Optimize the number of work-groups

[Narendra_V_]

I have a basic question on the number of work groups that can  run in parallel. According to the definition of compute unit each compute unit can have only one work group, so number of work groups which can be run concurrently depends "only" on the number of CU present  .But in the "OpenCL* Applications - Optimization Guide" it has been specified that number of work groups depends on the number of the work items in a group .    

[Narendra_V_]

Hi Robert, 

 In Gen7.5  Is it possible to partition the LLC between the GPU separate and CPU cores separate cache?? 

[Robert_I_Intel]

Hi Narendra,

Not that I know of.

[Narendra_V_]

Hi Robert, 

   I am running an application on I7 with intel iris pro graphics 5200. I have two buffers of size 68MB and 600MB. Irrespective of the order of creation of these buffers I am getting the same performance, does it mean that EDRAM is always used as cache ??   

[Robert_I_Intel]

Hi Narendra,

The 600 MB buffer won't fit into 128 MB EDRAM, so it will always end up in System Memory (DRAM).

The 68 MB buffer will always end up in EDRAM, since it fits perfectly. So you should get the same performance irrespective of the order of the buffer creation.

[Narendra_V_]

Hi Robert, 

There is a small correction. The big buffer which is of size 600MB is actually a collection of small buffers. So there is a possibility that 68MB might not fit into the edram if 600MB is allocated first. Is there any tool to know what fraction of edram is used as victim cache??.