Samsung SSD 840 (250GB) Reviewby Kristian Vättö on October 8, 2012 12:14 PM EST
Lower Endurance: Hardly an Issue
With perfect wear-leveling and write amplification of 1x, you would get 256,000GiB (that's ~275TB) of writes out of a 250GB Samsung 840 with TLC NAND and 1,000 P/E cycles. That is still a lot but wear-leveling and write amplification are never perfect. Giving any specific numbers for endurance is hard because every drive behaves differently and users have different workloads, but it's unlikely for a light consumer workload to see more than 10GiB of writes per day. That's 3,650GiB per year, which is only 1.4% out of 256,000GiB. In the real world NAND writes will be bigger than host writes but even with a write amplification factor of 10x, you will only end up writing 36,500GiB each year and exhausting ~143 P/E cycles out of the available 1,000. In other words, it would take roughly seven years for you to wear out the NAND.
|SSD Lifetime Estimation|
|NAND||MLC—3K P/E Cycles||TLC—1K P/E Cycles|
|Writes per Day||10GiB||10GiB||10GiB||10GiB|
|Total Estimated Lifespan||10.5 years||21.0 years||3.5 years||7.0 years|
For the 120GB Samsung 840, the lifespan is half of the 250GB model but we are still talking about years. Samsung doesn't offer a 60/64GB Samsung 840, although that makes sense as it wouldn't be hard to wear that out in less than three years, which is the warranty Samsung gives to the 840 SSD.
DSP to the Rescue
However, there is actually more to SSD endurance than just P/E cycles and write amplification. There has been a lot of talk lately about digital signal processing (DSP) in the industry, which is supposedly the solution for lower endurance NAND.
The basic idea behind DSP is very simple: you read changes in voltages and adapt to the changes. As I mentioned in the previous page, the voltages change as the NAND wears out and if your controller can't adapt to the changes, you'll be stressing the NAND even more. Each time you're trying to program or erase the cell, you are wearing it out, so you don't even have to succesfully program or erase the cell to cause damage. That's why the guess and test process for writing to NAND is so harmful; it may take multiple tries and each try will wear out the NAND even more.
Graphical presentation of a change in voltage state
However, if your controller can read the changes in program and erase voltages, you will know what voltages to use to program/erase the cell. Even though DSP doesn't make NAND immortal, it causes a lot less stress on the NAND, allowing it to last for more P/E cycles than what you would get without DSP.
Again, it's hard to give out any specific numbers of DSP usefulness in real world, but for example STEC is claiming that their CellCare technology can extend the endurance of regular 3K P/E cycle MLC up to as much as 60K. I've heard unofficial figures as high as 100K for some companies' DSPs, but I would take all figures with a grain of salt until they are tested by a third party. Either way, even if a good DSP is only able to double the endurance of NAND, it's a huge deal as we move to even smaller process nodes and possibly even more bits per cell.