News Western Digital claims its new QLC SSD beats its last-gen TLC drive — SN5000S is up to 16.5% faster than SN740

[Admin]

Western Digital's latest QLC-based SN5000S drive is apparently up to 16.5% faster than its TLC-based predecessor, the SN740.

Western Digital claims its new QLC SSD beats its last-gen TLC drive — SN5000S is up to 16.5% faster than SN740 : Read more

 

[GabrielFerrazTPU]

One correction, the SSDs doesn't uses Kioxia 96-layer QLC but instead its newest variants of BiCS6 1Tb QLC dies
https://www.techpowerup.com/ssd-specs/#WD SN5000S

 

[cyrusfox]

What about sustained performance, not the boosty SLC cache, as sustained performance is the main limiter for what I use these drives for(media recording at high data rates). I sure don't care about getting an extra boost for burst loads, for me the main characteristics of importance are sustained performance and efficiency(also related to sustained performance, overheating leads to throttling).

Based of the spec showing higher power consumption at load (6.9w vs 6.3w) and the knowledge that sustained performance always drops off with QLC vs TLC, I will keep buying SN740 until someone test these drives out and show me they can keep up on sustained writes. Would love to see power test, but hardly anyone properly puts SSD through their paces like they did 7+ years ago.

 

[DaveLTX]

What about sustained performance, not the boosty SLC cache, as sustained performance is the main limiter for what I use these drives for(media recording at high data rates). I sure don't care about getting an extra boost for burst loads, for me the main characteristics of importance are sustained performance and efficiency(also related to sustained performance, overheating leads to throttling).

Based of the spec showing higher power consumption at load (6.9w vs 6.3w) and the knowledge that sustained performance always drops off with QLC vs TLC, I will keep buying SN740 until someone test these drives out and show me they can keep up on sustained writes. Would love to see power test, but hardly anyone properly puts SSD through their paces like they did 7+ years ago.

Definitely worse. There's no example of a QLC drive in sustained being faster than a TLC

 

[cyrusfox]

Definitely worse. There's no example of a QLC drive in sustained being faster than a TLC

Have you seen the SN740 review... There is some flashes of brilliance from the Sabrent Rocket Q4 2tb (QLC drive) where it is besting the majority of the TLC drives, only Inland TN446 1TB appears to keep up (MP600 CORE MINI 1TB, SN740 and SN770M are the other TLC drives to compare to).

With some more controller magic as well as onpackage DRAM or software magic (System DRAM allocation),The next generation drives may only be QLC and likely superior to the current TLC best results. The ultimate limiter for QLC nand chips is the native write to QLC chips once pseudo QLC is exhausted, but with all of these tricks, that threshold can be pushed far out and give more opportunities for cache recovery.

I do find it interesting that while everyone else is climbing the layer count, WD seems to be regressing here(Others are at like 176+ layers WD in these products is going from 112 layers to 96 layers). Are they getting rid of old stock? Or did Kioxia and WD take a different path to scaling 3D Nand, perplexed... Would love to understand their positioning going forward to compete with Hynix, Samsung, and Micron.

 

[DaveLTX]

Have you seen the SN740 review... There is some flashes of brilliance from the Sabrent Rocket Q4 2tb (QLC drive) where it is besting the majority of the TLC drives, only Inland TN446 1TB appears to keep up (MP600 CORE MINI 1TB, SN740 and SN770M are the other TLC drives to compare to).

With some more controller magic as well as onpackage DRAM or software magic (System DRAM allocation),The next generation drives may only be QLC and likely superior to the current TLC best results. The ultimate limiter for QLC nand chips is the native write to QLC chips once pseudo QLC is exhausted, but with all of these tricks, that threshold can be pushed far out and give more opportunities for cache recovery.

I do find it interesting that while everyone else is climbing the layer count, WD seems to be regressing here(Others are at like 176+ layers WD in these products is going from 112 layers to 96 layers). Are they getting rid of old stock? Or did Kioxia and WD take a different path to scaling 3D Nand, perplexed... Would love to understand their positioning going forward to compete with Hynix, Samsung, and Micron.

Moments of brilliance doesn't equate to being better.
However, it appears they "regress" in layer count possibly because higher layer counts are possobly too dense to make multiple channels of QLC unless it's a very high cap SSD. Without multiple channels QLC is devastatingly slow.
And also possibly because higher layer count QLC can be more unreliable.
At the very least they're not pulling a solidigm and turning a PLC into a QLC. (D5 P5336)
Or QLC nand onto pSLC (D7 P5810)
pSLC isn't close to being SCM! Real SCM have short word lengths and consequently much lower access time than that trash called P5810