Just a heads up, I've noticed that many people say +500 or +600 or even +700 for their memory OC which is subjective depending on your system settings. Sort of learnt this the hard way:

In CUDA / compute tasks, a stock GTX 1070 runs memory at 7600 MHz (base 3800*2). So a +600 OC will result in (base 3800+600)*2 = 8800 MHz. In 3D tasks, this same +600 OC will result in (base 4004+600)*2 = 9208 MHz, which can be unstable /crash the system.

You can use Nvidia Profile Inspector to set "Force Cuda P2 State" to "OFF". This prevents the base memory clock from dropping to P2 state while mining (keeping it in P0). Result: +500MHz memory OC gets you to 9GHz in BOTH gaming and CUDA/compute tasks, which should be safe for most.