Please try updating to Crimson 17.12.1.
at first i mark the answer as UNCORRECT!!!!!!!! of amdmatt
I tested EVERY driver since i bought the card, now on Alpha 18.1.1 adrenaline
I tested for months with every possible setting, even borrowed a 200€ PSU 1000W be quiet dark power pro, it delayed the crashes a litte (maybee 10min)
I respont to my own thread with another response i posted before and i hope it is the reason, testing still in progress but it is the best bett
I have the same problem, sometimes my screen ends green or grey or black, but it is a crash
found a possible fix on the forum
i also tested 4 power supplies, 2 motherboards and oher things
my rx 480 runs totally fine so it must be the vega56
testen now the fixed hbm rate @ 500 in desktop while a game runs and youtube is playing
when it really is this switching issue of the ram then amd should fix it quickly with a bios flash or driver update
Update ( experince since card baught 1 week after launch and post December 8)
Tested Adrenalin 18.2.1, and 18.2.2 (and all between) no change with the crashes, only work arround is still fixing 500Mhz HBM as min state (state1)
Sounds like you have a faulty GPU.
The GPU is 100% stable, once i disable state 0 for the HBM.
Several users, in different forums have similar problems and the solution is to lock the HBM.
I assume it is either a design flaw or a driver error but i wouldn't think, that someone who rmas his card 3 times does always get a "faulty" gpu. Re: RX Vega64 Liquid -- Possible HBM2 Downclocking / Upclocking / Black-Screen issue
I think a certain combination of high resolution (4k@60Hz DP) and switching in low HBM states is to blame.
The reason for not many people reporting this errors is in the low saturation of 4k monitors in the market, and few people using the cards at low/med performance. most people mine or game (high load) or do just desktop work (low load).
I another point is, that the test lab notes on the drivers never mention a vega+ryzen combination for testing, there could be another problem.
I still think this problem could be fixed by drivers by using a bit more power at stae 0, or a little higher clocks (250 instead of 167) or a less agressive switching behaviour in low/med load scenarious.
I would appreciate it if you could forward this to AMD testing.
(Btw, newest driver, no changes, 15-20 drivers didn't change the behaviour)