Optimum Size Block

Mustaf

Hi,

I will use DualSPHysics with a Geforce GTX 980m (compute capabilty 5.2). In order to maximize the execution time, I had to modify the source (already done). And now, I just need to know the optimum size block for a compute capabilty equal to 5.2. For example we have (in: JSphGpu.cpp) for a compute capability 3.0 (or more):

if(compute>=30){
if(nreg<=32)return(256); // 1-32 -> 128:100% 256:100% 512:100%
else if(nreg<=40)return(256); //33-40 -> 128:75% 256:75% 384:75% 512:75%
else if(nreg<=48)return(256); //41-48 -> 128:63% 256:63%
else if(nreg<=56)return(128); //49-56 -> 128:56% 256:50% 384:56%
else if(nreg<=63)return(256); //49-63 -> 128:50% 256:50% 512:50%
else return(256);

If the benchmarks for the optimum block size (for this kind of capability) were already done (for example for the next release), I really would like to get them.

Thanks in advance for the comming help.

Mustaf