Samsung USB 3.1 Flash Drive BAR Plus MUF-256BE3/EU (WHL #45)
Wan Hung Lo #45 – Christmas Turd Edition!
Yeah…wasn’t planned to do that, but just like with the much more expensive “proper hardware” item from last year, the BenQ EX3501R that caused m u l t i p le revisits, I guess I found something worthy of in-depth reviewing. A simple USB flash drive.
Let me explain. I’m in the market for some 200GB+ device, as I’ve come to the point in the upgrade cycle where my 128GB Sandisk Extreme Pro (wank, wank) is no longer capable of holding all data that I want to carry around. Before I bought that one, I had the Sandisk Cruzer Extreme 64GB model, which (obviously) came after the Sandisk Cruzer Extreme 32GB (which I donated to my brother). All of these drives have significant performance advantages over regular USB drives in terms of random writes, typically benchmarked with 4KB block size like SSDs. And they are quite a bit more expensive than your run-of-the-mill flash drive that is advertised with similar sequential read/write speeds. But the Sandisk real-world performance is just so much better if your workloads aren’t just moving pirated movies from A to B, as now is shamelessly used in advertising. Imagine putting tons of small files onto such a drive (git repositories, anyone?) and write with 100 KB/s instead of 10 MB/s. Yeah.
So despite their performance advantage (mar-ke-ting! anyone?) and higher retail price (“performance premium” – for premium performance), the market is pretty scarce, especially for large capacities. I’ve been watching the Sandisk Extreme Pro 256GB hover around the 100€ mark for a while, but it won’t come down all that much. For future reference: The cheapest Sandisk 256GB USB drive currently retails for half of that, the cheapest overall drives are about one third. Same goes for SATA SSDs where cheap crap has reached 35€ and decent drives (Crucial MX500, Samsung 860 EVO) are around 45-50€. I won’t really touch “proper” USB SSDs, as they typically are quite bulky and block adjacent USB ports or even require your laptop to be jacked up if the port placement would make the laptop stand on the USB drive when plugged in. Also, they aren’t exactly cheaper…
Recently someone posted benchmark data of a Samsung drive that I have never heard of – the “BAR Plus”, MUF-256BE3/EU. I was hooked!
Very small form factor, much shorter than my Sandisk drives. Outer shell made from metal – I would still prefer a rubber case, but this one doesn’t have any slider mechanism that could break over time (and annoy you the entire time until it finally gives up). It also features a sturdy mounting hook and Samsung says it’s waterproof. Great!
Retail price? Just shy of 60€ delivered. So I had to try…
Side note: I later noticed that these are also available in grayish black color, the one I got is called “Champagne Silver”. Yeah, whatever, the black one also looks cool, but I cannot tell if scratches will destroy the look in no time.
Specs: “Up to” 300 MB/s reads, writes will be slower than that (just give us some number, you fools). Size 16x40x12mm³, 12g, 0°C to 60°C operating temperature, water proof for 72h (1m below sea level…), also not afraid of X-rays, magnetic fields (15k Gauss ~ 1.5 Tesla) or severe impacts (1500g over 0.5ms in all three axis). Five years warranty, compared to 30y / lifetime warranty of Sandisk.
Well, I don’t get it why they won’t advertise beefy hardware with more specs, especially when all of the competitors are shy as well. I mean, just guarantee like 1 MB/s of 4K writes and you’re the king of the USB flash drive world…?
What’s it like?
These benchmarks were performed after one full write cycle (sector number, sequential write) with HD Sentinel, which I very much recommend for all of your HDD/SSD needs. Contrary to the Sandisk sticks, the Samsung does NOT show any SMART data, meaning it does not appear as a regular hard disk / SSD, but instead acts like a dumb USB stick. There’s also no temperature reading.
As for the different readings, CDM always presents some different numbers than AS SSD does. That’s nothing to worry about, just stick to your tool of choice and compare results only to other results of the same application. Also, CDM changed benchmark types from version 5 to 6, so be aware of that. Fortunately, both tools display their version number in the title bar.
It’s also modest in power consumption. While it is relatively power hungry in idle at about 0.2 to 0.25W, it doesn’t go much beyond 0.3W when reading, and 0.6W when writing sequentially. Due to the tiny case, it does however get noticeably warm, that is, I’ve measured 43°C near the hook with a thermocouple, so 20-25K above ambient. That’s not in the unpleasant to touch realm, but it’s clearly warm to the touch.
So, is everything just fine and the Samsung is the best stick ever? Well, certainly not, but that took a while.
First, null the device again (with HD Sentinel). That should act like a factory reset.
Not much of a change. Filling it up again, using h2testw. 25%, then run the bench:
98%ish (AS SSD requires some 3GB free for 1GB of test data):
So fill state doesn’t have much of an impact on performance, which is very nice. Read speeds are down by like 1/7, but everything else is unchanged. Usually SSDs do not like high fill states, but given the huge overall capacity, this one still has 256GiB-256.6GB ~ 17GiB to work with, even when filled to the brim.
Everything’s fine man, just use it!
Well, not so fast. First of all, there’s no constant write speed. Instead, the device acts like it can take 100 MB/s – and then stops every so often. This was already noted in the initial forum post, and the HD Sentinel graph nicely shows this behaviour (the other performance graph maxes/averages this out, which I’d call a bug that needs to be addressed):
The darker the colour, the slower the write speeds. So at an average of 60 MB/s as shown via CDM/AS SSD and also derived from the completion time of 1h 18min over 256GB ~ 55 MB/s in HD Sentinel, every ~fourth block (25 MB each) is down to 25 MB/s, meaning the others have to be completed at a rate of 95 MB/s. Interestingly, this behaviour fades over time or over bytes written in sequential order.
When nulling the device again, this pattern is still visible, even though it’s not as pronounced as before. So the type of data written (zeroes, random data, compressible data) could be a factor, but the already stored data could be as well. As there’s no SMART information, one cannot issue a TRIM command and see how the device reacts to discarding large portions of the storage space. I’d much prefer a user-TRIM’able device over this.
The six or seven slow blocks in the middle of the device (no fixed position) also showed up in other write tests, not sure what to think of them…
And now for the ugly part. Base line, “steady state” speed. Basically every flash drive can only take so much before all blocks and pages are in a non-ideal state where for any new write, entire pages or even blocks have to be rewritten. Usually that takes place during idle time (“GC”, garbage collection, typically in conjunction with TRIMming the device), and aside from the annual firmware hiccup of my Samsung 830 (256GB) main SSD, I have never observed this strange behaviour on the limit myself.
Well, evil Santa punished the device. Random writes with random data, apparently these are 4K writes instead of those 25MB blocks that are shown. Doesn’t tell, I’ll ask for clarification. This is what you get:
…holy moly. On little writes, the device also employs this gulp-choke method as shown with sequential writes. So it takes in much more than it can handle, and then it breaks down. Not to 25 MB/s, not to the benched 12 MB/S random write speed. To MUCH lower speeds. 7h 50 minutes to write the entire device, averaging at 9 MB/s, dipping down to kilobytes per second.
I don’t have any screen shots of this, but AS SSD benchmarks after this (a couple of hours after the test completed at 6 a.m.) were terrible. Terrible like the ones shown in a minute. No change when nulling the device or anything. But when I shut down the computer overnight (USB port is powered permanently!) and restarted benches in the morning, THIS was the result:
Almost good as new
I couldn’t resist and did more writes; not random stuff, but two more full write cycles. Nothing. Then a 5GB random write with AS SSD. Voilà, results were down as before…but how fast will it recover?
No real recovery within 24h. The heck! Sometimes sequential writes are back up, which seems to be killed by the bench itself (every bench above 3GB ~ 1% with random data can probably induce this behaviour!). There has been time with the computer on, and there has been downtime with power applied to the drive, but no data pin activity. I’ve not been able to tell internal drive activity by monitoring power usage, as the general increase from a quarter watt of idling power to 0.3W when reading with screaming speeds is quite small. Furthermore, when the general device performance is down like this, the drive never consumes the 0.6W observed on writes before. It seems like either NAND or controller is mostly in a sleep state, waiting for the other party to complete the task.
Well, as time runs out for preparing the Christmas..umm, feast, and benching has been time consuming, I’ll call it quits and add an update when there’s more consistent data available. For now the conclusion has to be: Great device if you never intend to write more than 3GB of teeny-tiny files to it. So block backups (dd, anyone?) should be fine, but never file-based backups of large folders, let alone running virtual machines on this disk. If however you dare to use more than one percent of the entire capacity to write whatever you like to it – well, you’re probably not better off at all compared to a Wan Hung Lo flash device for 2/3 the money. Still, doesn’t it look fancy and upmarket?
Nerdy Christmas, everybody…!
[…] Well here’s a first I’m about to RMA a piece of hardware, and I needed performance data from when the thing was new. Probably got a screenshot from back then, but searching for it on a machine that is inconvenient to access…just search the interwebs instead. And so I did, and entered the Google image search, and found what I needed. Yeah, turns out I just found my very own website. […]