4060 ti vs 1650

[David67821]

I'm seeing something strange...
on the 1650 video card it finds 10 matches per template per day...
on the 4060 ti it finds 18-20 matches per template per day...

but the 4060 ti is 5 times faster! and shows 5 times more Mkey/s

I don't understand why it doesn't find 5 times more matches?

[joaoescribano]

As far i can tell, the code is bad optimized for newer GPUs (i'm usign a RTX 3080ti) and at the standard code compilation, i'm getting 1.4 GKey/s, after few updates at the cuda engine, i'm getting now 3 GKey/s.

in GPUEngie.cu#456 at function bool GPUEngine::callKernel() It uses nbThread / nbThreadPerGroup, nbThreadPerGroup as cuda parameters, i've been playing with it to find the best tune.

I've also changed the 8.6 Cores per SM at CPUEngie.cu#131 (_ConvertSMVer2Cores) to 1024, as my GPU can handle it.

Dunno if it's the case or not, but it helped me.