[SOLVED] PNY CS3030 NVMe SSD slow write-speed ?

Jan 21, 2022
6
0
10
Hi,

I have a PNY CS3030 NVME SSD for 6 months now and I just realized that the write speed does not match the specs.

I tried with default NVME conf in the BIOS as well as GEN 3 configuration as the CS3030 is a GEN 3 x4 SSD.

cs3030.png


More infos : I have multiples SSD 2.5 and HDD on the sata ports. Can it be a shared bandwith between NVMe and SATA disks ? Also, the NVMe is on M2_1 which is a Gen4 x4 PCIe port.

PC Specs :
  • CPU : AMD Ryzen 5900X
  • MB : Asus ROG STRIX B550-A GAMING
  • RAM : Crucial Ballistix, 3600 MHz, DDR4, 2x8Go
Do you have any ideas guys ?

Thx !
 
Last edited:
Hi,

I have a PNY CS3030 NVME SSD for 6 months now and I just realized that the write speed does not match the specs.

I tried with default NVME conf in the BIOS as well as GEN 3 configuration as the CS3030 is a GEN 3 x4 SSD.

cs3030.png


More infos : I have multiples SSD 2.5 and HDD on the sata ports. Can it be a shared bandwith between NVMe and SATA disks ? Also, the NVMe is on M2_1 which is a Gen4 x4 PCIe port.

PC Specs :
  • CPU : AMD Ryzen 5900X
  • MB : Asus ROG STRIX B550-A GAMING
  • RAM : Crucial Ballistix, 3600 MHz, DDR4, 2x8Go
Do you have any ideas guys ?

Thx !
WAG....is write caching enabled for the disk?

As a test I suppose you could unplug all sata disk to see if it had any effect.
 
Jan 21, 2022
6
0
10
I tried disconnecting all drives, update the bios, nothing works. I still have only 1Go/s while writing... Others hints ?
For info, I'm on Windows 11 and it's up to date. I know there has been some issues with NVMe wrinting speed that should have been fixed with the KB5007262 update. Maybe that's not enough yet.
 
You're hitting TLC rather than SLC.

The CS3030 is/was a typical E12(S) drive which tends to have ~24GB of dynamic SLC. This is fairly small and as such tends to feel more like a static cache in terms of consistency, although it will still shrink as the drive fills. At 85% usage, for example, it's probably closer to half that maximum size. As such it's fairly easy to fill it and hit direct-to-TLC speeds (~1 GB/s for 1TB E12 drives). For example, look at the results for the E12-based MP510 here - you can see where the red line is hitting TLC at around the same speed.

Dynamic SLC is common on consumer SLCs these days and this issue is not uncommon. The 980 PRO, for example, notoriously had a similar issue with benchmarks for months after its launch. The reason varies but in general there can be reasons to maintain data in SLC - reads are better, but it also reduces the potential for flash wear via deferred writes. In general SLC is a write cache and is emptied quickly when the drive is idle, though.

It's possible to fix this with a secure erase or equivalent (e.g. format + idle time). You should make sure the firmware is up-to-date also, keeping in mind an update may be destructive (erases data).
 
Jan 21, 2022
6
0
10
You're hitting TLC rather than SLC.

The CS3030 is/was a typical E12(S) drive which tends to have ~24GB of dynamic SLC. This is fairly small and as such tends to feel more like a static cache in terms of consistency, although it will still shrink as the drive fills. At 85% usage, for example, it's probably closer to half that maximum size. As such it's fairly easy to fill it and hit direct-to-TLC speeds (~1 GB/s for 1TB E12 drives). For example, look at the results for the E12-based MP510 here - you can see where the red line is hitting TLC at around the same speed.

Dynamic SLC is common on consumer SLCs these days and this issue is not uncommon. The 980 PRO, for example, notoriously had a similar issue with benchmarks for months after its launch. The reason varies but in general there can be reasons to maintain data in SLC - reads are better, but it also reduces the potential for flash wear via deferred writes. In general SLC is a write cache and is emptied quickly when the drive is idle, though.

It's possible to fix this with a secure erase or equivalent (e.g. format + idle time). You should make sure the firmware is up-to-date also, keeping in mind an update may be destructive (erases data).

I understand that the TLC access time is 3 times longer than the SLC access time because on each cell 3 data are stored. You wrote that the CS3030 has 24GB of SLC, however, it appers to be a TLC drive. Altough it is a TLC drive, the writing speed announced by PNY should be hit regardless of the cell technology used, am i right ?

I can try to wipe some games on this drive and see if it's better.

Also, how a secure erase (what's a "secure" erase btw?) will improve performances ? It has something to do with the 24GB of dynamic SLC which will be free to use ?
 
Last edited:
I understand that the TLC access time is 3 times longer than the SLC access time because on each cell 3 data are stored. You wrote that the CS3030 has 24GB of SLC, however, it appers to be a TLC drive. Altough it is a TLC drive, the writing speed announced by PNY should be hit regardless of the cell technology used, am i right ?

I can try to wipe some games on this drive and see if it's better.

Also, how a secure erase (what's a "secure" erase btw?) will improve performances ? It has something to do with the 24GB of dynamic SLC which will be free to use ?

Not exactly 3 times. TLC tR might be 50µs versus 25µs in pSLC while TLC tPROG averages 500µs versus 200µs for pSLC, but yeah there is a significant difference. pSLC being pseudo-SLC which is the TLC acting in SLC (single-bit) mode. Modern consumer drives trade temporary capacity for performance.

Drives tend to be marketed with their SLC write performance but native/TLC read (since SLC is a write cache and most reads come from native). TLC writes are slower. There's many reasons to retain data in SLC, though.

Secure erase wipes the mapping table and over some idle time the drive will erase the contents, too. This "resets" the counters which impact maintenance mode, which can help if the drive is caught in direct-to-TLC (which occurs when SLC is full or the drive feels it's more desirable to write to TLC as with sequential writes, the maintenance mode is triggered by free block thresholds, technical stuff).