HERETIC wrote:Pretty sure Sammy taking their 3D back to 30/40nm-endurance is comparable to planer MLC............
No, it isn't. Samsung documents its 950 and 960 endurance ratings and they are lower than any planar MLC drives I've seen endurance ratings for. They only reverted 3D to older process nodes because the endurance of 10nm class 3D NAND was not even marketable.
My understanding is IMFT isn't really a thing anymore, because Intel did not jointly buy into Micron's current generation of 3D NAND production. Micron fronted that node on its own and there's so far been no word if Intel will resume it's IMFT partnership for the next-generation of Micron NAND.
CScottG wrote:In essence Intel is running a very protective protocol for that drive, shutting-down writes LONG before the drive would fail. ALSO, it might be utterly dependent on the wear the SLC buffer is undergoing, not necessarily how degraded the vertical NAND has become.
To be fair to Intel, THG spamming 4K writes until the drive dies will blow the write amplification factor completely beyond normal usage scenarios, so I'd fully agree that 600p died way sooner than it should have. TR's own testing methods at least paused to allow drives recover and optimize the NAND every so often which is critical to minimize the write amplification factor.
That being said, the ~110TB failure is still close enough to the 140TB endurance rating that I would take the endurance rating at face value. And the point is moot because Intel locks the drive upon reaching the endurance rating anyway.
CScottG wrote: (..they did note that the SLC buffer was "inconsistent" - my guess is that after a few days that buffer was in operation a LOT, and that as the algorithm advanced and the wear level detected (becoming more worn), it became increasingly "inconsistent" in an effort to prolong drive life.)
The SLC cache buffer is a funny thing. SLC NAND has literally an exponentially longer endurance than TLC NAND because of how the voltage situation works. So while the SLC portion will see an absurd number of writes, it also has an absurd amount of endurance. I have not seen any PC hardware sites analyze this equation in depth so I can't really say how it balances out, but my feeling is the SLC cache portion will still be reasonably close to endurance parity with the rest of a TLC drive.
Any test that only spams continuous writes without pause is going to effectively break SLC caching anyway, so any odd perf results they saw don't mean anything. Perf inconsistency is expected during the last 10% too.