Here is a copy of my EWBF config (Mint Linux 18.2, Ubuntu derivative):
cuda_devices 0 1 2 3
intensity 64 64 64 64
You can leave intensity to the maximum (64), I didn’t see any impact on temperature. You have to control fan speed and to activate Nvidia CoolBit on your card (you can google it or check this thread: https://gist.github.com/bsodmike/369f8a202c5a5c97cfbd481264d549e9).
Then, you can limit power consumption and other settings (I run my 1070’s at 85% fan speed, and ~85% power and they run at 60-65 C).
I have this in a shell script to activate persistance mode (not sure I have to run more than once though…), and set power limits on all cards (nvidia-smi doc.: https://developer.nvidia.com/nvidia-system-management-interface):
# Set persistance mode for all GPU
nvidia-smi -pm 1
# Set GPU Max. power at 160w (86% TDP)
nvidia-smi -pl 160
./miner --config ./miner.cfg
Then, you can set fan speed and overclocking settings for each card using something like the following commands:
nvidia-settings -a '[gpu:0]/GPUPowerMizerMode=1'
nvidia-settings -a '[gpu:0]/GPUGraphicsClockOffset=100'
nvidia-settings -a '[gpu:0]/GPUMemoryTransferRateOffset=900'
nvidia-settings -a '[gpu:0]/GPUFanControlState=1'
nvidia-settings -a '[fan:0]/GPUTargetFanSpeed=85'
You have to repeat for each of your card, [gpu:0], [gpu:1], [fan:0], [fan:1], etc. for each card.
Monitor your system for any stability issues and adjust nvidia-settings accordingly. From experience, tweaking GPUGraphicsClockOffset don’t improve performance very much, but GPUMemoryTransferRateOffset does.