11-18-2018, 06:55 AM
https://www.gamersnexus.net/guides/3394-...ck-screens
Quote:Just for good measure, we also took apart several cards and did a cursory board-level inspection. We were only really looking for anything extremely out of place, like missing thermal pads, poor contact, burned or damage components, and so on. There was only one device that demonstrated any physical defect, but it was unrelated to the issue of artifacting and is something we may discuss later. The thermal pad contact on all of the cards looks fine – we can see indentations and clear contact being made to the pads – and there are no obviously damage components. Any defect is going to be something we don’t have the tools or knowledge to see, as it’s likely in the board or in the silicon.
What we’ve done today is primarily rule things out – or mostly out, as we can’t speak with 100% certainty – like thermals, firmware, and Windows. We’ve also determined that blue screens were a separate issue, caused by early driver compatibility problems with many GSync monitors. Visible defects on the boards were not present, although we have no means of inspecting the internals of the board or the silicon.
Ubuntu Linux also exhibits artifacting issues (410 and 415), as does the newest firmware revision on the bad cards, and artifacting still occurs even when a water block is on the cards. We have to assume that thermals are not the issue on the cards we had. The biggest take away is that there’s not some magical “TjMax” trip-point whereupon artifacting kicks in. Even running our known-good review sample at 100+ degrees on the memory – decidedly a bad thing for the memory’s health – we still did not encounter a hard shutdown, thermal fail-safe, or artifacting. For anyone who thought artifacting was triggered upon hitting a certain threshold, it would appear that this isn’t the case. Sticking thermocouples on the cards also did not produce offensively high thermal numbers, although I1 was getting bad.
Speaking with some engineers in the industry, we’re left with an assumption that this is either a board-level assembly issue or an in-silicon issue. NVIDIA has noted that it is replacing any affected cards, so there’s that. The company described these units as “test escapes” and seems to think that the problem is fixed going forward. We have exceedingly high confidence that this is a hardware-level issue, but it does not appear to be an epidemic.