[SOLVED] What performance decrease would you see with either a HDD or SSD coming to full storage?

[ShangWang]

Is there a difference in gaming performance when either a SSD or HDD is near 80% storage?

Will this only affect download and copying speeds, or does it have any impact on FPS, boot time, app responsiveness, or anything else?

When would you start to see any decrease in performance, and why? 50%?
Will a HDD or SSD suffer differently from each other?

What if you had both a SSD and HDD with the OS on SSD and games on HDD, what happens if one was more full/over 50%?

How would you benefit from having 2 hard drives? Does it increase performance in any way or is it just for managing storage?

 

[DimkaTsv]

Is there a difference in gaming performance when either a SSD or HDD is near 80% storage?

To be short... yes there is
To be long. Main difference will be in long sequential write speed with large block size.
With smaller block size difference will be much less.
With random performance difference will be minimal, but long queue will suffer a bit more than short queue.

Read performance will have close to no effect.
For example, i have Kingston KC 2500 1TB.
Crystal disk mark with 5 runs 4GB test file size
0% filled drive have:
Sequential 1mb Q8-T1: 3525 read, 3031 write
Seq 128kb Q32-T1: 3525 read, 3002 write
Random 4kb Q32-T16: 1426 read, 1441 write
Random 4kb Q1-T1: 69 read, 242 write

Spoiler: Proof

With 48% filled drive (maybe i managed to do something with optimisation i guess...):
Sequential 1mb Q8-T1: 3529 read, 3024 write (but sometimes it can be 2650 write)
Sequential 128kb Q32-T1: 3526 read, 2992 write
Random 4kb Q32-T16: 1470 read, 1438 write
Random 4kb Q1-T1: 68 read (up to same 69 that was before... i have stuff in background atm), 234 write.

Spoiler: Proof

So with almost 50% usage, difference in speed is... variable, but can even be non-existant.
But you still can get decrease in performance after 80% load as people said before

 

[ShangWang]

To be short... yes there is
To be long. Main difference will be in long sequential write speed with large block size.
With smaller block size difference will be much less.
With random performance difference will be minimal, but long queue will suffer a bit more than short queue.

Read performance will have close to no effect.
For example, i have Kingston KC 2500 1TB.
Crystal disk mark with 5 runs 4GB test file size
0% filled drive have:
Sequential 1mb Q8-T1: 3525 read, 3031 write
Seq 128kb Q32-T1: 3525 read, 3002 write
Random 4kb Q32-T16: 1426 read, 1441 write
Random 4kb Q1-T1: 69 read, 242 write

Spoiler: Proof

With 48% filled drive (maybe i managed to do something with optimisation i guess...):
Sequential 1mb Q8-T1: 3529 read, 3024 write (but sometimes it can be 2650 write)
Sequential 128kb Q32-T1: 3526 read, 2992 write
Random 4kb Q32-T16: 1470 read, 1438 write
Random 4kb Q1-T1: 68 read (up to same 69 that was before... i have stuff in background atm), 234 write.

Spoiler: Proof

So with almost 50% usage, difference in speed is... variable, but can even be non-existant.
But you still can get decrease in performance after 80% load as people said before

Thank you for those statistics!
So if you're under 50% space usage, would you say both wearing from writes and speed would not decrease whatsoever?

Any percentage used from 10-49% has no affect on performance or life because there's plenty of space?

If not, how come a SSD will still overwrite cells but to a lesser degree if there's 50% of space, which I assume is 50% of free cells to write over?

Edit: On a side note do you know if virtual memory from a page file will also decrease in performance from a fuller drive or not?

 

[USAFRet]

Thank you for those statistics!
So if you're under 50% space usage, would you say both wearing from writes and speed would not decrease whatsoever?

Any percentage used from 10-49% has no affect on performance or life because there's plenty of space?

If not, how come a SSD will still overwrite cells but to a lesser degree if there's 50% of space, which I assume is 50% of free cells to write over?

Edit: On a side note do you know if virtual memory from a page file will also decrease in performance from a fuller drive or not?

Any write action to that drive will 'suffer' if it is near full.
This includes the page file.

Percentages is irrelevant.
You can't just say "0-49% is OK, 51% is cause of death".

Just don;t fill it up past 75-80% or so, and don't stress.
Use it until it dies.

 

[DimkaTsv]

So if you're under 50% space usage, would you say both wearing from writes and speed would not decrease whatsoever?

I said nothing about wearing btw. Wearing speed will depend on how active you write data divided by how much space you have left. More data - faster. Less space left - faster.
Also remaining size of drive can affect size of "SLC" cache, because of trick manufacturers can use to emulate these speeds. (controller will use TLC as SLC until all space is full)
Some drives have decent SLC cache even besides this trick, though.
Same reasoning is why speed of sequential write will go down closer you go to filling the drive. Because there will be less space to use as SLC without swapping for different row of TLC cells (because something is written on top of them).
Also SSD doesn't quite wipe data. When you delete something, you just making controller forget that "there is actual data" so next time controller can overwrite what was there by resetting and filling with new data.

Not professional there, just something i picked up while looking for different stuff.
Also... Wearing SSD now is much harder job than you may expect compared to old times. Because besides cells that are wearing and weared already, all NVME (and most likely SSD too) have such thing as "SSD spare cells"... When some cell goes down because of wearing, one of these spare cells comes instead to replace it. And number of these is actually not that low.
So to get actually broken SSD you need to wear it AND run out of these spare cells

 

[ShangWang]

I said nothing about wearing btw. Wearing speed will depend on how active you write data divided by how much space you have left. More data - faster. Less space left - faster.
Also remaining size of drive can affect size of "SLC" cache, because of trick manufacturers can use to emulate these speeds. (controller will use TLC as SLC until all space is full)
Some drives have decent SLC cache even besides this trick, though.
Same reasoning is why speed of sequential write will go down closer you go to filling the drive. Because there will be less space to use as SLC without swapping for different row of TLC cells (because something is written on top of them).
Also SSD doesn't quite wipe data. When you delete something, you just making controller forget that "there is actual data" so next time controller can overwrite what was there by resetting and filling with new data.

Not professional there, just something i picked up while looking for different stuff.
Also... Wearing SSD now is much harder job than you may expect compared to old times. Because besides cells that are wearing and weared already, all NVME (and most likely SSD too) have such thing as "SSD spare cells"... When some cell goes down because of wearing, one of these spare cells comes instead to replace it. And number of these is actually not that low.
So to get actually broken SSD you need to wear it AND run out of these spare cells

Thanks for the info!

 

[ShangWang]

Any write action to that drive will 'suffer' if it is near full.
This includes the page file.

Percentages is irrelevant.
You can't just say "0-49% is OK, 51% is cause of death".

Just don;t fill it up past 75-80% or so, and don't stress.
Use it until it dies.

Ty
Edit: I'm not worried that my SSD will just suddenly lose life faster from being at 50% usage, but I'm just curious if theoretically MORE usage would cause more writes/less performance, but not by a relevant amount.
Say 50% vs 51%, there WILL be a difference, but the size of a speck of dust?