AnandTech Storage Bench - The Destroyer

The Destroyer is an extremely long test replicating the access patterns of very IO-intensive desktop usage. A detailed breakdown can be found in this article. Like real-world usage, the drives do get the occasional break that allows for some background garbage collection and flushing caches, but those idle times are limited to 25ms so that it doesn't take all week to run the test. These AnandTech Storage Bench (ATSB) tests do not involve running the actual applications that generated the workloads, so the scores are relatively insensitive to changes in CPU performance and RAM from our new testbed, but the jump to a newer version of Windows and the newer storage drivers can have an impact.

We quantify performance on this test by reporting the drive's average data throughput, the average latency of the I/O operations, and the total energy used by the drive over the course of the test.

ATSB - The Destroyer (Data Rate)

Samsung's dominance of this test wasn't being seriously challenged, but the 512GB 860 PRO does show improvement to the average data rate on The Destroyer, putting it up in the range of Samsung's multi-TB SATA drives. It's a small change, but SATA doesn't leave room for big gains.

ATSB - The Destroyer (Average Latency)ATSB - The Destroyer (99th Percentile Latency)

The good average and 99th percentile latency scores of the SanDisk Ultra 3D match or beat the best scores from the Samsung SATA drives. The 512GB 860 PRO shows substantial improvement in 99th percentile latency and more modest gains in average latency, relative to the 850 PRO.

ATSB - The Destroyer (Average Read Latency)ATSB - The Destroyer (Average Write Latency)

The Samsung 860 PROs show the best average read latencies in their respective product classes, but the SanDisk Ultra 3D isn't far behind. For average write latencies, the Ultra 3D takes a clear lead over the Samsung drives, and the Crucial BX300 is ahead of the Samsung drives by a hair.

ATSB - The Destroyer (99th Percentile Read Latency)ATSB - The Destroyer (99th Percentile Write Latency)

The 99th percentile read and write latencies of the 860 PRO show substantial improvements at 512GB, and smaller improvements among the multi-TB drives.

ATSB - The Destroyer (Power)

While the 4TB 860 PRO performed better on The Destroyer than the 512GB model by every measure, the 512GB model was more power efficient, and sets a new record for its class. The improvements relative to the 850 PRO are remarkable: the old 512GB 850 PRO required 60% more energy to complete The Destroyer than the new 512GB 860 PRO. Samsung has caught up with the modern competitors in terms of energy efficiency.

Introduction AnandTech Storage Bench - Heavy
Comments Locked

64 Comments

View All Comments

  • Lady Fitzgerald - Tuesday, March 13, 2018 - link

    DVDs are those coaster like things that have over 600 of my movies on them. I don't plan on getting rid of them anytime soon.
  • Lady Fitzgerald - Tuesday, March 13, 2018 - link

    Wow! What did Samsung ever do to you?

    A roughly 50% decrease in idle current usage is huge if one has very many of these in use (I currently have four 4TB 850 EVOs in my desktop machine but will probably replace them with five 4TB Pros later this year), even if one doesn't factor in the reduction in heat output that will likely occur. Even when in use, the 860 Pros will draw roughly 30% less current.
  • mapesdhs - Tuesday, December 29, 2020 - link

    Hello from the future! :) I was just curious, did you indeed in the end replace those 850 EVOs with 860 Pros?

    I was hunting for info on the 860 Pro (I bagged a 256GB on ebay a year ago), found this old review. I'm upgrading my daily desktop to a 2700X just now, was wading through my SSD pile to decide on the C-drive, use the 860 Pro I hadn't done anything with yet, or move stuff around and use an 850 Pro, or maybe a Vector (my existing system has had a Vector 256GB running for 7 years. :D)

    S'funny actually looking back at what the landscape used to be before the 850 era began:

    https://www.anandtech.com/show/6363/ocz-vector-rev...

    OCZ of course took some hefty brand image flak from earlier times due to dodgy early fw releases for its older Vertex models, which is a pity because its much later Vertex4 and Vectors were rather good, indeed the older Vertex2E/3 were fine with fixed fw (I used dozens of 2Es and 3s, still do for general testing/benching).

    I buy used 840 Pro units when I can, they're very good even today. Won a fair few 850 Pro 512GB/1TB units aswell. Was particularly pleased to nab four 1TB 850 Pros from a photo company which bought them for backup, because every drive had less than 50GB written. I look at modern QLC SATA products and it baffles me why anyone would buy them, I just hunt for used 850 drives, whether EVO or Pro, or an 840 Pro (I avoid the 840 EVO due to its data retention problems which thankfully never affected the 840 Pro). Sometimes I bid on an Extreme Pro aswell, they're still good.

    I have a lot of Samsung SSDs, but over time the focus of many comments here have proven true, pricing has become kinda crazy. Moving on to NVMe, I bought a few 950 Pro, 960 Pro and 970 EVO/Plus drives (the former two mostly via the used market), but after that the competition could no longer be ignored. My more recent 1TB/2TB NVMe purchases have all been Adata XPG SX8200 Pro, just 100 UKP for the 1TB model as I write this (vs. 170 for the 970 EVO Plus, or a completely ridiculous 289 UKP for the 970 Pro). I found the Adata to be faster, which in my case involves substantial sustained sequential writes (which naturally rules out all QLC models); it's also more power efficient. The Adata is TLC of course, but then so is the 970 EVO Plus, and the former actually has a higher write endurance rating (both have 5yr warranties).

    Samsung hasn't much moved their pricing though, so I guess despite the competition they're still able to sell the products they make, but I can't figure why anyone would buy a 970 Pro when it's almost three times the cost of the Adata or other models (they really are milking the perceived MLC advantage). I know the Sabrent Rocket is very popular, but so far I've not bought one as I've been unsure of the 4K block size issue, plus I've been able to find the Adata cheaper anyway.

    I just wish the capacities would properly get a move on. Seems to me vendors are not releasing anything better because they don't have to, people are still willing to pay solid premiums for existing 2TB/4TB models. It's all a far cry from Sandisk's old promise of an 8TB SATA3 SSD way back. I guess nobody wants to rock the boat; why release an 8TB NVMe when the market is happy to splurge on 4TB and below?

    I just think it's a shame how QLC has taken over, a race to the bottom via DRAMless designs, SATA and even NVMe models that tank once their SLC emulation phase is exceeded, in some cases giving performance slower than a rust spinner. It's bizarre to think that with modern benchmark suites an old SATA like a Vector, Vertex4, 840 Pro, Neutron GTX or Extreme Pro would actually be better in some cases, heck even the Samsung 830 would probably be quicker. Modern large capacity dies are killing performance by using so few memory channels. This could easily be resolved by allowing capacities to properly increase, but they just won't do it, not yet. They'll milk the 4TB for all its worth before considering 8TB. Makes me think the margins on modern SSDs must be very high vs. models from years ago, with the former using so few ICs on the PCB. Many modern SSD PCBs contain just a controller and one flash die.

    Btw, good comment below about halving the time for a particular task being less relevant if the duration is very short in absolute terms.
  • Lady Fitzgerald - Tuesday, March 13, 2018 - link

    Again, SATA is plenty fast for data storage. NVMe is an advantage only when used for the OS and programs; even then, it isn't all that much of an advantage. If you have a task that takes ten minutes to perform and you double the speed, that's taking it down to five minutes, which is huge. If you have a task that takes 10 seconds to perform and you double the speed, that will be now be five seconds, an improvement but nowhere nearly as as noticeable. If you have a task that takes 10 milliseconds to perform and you double the speed so it now takes only 5 milliseconds, you won't notice the difference. It won't be advantageous unless you move enormous amounts of data frequently.

Log in

Don't have an account? Sign up now