[ubernostrum]

as soon as you've written an amount of data roughly equal to the SSD's total capacity

After eight months of heavy daily use, in which I'm quite certain I've achieved more than 256GB of total writes, I've seen... no change in performance. Still zippy and happy and writing just fine.

Time for misguided hype based on anandtech's articles to go away now.

[greghines]

As I said, it depends on usage patterns. (I also oversimplified the problem description. My apologies.)

If you're only using a small fraction of the total capacity, chances are you won't see much performance degradation for a long while no matter how much you write to the drive because of minimal fragmentation at the block level. But given enough time, you'd still hit that wall at some point.

But if you're using a large fraction of the total capacity, then you're going to have much more fragmentation, which means pages that can't be erased without having to rewrite significant numbers of blocks in those pages. Those ephemeral writes will start taking a real performance hit.

Many newer SSDs now include logic in their controllers to help mitigate this problem, with varying degrees of success. But full TRIM support on the drive and in the OS make a significantly larger difference than anything done by the drive alone. (Even with TRIM, though, the problem doesn't go away completely.)

And this isn't just some kind of spin from Anandtech--TRIM exists for a reason. It's a well-tested property of MLC SSDs. The combination of wear leveling and write combining can cause serious performance issues without TRIM. This article describes it pretty well: http://www.pcper.com/article.php?aid=669&type=expert&...

[ubernostrum]

If you're only using a small fraction of the total capacity

At any given moment I'm using about half the total capacity. I'm not regularly doing huge writes, but as someone who works with a ton of remote and local repositories, pulling/merging/etc. many times a day, I do end up doing a huge number of small writes nearly continuously; I'd be surprised if I haven't hit every bit of that drive at least once by now. And I haven't seen any dramatic, or even noticeable, degradation.

And this isn't just some kind of spin from Anandtech

It's more that a bunch of people skimmed the anandtech articles, then started spouting a lot of gratuitous "X25-M is the One True SSD, everything else is shit" hype, which later morphed into "never use an SSD in an OS that doesn't do TRIM" hype. Of course, the hype doesn't quite match the reality...

[greghines]

Well, it's not so much which parts of the SSD have been touched, it's more about which pages have currently-in-use blocks stored in them. If you're using half the capacity of the drive, then it's likely you have a considerable number of empty (or at least erasable-without-needing-extra-writes) pages--especially if your drive's controller does some basic defragmentation.

And no, SSDs aren't worthless without TRIM. But if you're likely to use the majority of the storage space on the drive and do a lot of writes besides, it could be the difference between "fast" and "really damn fast." Either way it'll still beat the pants off most rotational disk drives.

I think the takeaway, though, is that if you've spent potentially-obscene amounts of money on an SSD, wouldn't you prefer that your OS take full advantage of its capabilities to consistently deliver top performance?

(And no, that's not a dig at OS X. I'd recommend that Windows XP users upgrade to Win7 if they're considering upgrading to an SSD simply for the TRIM support. It can make that much of a difference depending on the user.)

[jrockway]

As another datapoint, I had a laptop with a first-gen SSD and no TRIM or GC support. The performance hit a literal cliff; writes were fast one day, and the next day they were literally 300KiB/s. Download speeds were limited by disk write bandwidth, not network bandwidth! New SSD, writes are back up to 50MiB/s.

(The divide might not be at the capacity of the disk; this was a 16GiB SSD with probably 20x that in writes.)

[cpr]

Well, after 6-7 months of heavy use on my MBP (easily past 128GB worth of writing, many times), Xbench says I'm getting 100MB/sec for just about all disk activity with 256K blocks.