Are Micron chips causing most of the 2080ti problems?

[hdtvnut]

I originally set the OC on my ti Strix 011G to +1000 for memory, with core boost +100 and fans set to hit 100% at 70 deg, and was running benchmarks OK. But when I did stress testing, I got fails, and had to back off to +800. This was accompanied by a marked decrease in heat buildup from low-to-mid 70's down to 65 deg. Since I have MIcrons, I'm thinking these may be overheating if the clock is set much above +800, even with the thicker heatsink and three fans. So my question: if some are setting +1000 or more, could this be leading to early chip failure?

[Arne_Saknussemm]

Overclocking memory for gaming or normal daily use is a really bad idea. Even if you do do it, it is best to run multiple benchmarks to find the sweet spot which is often back from maximum, apparently stable, frequency. And yes it will shorten life...

For a 24/7 GPU OC it's best to see what you can get on the core "for free" i.e. stock voltage or with a slight bump in voltage and upping the power slider. Short benchmark runs OC the memory but stay away from excessive heat and of course artifacting/instability

[Silent_Scone]

hdtvnut wrote:
I originally set the OC on my ti Strix 011G to +1000 for memory, with core boost +100 and fans set to hit 100% at 70 deg, and was running benchmarks OK. But when I did stress testing, I got fails, and had to back off to +800. This was accompanied by a marked decrease in heat buildup from low-to-mid 70's down to 65 deg. Since I have MIcrons, I'm thinking these may be overheating if the clock is set much above +800, even with the thicker heatsink and three fans. So my question: if some are setting +1000 or more, could this be leading to early chip failure?

Are you referring to the OC Scanner when stress testing? The algorithms are designed to detect instability, not all samples will be able to obtain a stable 1000MHz offset on the memory. The hardware failures observed by users likely aren't directly synonymous with memory overclocking, but it is a memory failure none the less.

I've been running with an 800MHz offset from launch on water without issue.

[hdtvnut]

Sorry, I didn't say, but 3DMark Time Spy Extreme, etc., stress tests.

Arne, I'll follow your suggestions.

[Silent_Scone]

hdtvnut wrote:
Sorry, I didn't say, but 3DMark Time Spy Extreme, etc., stress tests.

Arne, I'll follow your suggestions.

Same applies here, too. If you're experiencing hanging then the memory offset likely isn't stable.

[hdtvnut]

Using the results of both stress tests, I eventually arrived at +100 core and +750 mem OC for this card in free air. To work as well in the closed case, I replaced the front two 120 case fans with TY-143 hi-speeds, and connected one of them to the card. It passes the Time Spy Extreme 20 min stability tests with final 71 deg temp; graphics score 7190.

[Korth]

"NVIDIA RTX 2080 Ti will use GDDR6 from Micron, Samsung, and Hynix."
https://www.pcgamesn.com/micron-gddr6-launch-partner-nvidia-rtx

Micron was the only real supplier of GDDR5X. It's not unreasonable to assume they'll be the leader with GDDR6.
Samsung made the very best DDR4 in the end. Hynix made the best DDR4 in the beginning. But Micron made (still makes) more good and more not-so-good DDR4 than anyone else.

Micron GDDR5 was known to limit overspec performance in GTX 1070/1080 cards. But this doesn't mean Micron only makes garbage silicon - they make low-spec and they make high-spec - it only means the card makers used not-high-spec (Micron) garbage when better silicon wasn't available within their price targets. There were plenty of cards which overclocked very well with high-spec Micron silicon but they didn't get much attention. The card makers can easily improve or wreck any card through their component choices.

https://rog.asus.com/forum/showthread.php?90119-Memory-Micron-vs-Samsung-vs-SK-Hynix-etc

We don't really know (yet) whether Micron GDDR6 is a great overclocker or not. But poor-perfoming products in the past which happened to (mis)use older-generation (and lower-spec) Micron aren't really relevant indicators in themselves. It's really more about what volume is being manufactured, more parts total equals more parts binned for top-spec, and Micron today has advantages over Samsung/Hynix in bulk production that they didn't a year or two back.

Just sayin' that maybe your card uses Micron, maybe not. But it's too early to know if it matters or which way Micron's dice have rolled this time around. Plus regardless of what's inside you could simply have an overclocking epic or an overclocking lemon, as always ... patterns don't emerge until lots of samples can be compared.

[hdtvnut]

Although getting good stability and temp numbers with an initial Time Spy Extreme stress test, I noticed upon repetitions with the system warmer, the numbers deteriorated. I decided to divide the mb with cardboard partitions so the cpu and gpu each was cooled only by its own fan. Front to back flow; no exhaust fans. This seemed to maintain the test values, about the same as with the case side off: 97.6% and 70 deg. I think the partitions are worthwhile, and plan to make some better looking ones out of black styrene sheeting. I have a U14S cooling the 8700K, and think this is a pretty good setup for all air cooling in a mid tower. More noise, but more sleep at night.