When I built the computer for my cousin, I set up Windows and browerse so that temp and cache and those type of files were written to the secondary HDD instead f fthe primary SSD. It's the constant writing/rewriting of data that wears out the SSD from what I read. Putting temp/cache folders on a secondary drive (i.e. not the drive OS is installed on) is what's recommended for higher speed regardless, because that way the two drives can work in tandem. Even if you have two HDDs and no SSD, it is better to put temp/cache folders on the secondary drive for improved performance.

The life expectance of HDD is affected byt shocks, being constantly powered and off. Writing and rewriting alone should not affect it the same way as for SSD. I think it also depends on the brand/model. I have Western Digital Black HDD and it's almost 10 years old and hasn't failed yet.

The ideal setup seems to be to have faster SSD for OS and programs, and larger HDD for temp/cache folders, documents, media, etc. That's what I would do if I build today.

I believe I have some clue, but I'm not an expert on ssd storage by any means.

Internally the individual units in the flash can only be erased so many times, and it's not actually a very high number. For maximum lifetime, the internal logic of the drive tries to level the abuse across the entire device, even when you just rewrite the same logical spot.

One thing that has changed since ssd hit the market is that the drives have gotten bigger. When there is more space to abuse, it can take more writes too.

Its not a wizardry to understand what impacts SSD life.

1) working temperature. Basic school physics, increase of temperature causes expansion - constant increase/shrink cycles put stress on soldered joints and current solder is hard (lead-free), so not very resistant. TL;DR: as with hdd - keep inner pc temperature within 25-45 Celcius range.

2) total write cycles (change of state). This is what manufacturers themselves base warranty upon. Multiply your drive total size by 4000 and you get the safe amount of total write cycles. Because devices have wear leveling, where you write on the drive does not matter. When you exceed it, chips start failing, drive starts relocating - thus as with HDD SMART, if you get increase in UBERs - SSD is not safe anymore.

3) write and read cycles also cause increase of temperature, so the way you use it also slightly affects life expectancy.

4) controller/pwm failures. as with (2), but is a roulette and drive dies instantly.

5) SLC=MLC basically.

Thank you for the interesting info dump, Draek. I'm no expert on hard drives so this is very useful to know. Of all the things that could happen, data loss would be the worst for me. And I didn't know about that virtual ram thing. Ironically, this means that some beloved classics like VTMB can't be played too often if one wishes to maintain hard disk expectancy!

I'd like to ask, what about direct downloading? Does it work the same way as bittorrent? How does it affect your hard drive read/write cycles?

When file is written its split into chunks by filesystem driver, while one file can occupy 1-to-n chunks, the n-files always occupy n-chunks unless filesystem can pack them into one chunk.
This means one file - either one (if it fits into 1 unit, typically 4KB) or many writes, where many files always take many writes, unless FS can pack tail / suballocate fragments.
But above does not apply on SSD, because - wear leveling.

Filesystem driver assumes that partition is stored linearly (0...n) on disk, but on SSD its projected (offset mapped) by wear leveling mechanism.

Means on SSD blocks 1,2,3 can be written as:
X,2,X,X,3,1 (where X is unused skipped block). And on deleting above and rewriting - next write will be:
3,X,1,2,X,X - SSD will not delete from chips, it will not use same chips, it will spread the writes. All writes are non-linear.

If one writes three sufficiently small files, the write is: 3, X , 2 , 1
And if one writes one big file that occupies three blocks its: 3, X, 2, 1
Same. Almost, because there is an additional factor when writing many small files - files also write additional metadata on filesystem, so one 1GiB file will produce same amount of write cycles as 0.8 GiB small files. Not much big difference.

However what Draek means is gigabytes of small files that constantly change, not by volume but by total write cycles. For this you are better with ram disk or hard disk if access speed is not critical. Torrent protocol does not write or overwrite much. It spreads payload into chunks to access them via checksums, but it does not change them. Read cycles do not affect SSD much.

---
Modern OS such as Linux, have two temporary locations:
/tmp/ - which is stored in RAM (basically a ramdisk), is for many small files like locks.
/var/tmp/ - which is stored on disk (basically regular directory), and is for big files.

This is done with SSD in mind, but can be changed via /etc/fstab or tmpfiles.d (man tmpfiles.d).

Actually, it's more like at least 1000x the capacity, and at least for Samsung it's almost this amount in the warranty: http://www.samsung.com/global/business/semiconductor/minisite/SSD/M2M/html/support/warranty.html

Also, every test I've read about this indicates that it usually takes at least two times, for some even as much as ten times this amount of data to be written on them before they start to fail.

By the way, my own 120GB SSD is now 4 years old and has about 14TB of data written on it. I've pretty much stopped worrying about this, if it fails, it probably won't be from too many write cycles. The "wear leveling count" has only decreased from 100 to 93 during this time, so if that is any indication for how long it will last at this rate (which, as far as I know, it is), it will still work for decades to come.

Here is what Microsoft has to say about SSDs and pagefile.(Just search for pagefile)

TLDR: Keep Windows' pagefile on the SSD.

Also keep in mind that the article is from 2009. Since then Microsoft has only gotten better at handling SSDs.

If you read that paragraph throughly, you'll notice that they've managed to contradict themselves even there.
TL;DR: its a good SSD killer.

Hmm.. I see. Perhaps after reading this you will understand?

Am I being trolled? From your very own article:
"Windows 10 writes pages out to disk only 50% as often as previous versions of the OS. If all goes according to plan, Windows users could be experiencing reduced waiting times for all devices as well as extended lifespans on systems that have flash-based hard drives."

You still haven't mention what the contradiction was in the original article.

Well, I had some share in developing the technology behind SSD so allow me to say...

(wait a minute, I have to read my legal contract again)

Nope, 43 years must pass before I'm allowed to discuss things I know from my previous job ;)

Just a general statement then: don't worry too much. Read a few statement how to use SSD effectively and you should be fine. Enjoy your life ;)

Its not my article, its article that I've linked. The 50% are marketing talk. Thing is:
- windows likes to read into memory, but very much dislikes to release it
- since windows 8, it reads and keeps entire applications (not just pages) in RAM and likewise aggressively swaps them out.
- some values in first article are vague and useless (40:1 ratio - the rations are useless like percentages, if "1" is not given).
- some values in first article are contradicting, as they themselves state that chunks are plenty and small. In reality windows will gladly write 1/2 of RAM (GB) of data - per session. Notice, that what matters is the total write IO size, not the allocated size; moving data around also contributes to writes. Has anyone done some metrics how much it moves around?

So it is an SSD killer. However:
- SSD also gives best random access times and this is very useful for swap,
- the SSD write cycles do not depend upon what or where is written, but how much is written in correlation to the total SSD size. So having a larger SSD gives more reliability than having a smaller one (unless controller burns, that is).

So if you pick a large SSD and limit pagefile to partition, it will be less.. killer, more profit. I don't think it allows to adjust more, like swappiness value? Its good to keep stuff in RAM - but SSD are NAND, not DRAM.. The best strategy would be to offload swap into compressed ramdisk, - and then offload into hibernation file on SSD on power off, whilst preferring IO cache in RAM to swap (low swappiness), but I doubt its possible with ms.

Okay, I see... I avoid discussing with conspiracy theorists, so this is the last I'll post about this.

The fact is that I have yet to find a reputable study detailing the negative effect Windows' pagefile has on SSDs. Also I don't view Microsoft as a great evil, when everything can be explained by them being stupid. So by default I still trust Microsoft when they say that their memory management solutions extend the lifespan of SSDs.

To each their own.

So far my reluctance to use SSDs (as a system drive) isn't really related to fearing of my data (as I would keep the most important data backed up on external HDDs anyway), but the price difference. I like having a big ass 2 terabyte system drive where I don't have to care how many gigantic games I've installed at the same time and crap downloaded from the internets, and it still costs too much as a SSD to my liking. I've been just waiting for prices to come down before jumping to the SSD bandwagon.

EDIT: My earlier thought of SSD umpowered data retention was apparently debunked already:

http://www.pcworld.com/article/2925173/debunked-your-ssd-wont-lose-data-if-left-unplugged-after-all.html