Question Sabrent Rocket Q PCI-E 3.0 x4 NVMe (QLC) (3200/2900MB/s) reviews / opinions?

[VirtualLarry]

They seem to be positioning these as performance / value leaders, being of similar pricing to the Intel 660p NVMe, but boasting way better read/write seq. specs.

Currently running 2x1TB 660p NVMe in RAID-0, would be interested in replacing with 1x 2TB Sabrent Rocket Q for $200.

 

[country2]

They seem to be positioning these as performance / value leaders, being of similar pricing to the Intel 660p NVMe, but boasting way better read/write seq. specs.

Currently running 2x1TB 660p NVMe in RAID-0, would be interested in replacing with 1x 2TB Sabrent Rocket Q for $200.

I'm running the 1TB Sabrant Rocket (SB-ROCKET-1TB) which is a TLC with no issues on a W10 build for about 8 months as a primary drive. They now have this Q version SB-RKTQ-1TB which is QLC and about to buy it also for another build as I have had no issues with this brand and most of the reviews are good.

 

[NewMaxx]

I've written about these on Reddit, but they're basically the new Rockets with QLC instead of TLC. The new Rockets use the E12S controller which is the E12 in a smaller form factor (and metal IHS to match). This design change was made to make more room on the PCB so they could have four NAND packages on one side, allowing for single-sided drives at higher capacities (for TLC). Unfortunately they also reduced the amount of DRAM on these drives, typically to 512MB (for 1/2 TB). In any case, the Rocket Q utilizes this same layout and basic hardware. My supposition based on their read/write specifications is that this uses the 96L Intel flash from the Intel 665p. Reasons for this are that it hits high write speeds at a relatively low capacity, which is more difficult with Toshiba's 96L QLC (which is denser) and Toshiba's 64L QLC is rarely seen. It could well be 64L Intel, but the speeds at 1TB - 8 dies - are more indicative of the faster (with same density) 96L QLC. It likely has the same SLC cache as the regular Rocket (new or old) which would be ~24GB of dynamic. However, it likely uses direct-to-QLC rather than folding outside of this state (unlike the 660p), so probably has faster base NAND writes. I made some predictions on the speed over at Reddit but ultimately put this drive in my "Moderate NVMe" category in my guides - that is, in-between Budget NVMe (660p) and Consumer NVMe (TLC Rocket).

 

[VirtualLarry]

but ultimately put this drive in my "Moderate NVMe" category in my guides - that is, in-between Budget NVMe (660p) and Consumer NVMe (TLC Rocket).

Without testing, that's about how I would estimate their performance as well, based on their components.

I'm running a pair of 660p 1TB NVMe SSDs in RAID-0, not for performance, but for capacity/$. Would like to eventually replace with a 2TB single drive, especially one with higher "spec" performance, like the Rocket Q 2TB.

 

[NewMaxx]

Without testing, that's about how I would estimate their performance as well, based on their components.

I'm running a pair of 660p 1TB NVMe SSDs in RAID-0, not for performance, but for capacity/$. Would like to eventually replace with a 2TB single drive, especially one with higher "spec" performance, like the Rocket Q 2TB.

The E12 is a more powerful controller. The 660p's SM2263 is similar to the SM2262/EN but with half the channels, but is still dual-core. The E12 is dual-CPU with co-processors, but these just offload I/O which is why those drives can push higher IOPS and generally do better when fuller. This is deceptive to some degree as having cores for host interaction, a la WD or Samsung, results in better workstation performance and efficiency.

Although SLC design also plays a role. The 660p has a dynamic gradient from 0% to 75% after which it is static, whilst the 665p grades from 50% to 75%, and the E12(S) is purely dynamic. Static SLC takes up some reserved/overprovisioned space but is inherently more robust because it doesn't convert to QLC and back. More importantly, the 660p relies on folding - incoming data must go through the SLC cache - which reduces performance but improves endurance as folding writes at 1/2 speed but sequentially (lower write amplification). My guess is the Rocket Q goes to direct-to-QLC which is faster, on top of that with Intel's 96L flash it would be faster yet. So it's probably more consistent with better full-drive performance, yet the cache is small enough (e.g. 24GB vs. 128GB dynamic at 1TB) not to be worrisome for endurance.

The last issue is DRAM. The 660p/665p has only 256MB, while the E12S should generally have 512MB. This isn't a huge deal as workloads that challenge the former's DRAM will be bottlenecked anyway by the flash and/or controller. But with the E12S and 96L QLC, plus a smaller SLC cache, it could act towards mitigating some of the QLC's downsides. So, on the whole, I place it in the Moderate category - however this is a preliminary finding as I have not seen the drive tested yet.