Archives Discussions

[evoliptic]

Hello everybody,

First of all, i would like to thank Chris and AMD as we managed to get a transfert between FPGA and GPU and got very good numbers for the transfer speed.

However, i have a concern about the size it seems we can allocate on the GPU with clMakeBufferResidentAMD:

in my tests, i get an error "CL_OUT_OF_RESOURCES" on clMakeBuffersResidentAMD whenever i try to allocate more than 100 659 200 bytes of memory, and thus getting 0 for marker and surface adress.

Is it normal?

thanks for any answer,

regards

[chm · Accepted solution]

Hi,

I am glad to hear that you got the transfers working. The behavior you are seeing is expected. When you enable DirectGMA you specify the size of the PCIE aperture that gets reserved for DirectGMA. I guess you used

   aticonfig --set-pcs-val=MCIL,DMAOGLExtensionApertureMB,96 

   aticonfig --set-pcs-u32=KERNEL,InitialPhysicalUswcUsageSize,96

This will set the max size to 96MB and trying to use more will return an error. You can set it to a max value of 128MB. Currently this is the maximum that is supported.

   aticonfig --set-pcs-val=MCIL,DMAOGLExtensionApertureMB,128

   aticonfig --set-pcs-u32=KERNEL,InitialPhysicalUswcUsageSize,128

Chris

[evoliptic]

oh well i forgot that point. thanks Chris.

However, is there any plan from your part to increase that number?