Does a SSD still slow down when you fill it up?

[Peterg]

Or no ?

[Cave Yeti Supreme]

Look up these terms - write amplification, garbage collection, and TRIM. Early SSD's (think 2006, 2007) were absolutely atrocious performers because these concepts hadn't been perfected, so the very earliest consumer SSD's using Marvell controllers (which makes modern controllers that aren't bad at all) were often worse performers than standard hard drives.

[vadim]

Writing speed decreases. Reading from old files remains the same. New files can be fragmented, so it will be accessed with lower speed.

[Bad 💀 Motha]

^This
Plus it can also be a matter of, does the ssd have over-provisioning pre set aside and/or has the user set this manually if allowed for said ssd, to a reasonable amount.

I used to hate on models of ssds in the early days due to the size, such as say Corsair 240gb. However I later realized I misunderstood a few things and that having am ssd such as Corsair 240gb in the end was probably better then having most other brands that came in a size of say 250 or 256 gb. Why? Because the Corsair 240gb still left you with a decent size after a format, and at the same time already had onboard over-provisioning set aside internally by default. Ssds such as Samsung 850 series did not, but allowed it to be setup via Samsung Magician. But if I were to set a decent enough size for that on a Samsung ssd, I would have lost alot of usable space in the end. Space I don't lose on the Corsair Force GT series because those already had it set aside for me.

[Peterg]

Thanks everyone!

One last question if I did fill up the SSD to lets say 90 % to 95% if I unfill it back down to 70% or less will writing speeds go back to normal or would I have to reformat the drive ?

[Bad 💀 Motha]

Yes of course that is fine to do if needed for the short term. Yes performance would increase as you free up space, as fragmentation overall should decrease. Sometimes an SSD TRIM may be needed for the performance to come back tbough. But if let's say before bed, you leave the PC on over night, you can force the TRIM manually and it'll be done way before you get up in a few hours.

I would not advise doing this on the OS drive though. That drive it is always best to never go below 40gb. But again that is mostly to leave ample room for Windows Updates. So a user could easily use common sense and still run the drive to nearly full if needed, again in the short term.

[Peterg]

Publicado originalmente por Bad 💀 Motha:

Yes of course that is fine to do if needed for the short term. Yes performance would increase as you free up space, as fragmentation overall should decrease. Sometimes an SSD TRIM may be needed for the performance to come back tbough. But if let's say before bed, you leave the PC on over night, you can force the TRIM manually and it'll be done way before you get up in a few hours.

I would not advise doing this on the OS drive though. That drive it is always best to never go below 40gb. But again that is mostly to leave ample room for Windows Updates. So a user could easily use common sense and still run the drive to nearly full if needed, again in the short term.

Makes sense thanks.

[Cave Yeti Supreme]

SSD's do not suffer from fragmentation. They suffer from write amplification, which "ties up" the controller on the SSD from doing writes and reads, exhibiting "slowness". Background garbage collection, and TRIM, resolve these issues.

Fragmentation is purely a spinning rust phenomena endemic to the mechanics of the medium, of which SSD's share virtually nothing.

[Bullwinkle]

You do not want to fill up an SSD, it can ruin it. They say 90% is max, but I won't go over 80%. If it gets to that point, I'm shopping around for another SSD. If you have it filled up and you start deleating files, it should speed up. Also, what type of SSD, is it a SATA 3.0, 2.0? Do you have it plugged in to a USB 3.0 or USB 2.0? These things matter. Want to test you SSD or Flash Drive Speed? You can download a free program called USBDeview from http://www.nirsoft.net

[vadim]

Publicado originalmente por ANACHRONiSM:

SSD's do not suffer from fragmentation. They suffer from write amplification, which "ties up" the controller on the SSD from doing writes and reads, exhibiting "slowness".

I quote: "write Amplification is the ratio of actual data written to the flash vs data requested by the host to write to the device". Why SSD actually writes more data than it should? Because when it has insufficient free space, it has no other choice than to write it down in parts.
SSD can write down by pages, but every page can be written only once. Before rewriting you need to erase it. But every single page cannot be erased. You need to erase whole physical block that is much more. Like 128 pages.
Over time, as the SSD is used, fragmented blocks accumulate, only part of the pages in which are in use. Such blocks called "journal block" as opposite to blocks tbat contain contiguous LBAs that called data blocks.
SSD can track only limited number of journal blocks, so to create new, it should convert some old journal blocks back to data. To do that, it reads all journal blocks write its pages to new block from the pool of free blocks as contiguous LBA sequence, erases journal blocks and add them to the pool of free blocks.
I.e. in-block fragmentation causes many unnecessary writes. But even worse, it causes erases that not only wear SSD cells, but also very slow. Typical time for erase operation is 2 ms.

[MancSoulja]

Just create a partition that's 10-15% of the disk, hide it and use your SSD normally.

[Cave Yeti Supreme]

Publicado originalmente por vadim:

Publicado originalmente por ANACHRONiSM:

SSD's do not suffer from fragmentation. They suffer from write amplification, which "ties up" the controller on the SSD from doing writes and reads, exhibiting "slowness".

I quote: "write Amplification is the ratio of actual data written to the flash vs data requested by the host to write to the device". Why SSD actually writes more data than it should? Because when it has insufficient free space, it has no other choice than to write it down in parts.
SSD can write down by pages, but every page can be written only once. Before rewriting you need to erase it. But every single page cannot be erased. You need to erase whole physical block that is much more. Like 128 pages.
Over time, as the SSD is used, fragmented blocks accumulate, only part of the pages in which are in use. Such blocks called "journal block" as opposite to blocks tbat contain contiguous LBAs that called data blocks.
SSD can track only limited number of journal blocks, so to create new, it should convert some old journal blocks back to data. To do that, it reads all journal blocks write its pages to new block from the pool of free blocks as contiguous LBA sequence, erases journal blocks and add them to the pool of free blocks.
I.e. in-block fragmentation causes many unnecessary writes. But even worse, it causes erases that not only wear SSD cells, but also very slow. Typical time for erase operation is 2 ms.

...And since SSD's have no moving parts (hence "solid state"), they suffer no read penalty from so-called fragmentation; modern, competent SSD controllers employed in low queue depth workloads handle wear leveling and garbage collection during idle condition between accesses transparently to regular users, and in conjunction with a TRIM-capable OS, will over time eliminate partially written blocks automatically.

Write amplification and issues with under-provisioning can become an issue as you raise the queue depth and hammer the medium with constant sequential writes in certain workloads, but this isn't due to "fragmentation", and it has virtually zero impact on reads. These scenarios are where you pony up the Samsung Tax and pay for the more capable controller and NAND durability of a Pro series.

[vadim]

Publicado originalmente por ANACHRONiSM:

...And since SSD's have no moving parts (hence "solid state"), they suffer no read penalty from so-called fragmentation

You seems didn't understand anything from my message. BTW, the absence of moving parts does not mean the absence of delays when writing to different places of the drive. Sequential write to SSD is always faster than random. For obvious reasons.
It's always faster to write to the adjacent logical block than to the block in different place of the drive, isn't it? The same about reading.
But my message wasn't about that.

[Bad 💀 Motha]

Publicado originalmente por MancSoulja:

Just create a partition that's 10-15% of the disk, hide it and use your SSD normally.

No just set that % as over provisioning.

Please look up what this does to your ssd and what it's purpose is please before just suggesting non-sense

[nullable]

Publicado originalmente por Bullwinkle:

You do not want to fill up an SSD, it can ruin it.

No. It may not perform optimally, but it won't be ruined. It's just space, you're not gaining anything making up crazy hoops to jump through.

Publicado originalmente por petergg26:

Thanks everyone!

One last question if I did fill up the SSD to lets say 90 % to 95% if I unfill it back down to 70% or less will writing speeds go back to normal or would I have to reformat the drive ?

Sure speed would go back to normal. But I think this is a non-issue. The lower speed from being near capacity doesn't make the drive unusable. You're still going to get lightning fast performance compared to a hard drive. It's just less than peak performance.

SSDs are nice, but there's this weird temptation for users to start fussing over minor details like they're critically important. My advice is just use your SSD like storage and if you get it really full AND you experience some performance consequences, sure, address it. And if you don't then you'll know a lot of the people preaching doom and gloom are just parrots.

[SHREDDER]

If it is a high end SSD like the MP 600 2TB that i got one week ago no. Right now it has 1460 GB left because of cloning windows from the 1TB HDD that i had them installed(so it copies also evrything else i had installed on the HDD to the SSD too). But this is a high end 430 euros SSD of course it will not slow down.