Yes indeed. We could go deep into the technical discussion here why, but I try to make the story short. Each shader processor has some shared resources shared within the local work group size threads that you want to adjust. Using more helps the GPU to hide memory latency. That is particularly true for Nvidia Pascal Mainstream Chips with 128 cores in one shader processor. (And here we have a huge difference to Tesla P100 / V100 or AMD Vega GPUs).

In lolMiner we handle things differently by ensuring that shared resources are enough to run with multiple work groups simultaneously. Overall speed is kind of product on how well the chips scheduler can run multiple tasks on one compute core and the way the kernel itself handles memory access. Lets say it this way: I took care about this issue, thats why lolMiner is so fast (for those users without stability issues ^^)

Hopefully tomorrow this time I will have a solution for the build program error you had *crossing fingers*.



Thanks for that numbers  (By the way: which OS?)

For adding CPU mining you must install the OpenCL drivers for your CPU. For the Intel CPUs there is an Intel CPU OpenCL driver at Intel website I used. The drivers for my Kaveri AMD APU came with the installation of the AMD graphics driver, which also added OpenCL support for the CPU.

But honestly: If not really necessary I would not recommend installing the AMD graphic drivers beside the Nvidia ones. That may be more painful then worth the effort, because that Ryzen is very quick, but on CPU we are mainly bounded by memory bandwidth and memory response time. So I won't expect more then about 600 sol/s. Furthermore when your CPU is mining simultaneously to the GPU that may cost GPU hashrate since the calculation against the target, the OpenCL framework and the stratum connection to the pool run on the CPU and I fear that penalty would cost more then the extra hash rate from CPU cores