There is not much one can do with an Intel GPU, but I've found the OpenCL sieving (gfn opencl sieving app) will work with one laptop's Intel HD Graphics 4000 GPU. Its pretty poor performance, but it does work. Better than nothing. With this i3 CPU I am able to get it up to 2.4p/day when running two simultaneous tasks. Doing about 1.2p/day each using B9. Any higher and it refuses to run. No B10 or greater. Hardly anything compared with a 1060 at 105p/day but it is still something.

There is not much one can do with an Intel GPU, but I've found the OpenCL sieving (gfn opencl sieving app) will work with one laptop's Intel HD Graphics 4000 GPU. Its pretty poor performance, but it does work. Better than nothing. With this i3 CPU I am able to get it up to 2.4p/day when running two simultaneous tasks. Doing about 1.2p/day each using B9. Any higher and it refuses to run. No B10 or greater. Hardly anything compared with a 1060 at 105p/day but it is still something.

Yes, sieving does work on iGPUs...

...but it's just not worth it. The thing is that it steals so much memory bandwidth that the CPU tasks get starved. From my testing, even SGS (which is pretty tamed with RAM, in comparison to the other LLR projects) loses performance: 1 unit was fine, 2 had a small loss, 3 and 4 just got destroyed. And keep in mind, that was with dual channel, dual rank, 3000mhz DDR4, so it's not as if I had slow RAM there to limit me.

If there was a sieving challenge, I guess you could use the iGPU to squeeze as much as you can. But outside of that, it's not a good idea to sacrifice on LLR to run GFN-SV.

There is not much one can do with an Intel GPU, but I've found the OpenCL sieving (gfn opencl sieving app) will work with one laptop's Intel HD Graphics 4000 GPU. Its pretty poor performance, but it does work. Better than nothing. With this i3 CPU I am able to get it up to 2.4p/day when running two simultaneous tasks. Doing about 1.2p/day each using B9. Any higher and it refuses to run. No B10 or greater. Hardly anything compared with a 1060 at 105p/day but it is still something.

Yes, sieving does work on iGPUs...

...but it's just not worth it. The thing is that it steals so much memory bandwidth that the CPU tasks get starved. From my testing, even SGS (which is pretty tamed with RAM, in comparison to the other LLR projects) loses performance: 1 unit was fine, 2 had a small loss, 3 and 4 just got destroyed. And keep in mind, that was with dual channel, dual rank, 3000mhz DDR4, so it's not as if I had slow RAM there to limit me.

If there was a sieving challenge, I guess you could use the iGPU to squeeze as much as you can. But outside of that, it's not a good idea to sacrifice on LLR to run GFN-SV.

I'll have to run several tasks and see what the performance loss is while running the sieves vs not running them. This little 2 core i3 does llr workunits as fast or faster than the i7 laptop and my fx 8350.

It's worth testing yourself, yes. I tested way back in the day, Skylake was "new", LLR did not have multithreading, and numbers were smaller on some of the projects.

1 more day of PPS multi-threaded LLR going with one sieve running for GFN18 then I will turn off the sieve and run a day of PPS multi-threaded LLR and see how it compares time wise.

Well I found I couldn't run them at B9 for long, kept freezing. So now I have them running at B8, about 1.0-1.1P/day per task times 2 tasks. Not very impressive, but it's something. I stopped running PPS LLR in exchange for ESP, but I still would like to run a comparison on how the sieving affects CPU LLR tasks.

Using the W1 parameter in the command line greatly reduces the CPU usage. Still it's sure no powerhouse at 2.2P/day total :)