384 is preferrable, I'll tell you why, on linux there is no GUI like windows to config the over-drive/over-clock stuff,

Only 384 has the nvidia-settings/nvidia-smi cmd-line properly setup to allow you to control fan and memory clock, thus if you want to get the full power out of your CUDA-CARDS I would be running 384

In summary if you care to roll up your sleeves in the cmd-line world of "nvidia-settings" then go for 384, if you don't have any desire to max the over-clock potential then stay with a older release.

If your a newb & don't know what I said above, then google the stuff below and learn what it does ..

#!/bin/bash

sudo nvidia-xconfig -a --cool-bits=28

[ the above is a special flag that enables the following ]

nvidia-settings -a '[gpu:2]/GPUGraphicsMemoryOffset[3]=100'

nvidia-settings -a '[gpu:2]/GPUMemoryTransferRateOffset[3]=100'

nvidia-settings -a '[gpu:2]/GPUFanControlState=1'

nvidia-settings -a '[fan:2]/GPUTargetFanSpeed=80'

# you need to run the above everytime you boot the machine

#!/usr/bin/env bash

nvidia-smi --format=csv --query-gpu=name,power.draw,fan.speed,temperature.gpu,clocks.video,clocks.mem

# this is just a diagnostic script, so you can see the current settings before you fiddle

# memory transfer rate is the critical on cuda for ethminer or equihash +500 for gtx-160 +1500 for Gtx-1070, and +1800 for Gtx-1080
# graphics is best kept below +150 it doesn't do much for equi or eth mining, same for all cards

# keep you power below 60% of max rating with "nvidia-smi -pl N -i M" where N is watts ( 95 for gtx-1060-6), and M is the i' card in smi
***

nvidia-xconfig -a --allow-empty-initial-configuration --cool-bits=31 --use-display-device="DFP-0" --connected-monitor="DFP-0"


#You need to run line everytime you add or remove cards.