Even in an era of cloud-everything, the humble USB stick still has a role to play. With the recent end-of-year sales events, there’s been a few different deals for such flash memory products. The Samsung Bar Plus is a favourite, but from time to time, the PNY Turbo Attaché 3 seems to pop up at a decent price – lately at AU$23.07 for a 256GB drive. This wasn’t all that much less than the AU$23.90 that a Samsung Bar Plus would command on sale. Is there any chance that this could be a sound competitor?
Unboxing
The unit comes in the common colour cardboard retail hanging package with the drive encased inside a plastic bubble. The product advertises “10x faster” although the comparator is hidden in the fine print on the rear – namely compared to USB 2.0 devices which they’ve quite unfairly down-rated to just 3MB/s write speed. So are they using this as a roundabout way to say 30MB/s write speed?
The rear has a list of relatively bland features and specifications. One thing does stand out – this drive is Made in Taiwan but only has a one-year warranty. This is perhaps the shortest warranty on flash memory devices I’ve seen as of late. This compares poorly to Samsung’s five-year warranty.
Cutting along the dotted line, the plastic bubble can be released out of the cardboard. The device is not sealed into the bubble. The device has a captive slide cap, made of smoked-translucent plastic, offsetting the light grey plastic body.
The branding is printed on the slide cap, with no sign of serialisation on the product’s exterior – a sign of the fact it’s such an inexpensive device. Compared to the Samsung Bar Plus, it feels a bit cheap, but at least it does have a proper USB-A 3.0 connector and is not a System-in-Package design which may make it easier to recover from certain sorts of damage.
CE and FCC certification logos are moulded into the rear tail end. An internal latch provides a detent for the slide cap.
The end of the cap is open, so dust can get in. There is a hole in the tail which allows for attachment to a lanyard.
Looking closer at the seam, it seems that the plastic is welded together, so it’s not going to fall apart easily.
Device Information
The product has a VID of 0x154B and PID of 0x1006 with REV 0x0110. In its default FAT32 format, there are 247,892,934,656 bytes of free space.
The drive arrives pre-partitioned in MBR format with a single partition formatted as FAT32 (rather than the more standard exFAT). The partition offset is 129536 sectors – 63.25MiB which is a bit of an odd figure.
The reported device size is 247,993,466,880 bytes total. The firmware revision as PMAP tells me this is a Phison-based device.
Using GetInfo, it appears that this is based around a Phison PS2251-19 with firmware 10.02.5D.H dated 2nd November 2023. It was origianlly produced using MPALL v6.00.04, but I suspect the production date and time are a bit off – could this really be a 10-year old device? The unit uses Hynix TLC flash – at least it’s not QLC.
The device is preformatted as FAT32 with a 64kiB cluster size according to its manufacturing data. ID_BLK version is 1.5.43.0, with a unit size (?) 98304kB and total device size of 236544MB.
Performance Tests
How well does it perform? Let’s see. Testing was performed with a Lenovo Legion 5 17″ (17ITH) based around an 11th-generation Intel Core i7 CPU running Windows 11.
HDTune
Out of the box, sequential reads averaged 126.6MB/s. Writes, however, was a very depressing 16.2MB/s and seemingly froze while testing burst rate and access time. I waited an hour, but still no result … but it didn’t error out and it was still blinking its red LED activity indicator. Let’s try again …
Second time around, reads averaged 121.5MB/s and writes averaged 14.7MB/s – similar to older Class 10 SD cards. Eww. This time, I let the drive complete the benchmark, which returned an impressively bad access time of 3236ms! That’s … really … really … slow. Is this thermal or lifetime throttling? Or more likely, a sign of a weak controller? That’s horrible if you’re working on lots of small files – this drive is really going to choke.
The sequential read of this latest write was 121.9MB/s, so overall, it’s a consistent ~121MB/s read. A full random fill and read-back took 4h 56m 1s, which meant an average throughput of around 26.6MB/s round-trip. Again, a test of patience.
Reads overall perform fairly consistently, but writes are surprisingly bad. Zero IOPS. ZERO. The cache plot seems to suggest that the drive may be doing strange things when it figures its being benchmarked – likely some kind of ATTO optimisation.
Read IOPS show a consistent maximum access time topping out at around 1k IOPS. Writes were bad to the point that the app simply could not compute a proper result, with over 10s stalls.
The file benchmark shows that writes proceed at a good 75MB/s speed with dips. A good chance that this is a small cache that’s managing the benchmark, or perhaps a benchmark optimisation. The block size measurement shows some inconsistency with higher sizes. So perhaps there is a small amount of cache.
CrystalDiskMark
The CDM default test results are actually not bad at all – but don’t get the wrong idea – you’re not going to see that >=86MB/s writes for long.
This is somewhat reflected in the peak-mix workload where the mix workload shows a big dip in speed.
The real-world-mix workload shows a similar slump in result.
ATTO
ATTO seems to show an inconsistent behaviour, peaking by 128KB for writes and 64kB for reads, but larger block sizes show writes dipping a little too.
Repeated tests show a similar inconsistency.
AS-SSD Benchmark
Sure, it’s not an SSD, but I felt like suffering so I ran AS SSD on the drive. This benchmark is not fooled. How’d you like a write score of just 2?
Anvil Pro
Anvil Pro also doesn’t seem to be fooled, but in both cases, the results are the lowest of the drives I have tested that I can recall.
H2testW
At least the drive’s capacity is honest, but the result shows a very similar time to fill and verify as the HDTune test.
Conclusion
The PNY Turbo Attaché 3 is an overall disappointment. It really doesn’t deserve its turbo moniker, nor does its claim of 10x faster than USB 2.0 drives (3MB/s) at writes seem to be upheld in sustained writes, achieving only in the 15-16MB/s range for writes. This is well behind what even modern SD/microSD cards can achieve. Small writes can be done quickly, likely because of a cache or some benchmark optimisations.
Sustained reads reach about 121MB/s, which is not much better. At the price that it sells for, if you have the chance to pay a tiny bit more for the Samsung Bar Plus, it’s a much superior offering compared to this, which feels a bit more like an old USB 2.0 drive when it’s writing. This is especially the case due to low write IOPS with long random write latency in the 3.5 to 10s range. The drive was literally so slow that it broke some benchmarks.
Very much not recommended especially in 2025 as better alternatives exist at a similar price.
It might not be the same model, but I have an effectively identical-looking flash drive here, also PNY and also 256GB, that’s either 10 years old or at least getting close to it.
That speed claim is pretty amusing though. How long until we see “1000x faster! (compared to a typical paper tape punch)”?
Haha … I betcha we can do 1-billion times faster!*
* as compared to caveman with rock and chisel.
– Gough