alright this was funny, thought completely unrelated to output
likely a skill issue on my part
glm 4.5 air q3_K_xl
>>106167608
as a 3060 owner myself, gemma3 with some spillover still gave an acceptable speed of ~6-7t/s at iq3_xxs
but if you spill over a lot like with a 32b and a higher quant you'll get 4t/s probably
cydonia with little spillover gives 10t/s (24b, iq4xs)
im pulling the speeds out of my ass because i dont remember them exactly but u get it