Question Silicon Power 1TB NVMe (SP001TBP34A80M28) Write Speed Issue

adikumar2010

Distinguished
Apr 17, 2009
62
0
18,530
I ran some benchmarks and I am getting very low write Seq. Q32T1 speed from few weeks in crystal disk mark. Originally Seq. Q32T1 write speed was always above 3000mbps when I bought it last year which is the advertised speed by company.

I am not sure what could be the issue. Maybe the SSD is fried ? Does anyone know?

Original Benchmark from last year:
old.png



Recent Bechmark :
New.png


-----------------------------------------------------------------------------------------------------------------------------

I could send the SSD for RMA but problem is I don't want to reinstall everything. I have 2 external HDD lying around. So I can clone full disk image of my SSD into a USB HDD then when I get it back I can restore image on it. Can anyone let me know good software to do that.

Also I think first I need to create image of SSD onto an external HDD and then when I get it back I will have to install Windows on another spare HDD so I can restore that image back to the replaced SSD. It's very long and tedious process. Let me know if there is any easier method. I am not sure if its possible to restore the cloned image without installing any OS.
 
The rated write speed is only for the SLC cache, which on that drive is up to 24GB. This cache will shrink as the drive is used down to a minimum amount, probably around 12GB. This isn't actual SLC, it's just each 3-bit TLC cell acting in 1-bit SLC mode. Outside of this cache the drive is using the actual TLC which is far slower than SLC. At 1TB for a drive with that flash, this is approximately 1050 MB/s, as you see in your CDM results.

You can even calculate this to some degree if you're so inclined - at 1TB you have 32x32GiB flash dies, each two-plane, for a total of 64-way interleaving possible. Given the standard page size (16KiB or .015625MiB) and program latency in TLC mode (933µs), this comes out to around 1072 MB/s - bit lower than your result.
 
  • Like
Reactions: adikumar2010

adikumar2010

Distinguished
Apr 17, 2009
62
0
18,530
The rated write speed is only for the SLC cache, which on that drive is up to 24GB. This cache will shrink as the drive is used down to a minimum amount, probably around 12GB. This isn't actual SLC, it's just each 3-bit TLC cell acting in 1-bit SLC mode. Outside of this cache the drive is using the actual TLC which is far slower than SLC. At 1TB for a drive with that flash, this is approximately 1050 MB/s, as you see in your CDM results.

You can even calculate this to some degree if you're so inclined - at 1TB you have 32x32GiB flash dies, each two-plane, for a total of 64-way interleaving possible. Given the standard page size (16KiB or .015625MiB) and program latency in TLC mode (933µs), this comes out to around 1072 MB/s - bit lower than your result.

Thanks for explaining. So no point getting it RMA'd as that replaced new drive will also become slow overtime and this slow speed is normal behavior & nothing can be done to fix it ?

Also someone suggested I can fix it by running a manual garbage collection. I have no idea what that is.
 
Go into Windows Defrag and optimize the SSD. Alternatively you can use PowerShell to do it:

Optimize-Volume -DriveLetter C -ReTrim -Verbose

The drive should "recover" to SLC mode automatically and quickly but there may be reasons it's writing to TLC. Perhaps something has been writing in the background, perhaps it thinks it's wise to maintain data in SLC, maybe it thinks TLC is better from a power-usage perspective (efficiency), etc. SLC algorithms are getting quite complex. You should be able to get it back to SLC though.