Phenon II (4 cores), Windows Vista 64.
Do you have any answer about?