A couple weeks back I unveiled the contenders in our planned battle for portable flash supremacy. Things have changed a bit since that simpler time, though, and my personal list of priorities has been, uh, trimmed.
This review was always going to be a longer term project. I’m aspiring to put my own spin on a mini version of Geoff’s iconic SSD Endurance Experiment. Albeit, one with a degree of portable wear and tear involved.
Since I’m off to a delayed start, and because I expect it’s going to be a long wait for more data, I want to share what I’ve got so far. Without further ado, let’s take a look at the setup.
The hub is an ORICO 10-port USB 3.0 model. Here’s a rundown of what you can see in that image, sorted left to right in order of the USB port they are plugged into.
- Intel Optane Memory M10 16 GB & ORICO Transparent NVMe M.2 USB 3.1 Type-C Gen2 enclosure (plugged directly into my PC)
- Kingston Digital 60 GB mS200 mSATA SSD & ORICO external mSATA UASP SSD enclosure
- Samsung 840 EVO 120GB SSD & StarTech SATA to USB 3.1 UASP hard drive adapter
- Patriot 64 GB Supersonic Rage Series USB 3.0 flash drive
- PNY Turbo 64 GB USB 3.0 flash drive
- SanDisk Ultra CZ48 64 GB USB 3.0 flash drive
- Corsair Flash Voyager Slider X1 64 GB USB 3.0 flash drive
- Kingston Digital 64 GB 100 G3 USB 3.0 DataTraveler
All of that plugged into my PC made for an interesting Disk Management window. After all, I’m usually a single drive kind of guy (not counting OneDrive or network drives). Before I gathered my initial numbers, I made sure all the drives had just a single NTFS partition on them that used all the space on the drive. I also did a full format on them before testing, why not?
For my first test with these new recruits I went back to some benchmarking comfort food, CrystalDiskMark 6. I intend to employ TR’s own RoboBench or something similar for longevity testing, though. I used my main rig to obtain these numbers, it runs Windows 10 Professional with all the latest updates for hardware and software. I’m not going to spell out the exact specifications in this piece, but you can check them out over here if you’d like. I set DiskMark to do nine passes at 500MiB for these tests, the same settings I’ve used in my other reviews.
Let’s look at some numbers (note the change to the x-axis as the tests change gears).
If you’re already thinking that it sure would be nice if Fish would add a cheap M.2 NVMe SSD into this mix, you’re not alone. I’m definitely fighting the review-feature-creep urge. We’ll see what happens. Anyway, I think the most interesting result here is the massive gap between the Optane’s read and write performance. I’m inclined to blame the controller in the enclosure, but without testing another… Gosh darn it, I’m going to have to buy another drive aren’t I?
Beyond that, it’s easy to see that if you’ve got an old SSD laying around you’ll get much better bang for your buck if you buy a $15 enclosure for it than buying a $15 flash drive. Speaking of the flash drives, the Patriot and the Corsair are the clear leaders in sequential performance, almost catching the mSATA drive in read speed and coming close to half its write speed.
I love the random 4K tests as a gut check for worst-case performance. With a queue depth of eight and eight threads to work with, there’s a massive throughput hit to everything—except for the Optane’s write speed. Hmm… Interestingly, one of the flash drives, the SanDisk Ultra, distinguishes itself with terrible but not tragic write performance. Frankly, I’m surprised the other drives even finished the benchmark.
As the number on the end of our x-axis continues to drop lower we can easily see why. At a queue depth of 32 and just one thread, the write speed on the Optane is still hitting the same wall but performance everywhere else is tanking. That said, even the over five-year old Kingston mSATA drive is whooping the flash drives from inside is aluminum UASP to-go box. The SanDisk Ultra’s meager 1.5 MB/s write speed is still remarkable among its peers, though.
With a queue depth of just one and a single thread to match, it doesn’t get much more worst-case than this—for performance anyway. Yeah, it’s not the most realistic test for a flash drive, but if a drive can handle this, you should be able to count on it handling whatever fiddly batch of files you want to move around with it. Notable developments are the Optane’s write speed finally stumbling and the SanDisk Ultra still cruising along with its mysteriously superior write speed. Finally, we have write speeds of the full-blown SSDs all surpassing the read speeds—caching at work.
That’s all I’ve got to share for now. If you must have a conclusion at this stage, mine would be that I like the plucky SanDisk Ultra the best of the flash drives. However, you’d be better served by digging an old mSATA drive out of your junk drawer and tossing it into a USAP-friendly housing. I’ll be back again once I’ve sufficiently abused these poor drives.
Share
Question & Answers (37)
Leave a Comment
ASM1351 is the only one I’ve been actually able to physically confirm mate, but Hardware IDs for a couple of others that play up match that one (I broke one down to sight the chip itself, but have no other fully-malfunctioning ones to compare it with).
That said, I did a bit more mucking around last night and discovered a couple more points of purportedly pertinent info:
1. The Fresco Logic USB 3 controller (or its driver) on my laptop may be skewing results. I ran SDI last week to do a sweeping update of drivers and there was an updated Fresco driver, so I went ahead. I ran a 20GB transfer last night via a BOT enclosure, then a UASP enclosure – both finished. I often leave the laptop on overnight to bulk upload to the NAS rather than the desktop – this may mean that while I’ve never experienced any BOT-related trouble via the laptop, it wasn’t running a UASP-ready driver. Who knows.
2. Alpine Ridge gave me issues, too. I’ve since swapped boards out where TB support isn’t present, but I couldn’t use the Type C port on my old board for anything other than BOT transfers – they’d crap out too.
In the overall scheme of things I don’t transfer significant amounts of data, so overnight or 48-hour transfers on the odd occasion don’t bother me. It might be valuable feedback for those considering UASP for large, regular goes though – if you see weird disconnects, you’re not alone.
I’d be curious to know what chipset those randomly disconnecting enclosures use. If that’s something you’ve kept track of.
The preview of a portable nand grudge match is a thing you must concern. If you are using a portable USB drive then you can go through this. The main thing is that if you are using a USB drive then there are many secure things you have to use for the USB drive. So I believe that if you know how to secure it then it’s easy to secure all information.
That’s valuable feedback mate, thankyou very much.
I’m a bit of a closet Orico fan – without necessarily meaning to, I’ve accumulated dozens of their 2.5 and 3.5 enclosures over time. I see similar weird disconnects with a couple of the later 3.5 units I have – I’ve deduced that it must be tied to the UASP-enabled controller (an ASMedia one, I think?). None of the legacy 2.5s or 3.5s exhibit the issue, and I do see the same on a UASP-enabled Yum Cha eBay 3.5 enclosure (but no way to pin that on ASM, as controller unknown).
I have no idea whether the JMicron controller in this particular unit of yours is prone to the same issues, or whether it’s just flaky UASP support somewhere.
It’s left a sour taste for UASP for me at home – when it works it works very quickly, but if I am leaving a big transfer to go overnight I always pick a non-UASP enclosure to do it – because I can almost guarantee that I’ll come back to a ‘location not found’ hang the next morning if I don’t.
It’s almost like if I’m not watching it, it disconnects. Very strange.
Cheers mate.
ah ok
happy flashing
Thanks! It is a neat read, get a M2 in there and it;d be even better!
7 NAND enter! 1 NAND leaves! https://youtu.be/9yDL0AKUCKo
Yet another indicator that I should look into an external case for my SSD or get an m2 external SSD “gumdrive”. (gumdrive term copyright me).
I bet longevity on the full SSD parts is a few orders of magnitude better as well.
One odd thing, in case you decide not to wait, is that it disconnects and reconnects “randomly.” Now, that’s not to say that it’s not there when I need it, it’s always there if I look for it and it doesn’t have any problems while being used—but I don’t know if it’s the enclosure or the Optane that is making it a bit wonky with what I assume is a power saving feature. I’ll come back to my PC after a while and just that drive will have a File Explorer opened up as if I’d just plugged it in.
Of course, there’s also the write speed issue. I’m inclined to blame that on Optane but I can’t know for sure unless I test with another drive.
USB-C has its own specification and was released about the time as USB 3.1. USB 3.1 brought SuperSpeed+ (10Gb/s with 128b/132b encoding) which was called USB 3.1 Gen2. To make that made sense, they renamed the existing USB 3.0 (5Gb/s 8b/10b encoding) to be Gen1.
Later, in USB 3.2in they introduced the ability to use two links in parallel. This gave USB 3.2 Gen 1×2 and Gen 2×2. To fill this out, they again renamed the previous USB 3.1 modes as USB 3.2 Gen 1×1 and Gen 2×1. As confusing as renaming things (again in the case of USB 3.0) is, it does lead us to a nice “Gen AxB” where A is the signaling technology (5Gb/s 8b/10b encoding for 1 and 10Gb/s 128b/132b for 2) and B is the number of links.
So, if you take any generation alone, it makes sense, but if you try to follow along and understand it historically, it’s a mess. It also confuses the strict numbering scheme people because a USB 3.2 (Gen 1×1) link is slower than a USB 3.1 (Gen 2) link. But, if we use the new naming scheme, it’s not as bad: USB 3.2 Gen 1×1 vs USB 3.2 Gen 2×1. It’s not hard to imagine the 2×1 is better than 1×1. There is the ordering: 1×1 < 1×2 < 2×1 < 2×2 which follows the normal sorting criteria.
Keen to hear your thoughts (when the time is right mate) on that little Orico M.2 enclosure Fish. Looking at a couple for spare sticks I have laying around – if the performance is worth it I’ll bite the bullet at some point, they’re pretty cool and I like the idea of a real heatsink.
USB 3.1 Gen 1 is really USB 3.0? What a mess. 🙁
Drat!
The adapters that support USB 3.1 Gen 2 were tested on both Gen2 and Gen1 (which is is the same as 3.0). Check the graphs, Gen 2 numbers top the charts, but it’s not dramatically different for the Samsung drive.
Best I can tell, as expected, there’s no sticks sporting super secret special SCSI sauce.
Sorry, real world tests will be part of the final piece, this was more of a first take/sneak peek/preemptive appeasement since it’s going to take a while to complete.
Wouldn’t mind the ability to click on graphs on mobile and get the image in a larger size. Really hard to read that text on my phone and zoom is locked on the page.
“If you must have a conclusion at this stage, mine would be that I like the plucky SanDisk Ultra the best of the flash drives. ”
Mine is why not no real world tests say, such as a thousand 1MB pics, and a windows ISO file??
Maybe usb 2.0 port vs 3.0 vs. 3.etc.?
Yes, exactly this. I was copying a bunch of stuff to a drive before heading out on vacation and this happened to me. Oh, I got such a good price on that 128GB USB3 sticks. I though, hey, it’s USB3, it’s got to be pretty fast or why bother being 3 when 2 would do? Why indeed. It’s completely inside the envelope of what USB2 can do. I’m solving the problem by no thinking of it as a R/W media, but as a R/O one. I’ll just put stuff on it once (overnight some night) and leave it there. 🙂
Two things:
The biggest limitation to performance on the USB sticks is that they’re just completely bottom of the bucket devices. Cost is about the only performance criteria in their design. SD cards used to be like this, but they actually got used in devices where you could notice if the performance was bad. So, they developed specifications to improve (and guarantee) the performance of cards. The whole “Class” and “UHS” certs and specs headed there and A1 (and A2) finished it off. It used to be common to see SD cards with performance like these USB sticks. Now, with A1 rated cards, it’s not unusual to see >1500 4k write/s. I literally have SD cards that got <1 4k write/s.
USB sticks have not hit that maturity. Some brand (Sandisk) do put some effort into making sure their drives perform not so badly, but they're still behind SD cards. And there's no guarantee that slight changes in a model will give you similar performance. "Oh, you got the XTS version? No, the fast one is the XTX, the S stands for slow…."
The second reason is that they all use the old bulk storage protocol which can only have one operation in flight at any time. Remember hard drives before NCQ? That's what bulk storage is like. The newer spec is UAS (USB attached SCSI) and it allows command queuing and other modern niceties. I don't have data handy, but it makes a huge difference in write speed performance on USB drives, but not as much on read speed. Sorry that I don't have numbers handy.
Throw those two together and you're in for an unplesant experience. I'm tempted to make my own USB stick the next time I need one. I'll either use an A1 rated SD card in a good performing adapter, an mSATA drive in an adapter, or (if I'm feeling rich) an NVME drive in an adapter. My fastest USB port is only 5Gb/s, but that's going to change when I rebuild my PC (for the first time in 5 years).
There’s a huge difference between the flash drives using USB 3.0 and the SSDs using USB 3.1 or mSATA. How much of the difference is because of the drive and how much is because of the difference in USB?
In other words, if I use an SSD with USB 3.0, will the performance be closer to the flash drives or to the SSDs?
Try copying a directory and then remembering you need to copy several other directories. Add them to the task. Do that a couple times and see the copy time go up seemingly exponentially.
Doh! Thanks for the correction/clarification. I’d forgotten it was the partition alignment. One of my first SSDs had terrible performance until I got it aligned right. As I recall, I made an empty partition in front to offset the formatted partition by the correct amount. Then I deleted the first partition. Good old Partition Magic.
There’s formatting and then there’s partitioning. Most solid state drives (thumb drives included) need to have partitions start on a boundary that is divisible by hardware block size for best performance (usually divisible by 512KB although the size may vary and is not inherently related to file system cluster size).
Most modern partitioning tools take care of this for you but it’s something to remember with SSDs.
Excellent write-up. I enjoyed the review once realized where it was going. Good idea to level the field with random read/writes. We all know USB sticks are terrible. Now we know specifically why.
That was a problem during the first couple generations of SSD’s. But its not a problem anymore AFAIK. Maybe USB sticks have remained more primitive.
Tell your sister, that Dr.Fish artwork is fabulous.
Are the USB ‘stick’ drives using Bulk Transport or do they use UASP? I assume the USB/SATA and USB/NVME adapters are UASP or you woun’t be seeing the speeds you’re seeing.
Could we get a table or something with that info? I have no idea how to check that on Windows, but under Linux you can use lsusb -v and look for ” bInterfaceProtocol 80 Bulk-Only” I can’t find my UAS capable USB/SATA adapter right now so I can’t verify what a UAS drive looks like there. I know you get an entry in the system log like “scsi host6: uas”.
Now, the new NVME 1.4 drives can have no DRAM but instead use a small chunk of host memory over PCI-E, right? How do those play in a USB to NVME enclosure? There’s no way to get any host memory.
I want to make a NVME/USB portable drive, but I fear I’ll be lucky enough to pick some incompatable drive and enclosure.
I used to do that, but I ran into some drives whose performance changed drastically when I did that, so I stopped.
Soundtrack to this review: https://youtu.be/Hj8YXtEGcwM
Interesting, I hadn’t considered that, but it makes sense. My personal habit is always to format portable drives to NTFS, so I never gave it a second thought. The full format was just a quick stab at leveling the playing field before running the test.
Because sometimes they pick the parameters of the filesystem format–block size (cluster size), # of FAT coppies, various padding parameters–so that it aligns well with the actual blocks in the flash. It’s not unsual for the first few meg of a USB drive to have different behavior than the bulk of the drive.
I typically image new drives and keep that around in case I ever need to erase the drive–because I’ll get back the filesystem exactly as the manufacturer designed it.
So, I wouldn’t be surprised if you’re not going to see performance similar to other people who have these same drives.
Fair, I’ve added “random” in front of the 4K tests when I introduce them.
Also might want to explain/distinguish between random and sequential reads/writes since…how many flash drives are being hammered by random read/write operations?
My Patriot Rage didn’t get the write speeds in your tests that I’ve seen (>70MB/s) for large file sizes (ie, a multi-GB movie). I can/will test my drive with your settings to confirm/deny these results, but perhaps the file sizes being tested here are quite small?
“There can only be one Flash!”