Originally Posted by Murlocke
...unless my math is incorrect...
It is. Each of your EVOs has 1TiB of NAND, or 1024GiB on the PCB. After some overprovisioning and allocation for the write buffer, you end up with 1000GB (1TB) or 931.5GiB available for storage, but 1024GiB being used. Now, if we assume Samsung's 1000 P/E cycle rating is accurate, then we get:
1000 cycles * 1024GiB/cycle * 1.024^3GB/GiB * 1PB/1 000 000GB = 1.1PB
of writes before death.
If we assume your claim of 2500 cycles, then it's just the above times 2.5 (2500 cycles / 1000 cycles), so 2.75PB
And if we look at the somewhat more accurate 1500 cycles (at least, that's my impression from the endurance tests) then it's the first one times 1.5, so 1.65PB
before kicking the bucket.
All of that is based on a single drive, and my numbers show one taking 30 years at 250GB daily. Your RAID array would take 60 years to kill, or only 15 at 1TB every day. That seems off, but I think it's actually right or at least reasonably close. Or totally wrong, because I really, really do not want to have to go back and check my math.
As I said, lifespan is not a real issue. It would be like if governments started taxing 95% of your income above [currency symbol]1 billion. Sure, there would be a huge outrage from all sides, but it would not affect more than a handful of people. Regarding slow write speeds, past a certain capacity, yes, the NAND will no longer be a bottleneck. Rather, the interface or the controller will. However, the only way the EVO compensates for TLC's speed in lower capacity models is by using it as if it were SLC. Otherwise, we'd be back to the barely-bottlenecked-by-SATA I speeds found in the vanilla 840. M500s aren't the best drive to compare with, since the 128Gib modules roughly half the potential speed compared to 64Gib modules. SanDisk Ultra Plus? Fair comparison, and they smash 840s and post-buffer EVOs. I'm not sure of any other Marvell drives off the top of my head unfortunately.