hi, im really curious that gs does actually affect anything?

gs defines together with the blocksize the number of threads to execute. Blocksize is calculated and you cannot change it.
hmm, still dont get it. it affect the lifespan of gpu or use more cpu & gpu resources with gs size?
Maybe I'm too bad in explaining. Please check this information:
https://www.youtube.com/watch?v=kzXjRFL-gjohttp://cuda-programming.blogspot.com/2013/01/threads-and-blocks-in-detail-in-cuda.htmlThere are much more - just try to google 'cuda grid block'. You will get lots of information about grid/block/threads and how they are related to each other in GPU kernel programming.
Hope the helps more!
OK I'm no rocket surgeon here

If I have a 7 card rig with decent GPU's (1070ti's) with a low-end CPU (G4400 dual-core Pentium) ...
Should I focus on smaller gridsize (perhaps between 4096-8192)?
And if I had a better CPU, a larger gridsize may or should work better?
Hi,
I can only give you my personal opinion and preference only. there will be others out there with very different opinions, I guess you have to find the best settings for you and your configuration.
For the algos that are supported by TT the CPU doesn't play a big role as well as RAM. I know of systems that works well with 8GPUs and have a celeron and 4GB RAM only. You CPU, RAM-size and HD/SSD size will not have a very big impact on your overall performance. Changing the grid-size changes the configuration of the kernel that runs on your GPU. I prefer lower values since I noticed that on my system I see much faster stable hashrates. With higher/very high (over 8192) gridsizes I see the hashrate dropping over a long period until a stable level is reached, so I chose in most cases values between 256 and 4096. But this observation is very depended of the system and GPU you use. Also please to note that I do not run hundreds of GPUs - my experience if 1070 and 1060/6GB ONLY. I never made some investigations how different gridsizes impact the lifetime of a GPU - I leave that to someone else.
So my best advise for you is to make your own tests and find the settings that works best for you and the system you have.