The Intel SSD 660p SSD Review: QLC NAND Arrives For Consumer SSDs
by Billy Tallis on August 7, 2018 11:00 AM ESTAnandTech Storage Bench - Heavy
Our Heavy storage benchmark is proportionally more write-heavy than The Destroyer, but much shorter overall. The total writes in the Heavy test aren't enough to fill the drive, so performance never drops down to steady state. This test is far more representative of a power user's day to day usage, and is heavily influenced by the drive's peak performance. The Heavy workload test details can be found here. This test is run twice, once on a freshly erased drive and once after filling the drive with sequential writes.
When the Heavy test is run on an empty Intel SSD 660p, the test is able to operate almost entirely within the large SLC cache and the average data rate is competitive with many high-end NVMe SSDs. When the drive is full and the SLC cache is small, the low performance of the QLC NAND shows through with an average data rate that is slower than the 600p or Crucial MX500, but still far faster than a mechanical hard drive.
The average and 99th percentile latency scores of the 660p on the empty-drive test run are clearly high-end; the use of a four-channel controller doesn't seem to be holding back the performance of the SLC cache. The full-drive latency scores are an order of magnitude higher and worse than other SSDs of comparable capacity, but not worse than some of the slowest low-capacity TLC drives we've tested.
The average read latency of the Intel 660p on the Heavy test is about 2.5x higher for the full-drive test run than when the test is run on a freshly-erased drive. Neither score is unprecedented for a NVMe drive, and it's not quite the largest disparity we've seen between full and empty performance. The average write latency is where the 660p suffers most from being full, with latency that's about 60% higher than the already-slow 600p.
The 99th percentile read latency scores from the 660p are fine for a low-end NVMe drive, and close to high-end for the empty-drive test run that is mostly using the SLC cache. The 99th percentile write latency is similarly great when using the SLC cache, but almost 20 times worse when the drive is full. This is pretty bad in comparison to other current-generation NVMe drives or mainstream SATA drives, but is actually slightly better than the Intel 600p's best case for 99th percentile write latency.
The Intel SSD 660p shows above average power efficiency on the Heavy test, by NVMe standards. Even the full-drive test run energy usage is lower than several high-end drives.
86 Comments
View All Comments
StrangerGuy - Tuesday, August 7, 2018 - link
"I am a TRUE PROFESSIONAL who can't pay more endurance for my EXTREME SSD WORKLOADS by either from my employer or by myself, I'm the poor 0.01% who is being oppressed by QLC!"Oxford Guy - Tuesday, August 7, 2018 - link
Memes didn't make the IBM Deathstar drives fun and games.StrangerGuy - Tuesday, August 7, 2018 - link
I'm sure you were the true prophetic one warning us about those crappy those 75GXPs before they were released, oh wait.I'm sorry why are you here and why should anyone listen to you again?
Oxford Guy - Tuesday, August 7, 2018 - link
Memes and trolling may be entertaining but this isn't really the place for it.jjj - Tuesday, August 7, 2018 - link
Not bad, at least for now when there are no QLC competitors.The pressure QLC will put on HDDs is gonna be interesting too.
damianrobertjones - Tuesday, August 7, 2018 - link
These drives will fill the bottom end... allowing the mid and high tiers to increase in price. Usual.Valantar - Wednesday, August 8, 2018 - link
Only if the performance difference is large enough to make them worth it - which it isn't, at least in this case. While the advent of TLC did push MLC prices up (mainly due to reduced production and sales volume), it seems unlikely for the same to happen here, as these drives aim for a market segment that has so far been largely unoccupied. (It's also worth mentioning here that silicon prices have been rising for quite a while, and also affects this.) There are a few TLC drives in the same segment, but those are also quite bad. This, on the other hand, competes with faster drives unless you fill it or the SLC cache. In other words, higher-end drives will have to either aim for customers with heavier workloads (which might imply higher prices, but would also mean optimizations for non-consumer usage scenarios) or push prices lower to compete.romrunning - Wednesday, August 8, 2018 - link
Well, QLC will slowly push out TLC, which was already pushing out MLC. It's not just pushing the prices of MLC/TLC up, mfgs are slowing phasing those lines out entirely. So even if I want a specific type, I may not be able to purchase it in consumerspace (maybe enterprise, with the resultant price hit).I hate that we're getting lower-performing items for the cheaper price - I'd rather get higher-performing at cheaper prices! :)
rpg1966 - Tuesday, August 7, 2018 - link
"In the past year, the deployment of 64-layer 3D NAND flash has allowed almost all of the SSD industry to adopt three bit per cell TLC flash"What does this mean? n-layer NAND isn't a requirement for TLC is it?
Ryan Smith - Tuesday, August 7, 2018 - link
3D NAND is not a requirement for TLC. However most of the 32/48 layer processes weren't very good, resulting in poorly performing TLC NAND. The 64 layer stuff has turned out much better, finally making TLC viable from all manufacturers.