Crucial MX300 525GB and 1050GB SSD Review

Conclusion

With the Crucial MX300, you have to take the good with the bad. There are several points to consider, and the pro's and cons each have several items that give us a full picture of the overall drive.

Let's start at the SSD controller, which is the heart of the system. The controller sits between the SATA bus and the NAND. The Marvell 88SS1074 "Dean" utilizes four channels but also features Low-Density Parity Check (LDPC) to extend the life of the flash. This is the first time Crucial has used a 4-channel controller in a mainstream class MX Series SSD. Some of the older, smaller capacity drives only used four channels, but they still shipped with 8-channel controllers. 4-Channel controllers are cheaper to design, manufacture and use less power. They are also slower than true 8-channel controllers are and limit scale-out (addressing a large number of NAND die) and parallel IO.

When Micron announced the first generation 3D NAND, we learned that it doubled the density of MLC and tripled the density of TLC. At first, that sounded great! Until now, vendors based SSD pricing on how much space each die consumed on the wafer and the number of NAND die in the drive. When IMFT moved from 25nm to 20nm, the retail SSDs experienced a large price reduction. The same thing happened from 20nm to 16nm. Historically, you could go back, examine pricing data, and easily spot when a new lithography shrink came to market. If that were true with the new 3D NAND, the Limited Edition 750GB would cost about 80 Dollars rather than 185. IMFT is likely recouping some of the expensive R&D investments early on, and the current NAND flash shortage helps to keep prices high right now, too.

When it comes to 3D NAND's performance, the updated ONFI interface increases the flash's bandwidth through the interface back to the controller. At that critical link in the chain, Micron's new 3D is faster than any flash it had before. The increased ONFI bandwidth is the primary reason why Crucial was able to use a 4-channel controller in a mainstream SSD.

The faster interface speed is nice, but it still doesn't deliver the 500+ megabytes per second we get with these products. Micron's 3D NAND uses quad planes, and that means your data has four paths off the die instead of just one, but that still doesn't add up to 500+ MB/s. For that, you need massively parallel operations, which means reading and writing to a large number of die simultaneously. Due to the 3x increase in die density to 384Gb (48GB), the MX300 will not show us what Micron's 3D NAND is really capable of until we see the 2TB model next month.

Crucial used twenty-four die to hit 1152GB of raw NAND capacity. The 1050GB model we tested today looks really good considering the number of die used to reach the capacity point. In contrast, the Samsung 850 EVO 500GB uses thirty-two NAND die in tandem with an efficient 8-channel controller. The two SSDs deliver nearly identical performance outside of the low QD random read tests.

The MX300 525GB uses just twelve NAND die, but it compares well to the 850 EVO 500GB, too. The Dynamic Write Acceleration helps both MX300 SSDs stay competitive even when, on paper, they shouldn't. The MX300 750GB was a little rushed, and Crucial really should have held it until the firmware was as mature as it is now with the two new models we tested. The underlying technology in combination with the newer firmware makes the MX300 SSDs more attractive than we initially expected.

MORE: Best SSDs
MORE: Latest Storage News
MORE: Storage in the Forums

Follow us on Facebook, Google+, RSS, Twitter and YouTube.

Create a new thread in the Reviews comments forum about this subject
This thread is closed for comments
29 comments
Comment from the forums
    Your comment
  • hannibal
    Nice! Competition to 3d nand is just arriving! The Samsung did have monopoly too long time, now there is competition ones more.
  • alextheblue
    Drives like this don't appeal as much to the enthusiast directly, but they still benefit everyone. Performance is plenty good for most mainstream systems, and as cost per GB decreases they displace HDDs in more OEM systems. This is great because it makes it a lot easier to suggest an SSD model to budget-minded purchasers. We're still not quite where I want to be, but getting there. In the not-so-distant future I can envision building a system with a next-gen NV boot drive and 3D TLC storage drive(s), taking the place of my current SSD/HDD setup.

    Oh, and I'm really hoping the cable part of the PCIe 4.0 standard takes hold. I'd rather go that route than multiple M.2 drives.
  • FelixtheCat
    Thanks for the review - you have convinced me to keep my Samsung Evo 500GB!
  • dstarr3
    Added to my Christmas list.
  • bit_user
    Thanks for the benchies, Chris. I wish the SSD Charts would still be getting updated! (hint, hint)

    BTW, I always liked the end-to-end data protection features in Crucial's MX series. To my knowledge, they and Intel's 500-series drives are the only ones to offer this in the consumer segment. Is that (still) correct?
  • CaedenV
    Man, I take my eyes of the SSD segment for a little bit and the whole landscape changed! I bought all of the SSDs in my house back when they were ~$1/GB. For what I paid for my 500GB of SSD storage I could be looking at 2TB next month. That is just crazy.

    I guess the real question in my mind is when I rebuild my rig in ~2 years will I go for a 500GB-1TB performance m.2? Or a larger 2TB SATA drive? Or, if things keep droping in price like this will I be able to afford both?
    Or better yet, will these drop in price enough to start replacing HDDs in my NAS box? Hmm... decisions decisions.
  • bit_user
    Anonymous said:
    Man, I take my eyes of the SSD segment for a little bit and the whole landscape changed!
    ...

    I guess the real question in my mind is when I rebuild my rig in ~2 years will I ...
    The next 2 years will make the previous 2 years look static, by comparison. Expect $/GB to drop much further, and significant performance improvements. nvDIMMs will be on the scene, too.

    BTW, M.2 doesn't necessarily connote NVMe. Although, by the time you upgrade, it'll probably be the norm.
  • 10tacle
    I have the 500GB EVO 850 and am happy with it, but if I were looking for a new SSD, the 1050 would be on my short list for a 1TB drive. Absolute raw performance numbers are not be as important to me as having that extra 50GB which is an entire Steam game install these days.
  • c0rr0sive
    I can't wait for the day that I can replace all my 4TB and 8TB disks with some solid storage, right now I can either go with enterprise grade disks for $7800, or go the SSD route for about $9800. Soon, hopefully soon it will be cheaper to go SSD for my needs.
  • CRamseyer
    The MX300 2TB is coming out of testing tomorrow morning. It will be interesting to see it compares to other products shipping today.

    As always, thanks for reading and the comments.
  • clifftam
    I like how low cost shows up both for the pro and the con :).

    Can't wait for the 1 TB SSD to drop even more.
  • CaedenV
    Anonymous said:
    [The next 2 years will make the previous 2 years look static, by comparison. Expect $/GB to drop much further, and significant performance improvements. nvDIMMs will be on the scene, too.

    BTW, M.2 doesn't necessarily connote NVMe. Although, by the time you upgrade, it'll probably be the norm.

    Well, all of the M.2 I am considering are all NVMe ;)

    I would love to see nvDIMM tech like 3D x point, or HMC to come to the consumer space, but I am pretty sure it won't be coming in any useful way for another 5+ years. In the short term is it just a server tech. Once they get the price down I think we will see 32-64GB dimms for use in consumer PCs to replace both RAM and SSDs, but it will be a cheaper version that is only a bit faster than SSDs. To get this tech cheap enough for consumer use, and large enough to replace SSDs in the enthusiast market (250+GB) I think we will have a very long wait.
  • TJohn
    Good grief... talk about fast technical evolution. Now I have to start figuring out what to do with my existing SSDs. I have the need for speed and space. Free coasters? Again?
  • rmse17
    Where are all of the 3D NAND MLC drives? I see everyone paddling the TLC, but isn't only Samsung selling the 3D MLC in their PRO lineup?
  • CRamseyer
    Outside of a few corner cases, MLC for the consumer is dead. Several companies have said it behind closed doors. I'm not ready to buy a TLC drive other than an 850 EVO and I know many of you feel the same way.
  • kalmquist
    The low random read performance of the MX300 at queue depth 1 (compared to the 850 EVO) cannot be due to the number of channels or the number of dies. In that benchmark the SSD is only performing one read at a time. When the 750GB version of the MX300 was released, I was thinking that the difference was in the speed of the flash. But with the newer firmware tested in this review, the MX300 does quite well in some of the other benchmarks, so I'm now guessing that Micron's flash is as fast as Samsung's.

    If it's not the flash, the only other alternative I can see is controller overhead. The difference between 7620 IOPS (the MX300) and 10855 IOPS (the 850 EVO) is 39 microseconds per operation. I can believe that the Marvell controller is slower than Samsung's controller, but an additional 39 microseconds per read operation is a lot of time.
  • CRamseyer
    We will have a better look at the flash in a few days. We have the Intel 600p NVMe with IMFT TLC and the Samsung PM961 also with NVMe and TLC.
  • bit_user
    Anonymous said:
    The low random read performance of the MX300 at queue depth 1 (compared to the 850 EVO) cannot be due to the number of channels or the number of dies. In that benchmark the SSD is only performing one read at a time.
    For all the benchmarks included in articles like this, random read perf at low queue depths is the main stat that users actually feel. For everything else, OS-level prefetching, caching, and write buffering do a good job of hiding the performance of the underlying storage from end users.

    Obviously, sequential read/write matters, but is often interface-constrained.
  • max0x7ba
    Sequential read performance chart 2 (the bar chart) is an example of misleading statistics: the top bar is more than twice as long as the bottom bar, although the real difference between them is around 15%,
  • bit_user
    Anonymous said:
    Sequential read performance chart 2 (the bar chart) is an example of misleading statistics: the top bar is more than twice as long as the bottom bar, although the real difference between them is around 15%,
    Yeah, I noticed that. Well, the axis is labelled, so it's not that bad.

    I think they were trying to illuminate the differences - not skew the article one way or another. Because all of the drives are near the interface speed, they're all pretty similar in this benchmark.