From lambda cloud:

3 hours * 4 ~22$/hr for 8x A100 ~= $265

So yeah not too expensive even for a native fine tune (obviously this ignores all other costs other than the GPUs)