Quote from: Andre_25 on Today at 12:51:45 PM

Quote from: WanderingPhilospher on September 16, 2021, 09:57:28 PM

Quote

Why does appear (38X0 cores) ?
help me please understand a little.

The code was built before the new cards came out. If you want to adjust, go into GPUEngine.cu and change the code to this:

Code:

sSMtoCores nGpuArchCoresPerSM[] = {
{0x20, 32}, // Fermi Generation (SM 2.0) GF100 class
{0x21, 48}, // Fermi Generation (SM 2.1) GF10x class
{0x30, 192},
{0x32, 192},
{0x35, 192},
{0x37, 192},
{0x50, 128},
{0x52, 128},
{0x53, 128},
{0x60, 64},
{0x61, 128},
{0x62, 128},
{0x70, 64},
{0x72, 64},
{0x75, 64},
{0x86, 128},
{-1, -1} };

The {0x86, 128}, part is what will update and show you the proper grid size for your newer graphics card.

Ok, thank you very much for answering.
With that, as I understand it, I will allow to show the correct number of cores, but then the current results, with the 3060 ti are less than the 2080, for example? or do I still need to correct something in it to get better results?

Thanks again!

You can play around with the grid size to see if you can tweak out more MKey/s, but here's the deal, IMO, most of these programs were releases prior to 30xx cards so the programs probably do not best optimize the use of the cards GPU structure. I have only used one program that did and that was an OpenCL program versus Cuda.