Does the ATI's OpenCL implementation a watchdog in the drivers?
What happens in the future OpenCL- GPU accelerated implementation if a kernel takes more than 5 seconds to run?
thx
This really is an operating system issue as Windows enforces a watchdog timer and there is nothing that can be done to avoid it. This is not the case for Linux and the ATI implementation do not enforce a watchdog timer in this case.
Originally posted by: omkaranathan This really is an operating system issue as Windows enforces a watchdog timer and there is nothing that can be done to avoid it.
What about dedicating some fixed Mhz to DX? For example, if the GPU runs at 900Mhz then use 800Mhz for OpenCL and reserve 100Mhz for painting the Windows's UI.
could you please explain what you are expecting to do?
are you expecting to do both computing work and graphs work simultaneously?
Originally posted by: bubu
What about dedicating some fixed Mhz to DX? For example, if the GPU runs at 900Mhz then use 800Mhz for OpenCL and reserve 100Mhz for painting the Windows's UI.
It's not possible to divide up a GPU like that currently. The best option is for the user to break up their kernels to stay under the limit for production code. For development purposes, there is an article on MSDN on how to disable/adjust the watchdog timer.