Random Read Performance

Our first test of random read performance uses very short bursts of operations issued one at a time with no queuing. The drives are given enough idle time between bursts to yield an overall duty cycle of 20%, so thermal throttling is impossible. Each burst consists of a total of 32MB of 4kB random reads, from a 16GB span of the disk. The total data read is 1GB.

Burst 4kB Random Read (Queue Depth 1)

The Optane SSD 900P doesn't break the record for QD1 random reads, but only because we've also tested the 32GB Optane Memory M.2, which is about two microseconds faster on average for each 4kB read. The Optane SSD 900P is still about 7 times faster than any flash-based SSD.

Our sustained random read performance is similar to the random read test from our 2015 test suite: queue depths from 1 to 32 are tested, and the average performance and power efficiency across QD1, QD2 and QD4 are reported as the primary scores. Each queue depth is tested for one minute or 32GB of data transferred, whichever is shorter. After each queue depth is tested, the drive is given up to one minute to cool off so that the higher queue depths are unlikely to be affected by accumulated heat build-up. The individual read operations are again 4kB, and cover a 64GB span of the drive.

Sustained 4kB Random Read

When longer transfers and higher queue depths come into play, the Optane SSD 900P passes the Optane Memory M.2 and remains more than 6 times faster for random reads than any flash-based SSD.

Both Optane devices more or less level off at queue depths of 8 or higher. The Optane SSD 900P saturates at about 1800 MB/s while the Optane Memory tops out around 1300 MB/s. The Samsung 960 PRO 2TB hasn't caught up by QD32, and doesn't surpass the QD1 random read performance of the Optane SSD until the Samsung reaches a queue depth of about 8.

Random Write Performance

Our test of random write burst performance is structured similarly to the random read burst test, but each burst is only 4MB and the total test length is 128MB. The 4kB random write operations are distributed over a 16GB span of the drive, and the operations are issued one at a time with no queuing.

Burst 4kB Random Write (Queue Depth 1)

The burst random write performance of the Optane SSD 900P is slightly higher than the Intel SSD 750 1.2TB, and about 14% faster than Samsung's fastest.

As with the sustained random read test, our sustained 4kB random write test runs for up to one minute or 32GB per queue depth, covering a 64GB span of the drive and giving the drive up to 1 minute of idle time between queue depths to allow for write caches to be flushed and for the drive to cool down.

Sustained 4kB Random Write

With higher queue depths in play, the Optane SSD 900P scales up faster than the Intel SSD 750 1.2TB, leaving the Optane SSD with a 7-10% lead over the Samsung 960s and Intel 750.

Samsung's 960 PROs and the larger 960 EVO all trail slightly behind the Optane SSD's random write performance for queue depths 1 to 4, then the Samsung drives level off and leave the Optane SSD with a substantial performance advantage at high queue depths. The Intel 750 is slightly faster at QD1 and QD2, but saturates at an even lower performance level than the Samsung 960s.

AnandTech Storage Bench - Light Sequential Performance


View All Comments

  • phaethon1 - Tuesday, November 14, 2017 - link

    Nice post,

    I read in multiple channels about this SSD being able to be used as extra RAM. Then I contacted the technical support of Intel, and they do not have any clue about a software to enable this feature. Any ideas?
  • extide - Friday, October 27, 2017 - link

    SLC might be faster in sequential, but if you want sequential stripe a bunch of platters ..

    Also, I didn't say bit-level, I said block level. They present 512b blocks so you would assume the drive manages 'pages' in the size of 512b even though the underlying memory can be more fine-grained. SLC can't do that, plus there is still the whole garbage collection thing. If your hypothetical drive was actually a good idea, someone would make it. That's proof enough that it's not.
  • ddriver - Friday, October 27, 2017 - link

    They didn't make one not because it is not good enough, but because it would be too good.

    That would set a bad precedent. Before you know it, people will start demanding quality rather than being content with what the industry dictates to them.

    Of course, if hypetane manages to make enough a hole in the pockets of big players, we will definitely be seeing some of that long-possible, deliberately untapped potential coming to life.

    "That's proof enough that it's not."

    You know, they make trucks that drive 24/7, under huge loads, and can go much long without maintenance than a regular personal vehicle. That's proof enough that the industry doesn't make things as good as it can, obviously, if it can make a heavily used and loaded truck more durable, that would be not only possible, but actually much easier to achieve for a regular car that's driven less, under less load. Yet they don't make it, even if that ends up costing human lives. And the reason for that is moar profit. Which is why they chose to only overbuild trucks, because that too maximizes profits. But not cars. Cars are far more profitable if need more servicing, and that doesn't result in profit losses as it would if it was commercial trucks, and if underbuilt cars end up costing human lives, that's a small price to pay for more profit. Engineering wise, is entirely possible and easily doable to make a car about 10 times more durable, and requiring 10 times less maintenance, and 10 times safer too, but they'd rather get the extra profit. And keep good engineering exclusive to military and commercial production.

    The reason they haven't made it is they didn't have a reason to make it. And the reason intel did hypetane is only because it has been a very long time since they did anything new. They had that in the works, and decided to release it in order to demonstrate some innovation, unfortunately, not without shamelessly lying about how well it will perform in advance.
  • Xpl1c1t - Friday, October 27, 2017 - link

    ddriver, i like your analysis. maybe the review system just wasnt equipped with rgb lighting, that would explain at least one order of magnitude of error in their results vs Intel's promises Reply
  • jospoortvliet - Friday, November 3, 2017 - link

    > Engineering wise, is entirely possible and easily doable to make a car about 10 times more durable, and requiring 10 times less maintenance, and 10 times safer too, but they'd rather get the extra profit. And keep good engineering exclusive to military and commercial production.

    Well, yes, they care about their profits: nobody would buy such a super-durable car because it would cost 5-10x the price and people will go for the cheaper car, even though it has higher maintenance cost. This is true for nearly ANY product on the market: sure, you could built houses more durable, or bikes, or... you name it. But people prefer 'good enough' over 'perfect', always have. And they're not entirely stupid - many products' practical life time is fine, people quite like buying a new car every 3-5 years. Or new cups. Or new forks and knives.

    Yes, some folks pay the 10x price to get the perfect, durable stuff. But most buy pressed wood closets at Ikea and are happy with it.
  • Gastec - Saturday, March 10, 2018 - link

    I'm not sure if you are both ironic or are just too rich to think straight. Reply
  • AlishaScott - Sunday, October 29, 2017 - link

    I just got paid $6784 working off my laptop this month. And if you think that’s cool, my divorced friend has twin toddlers and made over $9k her first month. It feels so good making so much money when other people have to work for so much less. This is what I do... http://cutt.us/O5gex Reply
  • Nails6365 - Monday, November 6, 2017 - link

    Thank you for your in-depth analysis.

    Given the opportunity to make a high-end rig. What would you choose ?
  • Jared13000 - Tuesday, May 22, 2018 - link

    You’re not giving Optane enough credit, you don't necessarily compare a NAND based drive to an Optane based drive. Compare NAND to Optane, as NAND has had years of development pored into hiding its short comings that Optane has not yet had.

    I just built a small all flash hyper converged cluster and after setup I was getting about 500,000 random read IOPS on a quad node cluster with triple mirrored storage. Write speeds were about 1,000 IOPS, basically hard drive speeds across the 16 SSDs in the cluster.

    Was it bad drivers, miss configuration, ethernet flow control issues?

    None of the above. It was the drive cache. Storage spaces disabled it due to the drives not having power loss protection. Enabled the cache on all the drives to avoid direct NAND writes and now the cluster can push nearly 280,000 write IOPS. This mean with cache the drives are over 200 times faster than just writing directly to NAND.

    What does this have to do with Optane? As far as I have been able to find, Optane drives don't have or need a cache. Their performance is direct to storage, without cache!

    Taken in the context of NAND vs Optane, 1,000x may be embellished, but probably not by much. At this point PCIe overhead and lack of software optimization may be the only reason it’s not 1,000x faster when comparing modern NAND memory.

    It's not that much faster comparing a whole NAND drive with well implemented cache to an Optane drive, but some situations can't rely on cache. Also, a simpler drive should be more reliable, in theory.

    As it is Optane is unrivaled until someone manages to bring a drive to market with SLC NAND and nonvolatile cache like MRAM for about twice the cost of a 970 PRO.

    Just a thought, a 970 PRO 512 GB has an MSRP of $329 and the Intel 900p 280 GB has it’s MSRP at $329 as well. That is 256 GB of SLC flash vs 280 GB of Optane. Comparing an MLC drive to an SLC drive at half the capacity is a bit like comparing apples and oranges, but it’s a start for an estimate. Trade the DRAM for MRAM and bump the capacity a bit and it’s hard to imagine that a SLC NAND based drive with MRAM wouldn’t cost more than Optane.

    If you expected 1,000 times SSDs that are getting multiple gigabytes per second reads and writes, then you were looking for a drive faster than CPU cache. Intel really needs to watch their wording, but that does not make this a bad product.
  • CheapSushi - Friday, October 27, 2017 - link

    Well, then wait for Samsung's Z-NAND, which is MLC/TLC NAND treated like SLC. Reply

Log in

Don't have an account? Sign up now