Hello. Can you add argument --intensity, or "intensity":value in JSON-config,
to set intensity of mining on each core, and decrease this intensity, using trottling, or some delay?
Because limitation of number of the CPU-threads, decreasing usage of CPU-cores,
but loading on this corea is 100%, and temperatures is maximum, on each loaded CPU-core.
Not good...
What if need to load 8 cores up to 25%, and no more?

Thanks.

@wisnia37 CUDA doesn't use intensity or worksize you're thinking of OpenCL. As the autoconfig for CUDA is generated inside the xmrig-cuda plugin there are no commandline options for threads/blocks, and the ones for bfactor/bsleep don't seem to work either:

      --cuda-bfactor-hint=N     bfactor hint for autoconfig (0-12)
      --cuda-bsleep-hint=N      bsleep hint for autoconfig

So then let it generate the highest possible threads/blocks and turn them down (and/or add bsleep/bfactor) by hand.

@username1565 Repaste your heatsinks if you can't run 100% 24/7 without hitting thermal throttling.

25% would be running autoconfig once for a "full" setup and then go in and delete 75% of the threads in some balanced way.

There is an option for this (--cpu-max-threads-hint=N maximum CPU threads count (in percentage) hint for autoconfig ) but I've never had it work right, therefore I just let it calculate 100% and then carve it away by hand.

@Spudz76, yes, now I using threads:2 in algo config-section, instead of --cpu-max-threads-hint=N, and I see 2 cores from 8 was been loaded. But this is loaded up to 100% and temperatures there about 100 celsium.
I just wanted to use something, like --max-cpu-usage=25 to set 25% of loading for each core, and do not see so high temperatures, on each core. Maybe there is possible to use trottling, or time delays, to implement this.

Turned out it is straightforward to slow down the mining process, by adding some sleep code here:

xmrig::Job xmrig::Miner::job() const
{
    std::lock_guard<std::mutex> lock(mutex);

    return d_ptr->job;
}

I will make a pull request ASAP.