An SSD in a USB 3.0 Enclosure: Part 2 – Transcend SSD340 256Gb + StoreJet 25S3

Having just done a review and teardown of the Transcend SSD340 256Gb SSD, astute readers will note that the product was purchased in a bundle with a Transcend branded USB 3.0 2.5″ enclosure. While the inclusion of the enclosure is generally targeted at users who are looking at an easy way to migrate the data from an internal drive to their SSD (say, laptop users), it opens up a new possibility – turning it into a USB 3.0 SSD.

One of the more popular articles was my former, Am I Crazy? An SSD in a USB 3.0 Enclosure. In that article, I cover the motivations behind creating a USB 3.0 connected SSD and how it pays performance dividends compared to most USB memory key devices. However, I also noted some cautions with unexpected power loss in another article published soon-after.

I’m pleased to report my Kingston SSDnow V300 (Sandforce SF2281 based) SSD in an enclosure is going great – no data loss has been experienced and performance degradation has not been apparent. In general, I have preferred SandForce/LSI based drives for these roles because of their compression in reducing write amplification, their generous over-provision and effective block management even in case of non-TRIM environments.

As the Transcend SSD340 came in a bundle with an enclosure, and I was going to review the enclosure anyway, this would be the ideal circumstance in determining whether the Transcend SSD340 would make a good USB-connected SSD.

Unboxing the Enclosure

transcendstorejet25s3box-front

The enclosure is the Transcend StoreJet 25S3. It claims to support SATA 6Gb/s connectivity with HDDs and SSDs. It features a one touch backup button and file encryption software (by download). It is 13mm in height, and features a rounded aluminium shell with plastic ends which are secured by two phillips head screws.

transcendstorejet25s3box-rear

The enclosure is covered by a 2 year warranty, and is backwards compatible with USB 2.0. The package includes the enclosure, USB cable, quick start guide and warranty card. The software can be downloaded online.

DSC_7964

Also included in the package are some promotional leaflets for other products in their catalogue.

DSC_7965

The enclosure itself is a rounded aluminium shell with the logos printed on one side, and nothing printed on the other. A small cutout is made for a window that exposes the activity LED.

DSC_7967 DSC_7966

The front cap has the USB 3.0 micro-B connector, and backup button, with the other end having the batch number and product name/approvals moulded in. Both ends are secured by small philips screws (you will need to supply your own screwdriver).

DSC_7968

The bridgeboard PCB utilizes an Asmedia ASM1153E third-generation USB 3.0 to SATA controller. This chip supports SATA III 6Gbit/s connectivity and UASP. There is also a serial EEPROM that holds the configuration for the chip. This is a much nicer chip compared to the first generation Asmedia ASM1051 (from archive.org) I used in my first attempt which only supports SATA II and no UASP.

DSC_7969

The underside seems to show there is a MOSFET switch to disconnect the drive from the bus to comply with USB standby power requirements, and a polyfuse F2 to protect against over-current.

The enclosure was designed for regular height laptop drives (9.5mm), and thus inserting the 7mm SSD without an adapter results in some rattling if shaken hard. Users are advised to pad out the extra space with some paper to prevent movement fatiguing solder joints on the connector.

Performance Testing – Non UASP

USB Attached SCSI Protocol is a new protocol for external devices on the USB 3.0 bus which improves performance by allowing command interleaving. Unfortunately, UASP is only available on USB 3.0 controllers with appropriate drivers, and under Windows 8 or higher. Unfortunately, that means that the vast majority of the computers running Windows 7 is unable to benefit from UASP. The resulting performance is illustrated below.

HDTune Pro Sequential Read

USBR-HDT-Read

The drive only reaches an average read rate of 260.6Mb/s – something similar to about SATA II performance.

HDTune Pro Sequential Write

USBR-HDT-Write

Without UASP, the write speed averages 183.4Mb/s – which is also SATA I/II grade performance. This is all due to the overheads imposed by the USB bus so far.

CrystalDiskMark

USBR-CDM

CrystalDiskMark reflects a similar outcome, however, for some reason, the read rate seems to have suffered. The write rate varies significantly, and this may be due to lack of TRIM in combination with DRAM cache. No significant benefits were seen with queued accesses, indicating the lack of UASP.

H2testw

USBR-H2testW

No corruption was encountered with H2testw, however, the performance degraded somewhat further to 106Mb/s write rate.

Performance Testing – With UASP

As I also have a high performance laptop with Windows 8.1 and USB 3.0 drivers which support UASP (i.e. the Refurbished Asus laptop), I was also able to determine the performance of the drive with UASP and illustrate the performance differential.

HDTune Pro Sequential Read

USBI-HDT-Read

The read performance has shot up and is now much more comparable to the SATAIII connection. Noting that the USB 3.0 connection is 5Gbit/s, and the SATA III connection is 6Gbit/s, it seems that the average of 430.8Mb/s is right on the money when the ratio is considered.

HDTune Sequential Write

USBI-HDT-Write-SecE

The write performance with UASP mirrors that of the internal performance – very impressive indeed. However, this result was only achieved after a secure erase. Before that, I saw this …

USBI-HDT-Write-TRIMDirty1

Not believing the degradation, I ran it again.

USBI-HDT-Write-TRIMDirty2

Once the drive is filled, and all user accessible sectors are filled, the write performance of the drive degrades to the point where it is slower than even most USB keys. This occurs even if the written pattern is 0×00′s as HDTune Pro does – so no amount of “manual” data shredding will restore performance.

SSDScope (Transcend’s SSD Utility) refuses to detect the drive when it’s inside an enclosure, and thus refuses to TRIM or secure erase. Deleting and creating a partition on the drive using Control Panel doesn’t restore performance either – indicating Windows does not perform TRIM on USB connected SSDs.

After a manual disassembly and secure erase, I continued with the next benchmarks.

CrystalDiskMark

USBI-CDM-UASP

The impact of UASP is shown in the queued commands having much better throughput than the unqueued benchmark. It does seem like the choice of USB controller may have influenced the 4k result as well, so unqueued requests on this controller (Intel) seems to be slower than Bulk-Only-Transport unqueued requests on the other (NEC/Renesas) controller.

H2testW

USBI-H2TestW

No corruption was found, although it seems that the verification speed was limited by CPU-limitations of the program.

Conclusion

The impact of a lack of overprovisioning on this drive is clear. The Jmicron controller isn’t able to manage the blocks in such a way that once the drive nears (or reaches) complete usage, the write performance degrades unacceptably. It was also determined that TRIM does not work via USB 3.0 enclosure regardless of UASP support, and that manual TRIM is not possible either as SSDScope will not detect the drive inside the enclosure. Only by removing the drive into a desktop connected by a SATA connection can TRIM or secure erase be undertaken to recover full performance (hence, the degradation is not due to write throttling).

Unfortunately, the situation of operating without TRIM and at near full capacity increases the write amplification which further accelerates wearout of the SSD. Operation without TRIM occurs when the SSD is connected through a USB 3.0 case, under operating systems not supporting TRIM or using SATA/RAID controllers with custom drivers, or in a RAID array.

As a result, the Transcend SSD340 is not recommended for use in those environments. I still continue to advocate the use of Sandforce SF2281 based solutions in USB enclosures for this reason.

Posted in Computing, Flash Memory | Tagged , , , , , | Leave a comment

Review, Teardown: Transcend SSD340 256Gb 2.5″ Solid-State Drive (TS256GSSD340)

Solid state drives are now a vital part of every computer I build and use. Most of my commonly used machines in my fleet have already been outfitted with SSDs, and it has paid dividends in terms of faster response times, faster loading times, lower power consumption, better shock resistance and improved reliability. In fact, they’re so important that I have ten SSDs in my fleet and I’ve never looked back. To date, only one reallocated block has been seen across the whole fleet, with no outright failures and a write total across all the drives of about 32TiB.

Reducing prices have only made them more attractive to enthusiasts and mainstream users. In fact, drives can be found below the 50c/Gb mark nowadays, especially in larger capacities. It makes sense for me to buy some drives to hold in reserve in case I need more high speed storage, or for new machines.

As a result of this, and my desire to understand about the performance characteristics of the drives on the market, I have decided to diversify my SSD portfolio by purchasing drives from the value segment, where competition is heating up.

In this review, we will be focusing on the Transcend SSD340 256Gb 2.5″ Solid-State Drive with model number TS256GSSD340. Transcend has been a long-time flash memory vendor, with a history of very competitive pricing but varying quality and compatibility. It is the largest drive in the Transcend SSD340 series, and is labelled as a “premium” drive by the manufacturer, but is priced with a price tag to compete with the value-series drives (e.g. Samsung Evo, Crucial M500, Kingmax XValue).

The Transcend SSD340 was on special at Mwave, in a promotional bundle with a Transcend USB External Enclosure (see next review) for a total cost of about AU$147 after shipping and insurance. The SSD itself is listed at Mwave for AU$129 (plus postage and insurance, it works out to be about AU$141). As a result, it’s almost as if they’re giving away the enclosure, which is always handy to have and helps with migration.

The drive is very well priced, however, is the Transcend SSD340 a worthy contender? Lets find out.

Unboxing

transcend340box-front

The drive comes in a retail kit configuration, inside a matte finish colour-printed cardboard box. The capacity is stuck on the front with a label. The front of the box clearly indicates the use of a DDR3 DRAM cache, DevSleep support, 7mm height, and included software. The front also confirms the inclusion of a 2.5″ to 3.5″ adapter bracket. In general, this is a pretty generous inclusion for an SSD retail kit, only bettered by the Kingmax for inclusion of a SATA cable. However, most SSDs ship pretty bare – some “drive only”, so Transcend can be commended for putting so much into this price point.

transcend340box-rear

The rear of the box also makes mention of included Quick Installation Guide, Warranty Card and mounting screws. The drive itself weighs 52 grams and utilizes Synchronous MLC Flash. This is considered “middle” ground, whereas Toggle NAND takes out the high-end, and Asynchronous NAND takes out the bottom-end. The use of MLC is also notable, as MLC is generally more durable than TLC solutions when it comes to write cycles per cell (such as the Samsung Evo and Sandisk Ultra II).

The drive seems to be quite extreme in specifications when it comes to temperature as well – being rated for operation up to 70 degrees C. I don’t think you need to worry about that happening even in the most poorly ventilated laptops!

The box doesn’t shy away from providing performance data as well – claiming a 520Mb/s read and 290Mb/s write speed under ATTO, and 485Mb/s and 270Mb/s respectively under AS SSD. It also claims an IOmeter IOPS of 68k for 4k Random Read, and 69k for 4k Random Write. These give us some solid numbers to compare with, with the IOPS value pretty much average compared to some other competitors’ marketing literature in this “value” segment.

The drive itself is Made in Taiwan, and is provided with a three year warranty, provided the wear-out indicator in their monitoring software (SSDScope) does not show 0%.

transcend340box-side1

transcend340box-side2

The sides of the box also indicate support for Trim, and the availability of system cloning software as part of the SSD Scope package which must be downloaded through the Internet.

DSC_7970

DSC_7971Within the package, as promised, there are two sets of screws (one to secure the SSD to the adapter bracket, and the other to secure the bracket to your computer), the warranty card, a quick start guide and three promotional leaflets with their product catalogue.

The bracket is also provided, which is made of metal, and black in colour with white logo print. It seems very similar to the Kingmax bracket in design, and it serves its purpose quite well.

Of course, you get the drive, inside an anti-static shielding bag.

DSC_7972 DSC_7973

The drive itself claims to require 1.2A from the 5v rail to operate. The drive itself feels very light, and this is due to the use of plastic to manufacture the body. This is also why all the screwholes are formed by sinking brass threaded studs into the plastic mold. The drive itself is sealed with a warranty label on the underside.

Aside from a serial number, there doesn’t seem to be any other numbers or identification which alludes to drive encryption ability. I suppose that could be considered a drawback, if you intend to use self-encrypting drive abilities, however, most users do not.

Teardown

It might seem strange, but this idiot (yeah, me) decided to void the warranty on day 1. Under the shiny holographic seal sits just one screw, which needs to be removed. After that, you pry along the edges to free the plastic backplate from the frame of the drive to reveal its innards.

DSC_7974

Already, we can see that the drive itself is pretty empty, and has adhesive foam rubber to keep it from accidentally contacting the rear plastic plate. The PCB itself is secured to the other frame by a single black screw. The design is extremely minimalist! Lets take a much closer look at the PCB.

DSC_7975

The main controller of the drive is a Jmicron JMF667H, very recently released to the market (mid-2014). Jmicron’s reputation when it comes to their controllers typically is mixed, and it seems the JMF667H is a value controller featuring four channels with up to eight chip enable signals per channel. It seems Anandtech reviewed this exact model of SSD, but their unit was different in its flash. They seem to claim a limit of 256Gb on the JMF667H controller, despite the unpopulated BGA pads.

The flash modules come from SpecTek (a Micron subsidiary, formerly reseller, which was founded upon selling partially defective devices, but has since grown to selling full-spec devices as well). The devices themselves are marked with PF567-10AL, which, when PF567 is entered into their laser mark decoder, it provides a part number of FBNL85A91KDMABH7. Plugging FBNL85A91KDMABH7-10AL into their MPN decoder provides the breakdown that the device is MLC 512Gib with Density Grade 1 (94-100%) and 4 dies per package with 4 CE pins and 2 I/O channels operable at 200MT/s Synchronous or Asynchronous. The AL marking indicates the device is Full spec with tighter requirements, detailed in their document here.

Given the four packages for four channels, it is expected that this drive will demonstrate the maximum throughput capabilities of this controller. In theory, each channel still has 4 CE outputs unused, so if not for the 256Gb limit of the controller, it would be possible to build a 512Gb SSD using Flash of this density. The drive itself has exactly 256GiB of flash memory, and seeks to offer 256GB of storage space. This results in an overprovision of the difference between binary and decimal (i.e. about 4.8%) which is fairly small and will likely impact upon performance to some degree when full.

Each flash chip appears to be bypassed by five small capacitors surrounding the chip.

The DRAM cache is supplied by Samsung, and is marked K4B2G1646Q-BCK0, which appears to be a 2Gbit (256MiB) DDR3 (latency 11-11-11) 1600Mhz chip.

The date codes are slightly perplexing, as for such a recently “unveiled” controller, the date code implies it was manufactured week 17 of 2013 (a long time ago for this kind of technology), whereas the flash was manufactured much more recently in week 4 of 2014.

The underside also shows little power conversion circuitry near the black single screw, which appears to be a Texas Instruments switching converter of some sort. There doesn’t seem to be many large bulk capacitors (C135-C137 don’t look very big), and as a result, data loss during unexpected powerdown may be a risk with this model.

Interestingly, I spot what appears to be an activity LED mounted on the PCB (D1), despite the fact that it is mounted inside an opaque black plastic enclosure. It’s probably used during testing, but is not strictly necessary. I did not investigate whether the LED is actually active during drive access.

DSC_7976

The rear of the PCB appears to be ready to accommodate many more flash packages. This might be necessary if they have higher capacity drives based on a pin-compatible controller, or if they choose to use lower density flash packages (note the Spectek ones have four dies per package, so if single-die packages were used, then every slot might be populated).

The underside features an EEPROM, likely for configuration, U4. The underside has many inductors, for filtering the power that comes in and for the switching converters to derive lower voltages to drive the controller and flash. No significant capacitors are seen, and it seems there is a slot for a fuse that is unpopulated.

The PCB itself is silkscreened with Transcend’s brand name, implying that this is a home-spun design, specific to them. The product code is 29-3271 and is Version 1.0. The PCB itself was manufactured week 2 of 2014, and is also marked with VIC 2-9.

Overall, it seems like a very average “value” line SSD design. Notable is the absent of any form of thermal dissipative foam or heatsinks for any of the packages, which while not strictly necessary, may be a desirable feature.

Performance Testing

Testing was performed as per the previous SSD tests. The testing platform is an AMD Phenom II x6 1090T BE @ 3.90Ghz running on a Gigabyte 890FXA-UD7 running Windows 7 64-bit edition with the latest patches. The on-board chipset SATA 6Gb/s ports were used with the AMD SATA drivers to interface with the drive. TRIM was enabled on the system. The drive was subjected to a full random fill, and multiple-readback to verify data integrity of all sectors, which it passed.

SMART Data

SMART-Initial SMART-Final

SMART data was read using CrystalDiskInfo v6.1.14 (latest at the time of writing). The left screenshot shows the data of the drive as new, where it shows 6 unsafe shutdowns already recorded. After running the barrage of tests, several values did grow – notably E9, EA, F1 and F2 appear to be directly related to the flash writes used. I would guess that EA is the raw LBA read, with F2 as the host-LBA read count. E9 would be the raw LBA writes, with F1 being the host-LBA write count. A9 and AD seem to be indicative of flash cell cycles in some way, and probably lifetime, but their relationship is not known at this time.

After even further testing (after this article’s tests were completed), I did load SSDScope (Transcend’s own tool) which provided the following data:

SSDScope-SMART SSDScope-MWI

It seems as if this drive isn’t very good on write amplification, as during the course of testing, I really only put in about 5 full-surface writes, but the average cycle count is already up to 36. You will understand some more about this in the next few tests. The media wearout indicator is already down to 99% (no SSD of mine to date has demonstrated any wear during testing runs, and none of mine have wear >1% at this time).

Notably, you may suffer from the “frozen” SSD state problem when it comes to secure erasing your drive. Interestingly, Transcend actively advocates hot-unplug and replug of the drive’s power. Most other utilities advocate sleeping/waking the system instead.

SSDScope-Hotplug

HD Tune Pro Sequential Read

HDT-Read

The drive managed to churn out fairly normal numbers when it comes to reading, with an average of 512.9Mb/s. In the current SSD market, saturating the SATA3 link is considered an expectation.

Drive             Average Read  Access Time
Transcend SSD340 256Gb   512.9        0.069
Intel 730 240Gb          512.6        0.104
Kingmax Xvalue 240Gb     514.7        0.061
Samsung 840Pro 256Gb     527.3        0.047
Crucial M500 240Gb       296.4        0.037

HD Tune Pro Sequential Write

HDT-Write

The sequential writes on this drive demonstrate the problem with lack of overprovision. When the drive fills to completion, the drive itself starts having problems managing the flash translation layer block lists, and finding free blocks to use. As a result, the write speed takes a dive at the end. The overall average speed was 297.7Mb/s, which is very respectable for a value line SSD. Notably, it has a similar sort of “hump” as on the Kingmax Xvalue SSD which is based on a SiliconMotion controller. I wonder why this is the case?

Drive             Average Write  Access Time
Transcend SSD340 256Gb    297.7        0.052
Intel 730 240Gb           263.0        0.042
Kingmax Xvalue 240Gb      265.7        0.038
Samsung 840Pro 256Gb      440.0        0.041
Crucial M500 240Gb        209.6        0.045

HD Tune Random Access Read

HDT-RARead

It seems that the drive puts up a very competitive performance for a value drive, which was surprising. Compared to the other drives I have tested, it doesn’t fare badly at all.

Drive                   512b   4K     64K    1Mb  Rand IOPS
Transcend SSD340 256Gb 12240   6992   3617   466  857
Intel 730 240Gb         9624   8243   2956   440  763
Kingmax Xvalue 240Gb   13721   6519   1999   430  749 
Samsung 840Pro 256Gb    8340   8145   4093   477  879
Crucial M500 240Gb       N/A    N/A    N/A   N/A  N/A

HD Tune Random Access Write

HDT-RAWrite

While the drive puts up a very competitive read performance, the drive’s write performance seems to be fairly poor despite having TRIM enabled and on a freshly partitioned drive (all blocks TRIM-ed).

Drive                   512b   4K     64K    1Mb  Rand IOPS
Transcend SSD340 256Gb 26490   16397  2240   117  237
Intel 730 240Gb        22980   19414  5534   311  669
Kingmax Xvalue 240Gb   26120   21794  5489   255  501 
Samsung 840Pro 256Gb   23211   19755  5596   430  827
Crucial M500 240Gb       N/A    N/A    N/A   N/A  N/A

CrystalDiskMark

CDM-Transcend340

The drive’s sequential read and write performance continue to stand out in the value segment, as well as its medium block 512kB accesses. However, at the 4k small-blocks, the drive is generally an average performer.

Drive                 SeqR  SeqW  512kR 512kW 4kR   4kW   4kR32 4kW32
Transcend SSD340 256Gb506.6 315.2 440.7 317.6 29.53 80.50 246.2 244.4
Intel 730 240Gb       462.9 296.8 383.2 295.6 33.78 85.43 285.3 251.9
Kingmax Xvalue 240Gb  516.8 280.9 381.1 280.8 27.27 85.56 261.8 232.5
Samsung 840Pro 256Gb  523.5 448.6 329.5 427.6 23.20 81.28 190.1 248.8
Crucial M500 240Gb    475.2 283.6 421.2 282.5 26.23 86.09 255.9 242.6

AS SSD Benchmark

ASSSD-Transcend340

The AS SSD benchmarks can be compared to that of the box, in which case, this drive actually goes above and beyond the specifications (485Mb/s and 270Mb/s claimed, 514.21Mb/s and 299.96Mb/s measured). The performance measurements mirror the same results as for CrystalDiskMark. In terms of score, it ranks ahead of the Crucial M500 by a hair, and is one point behind the Kingmax Xvalue.

Drive                SeqR  SeqW  4kR  4kW  4kR64 4kW64 AcR   AcW   Score
Transcend SSD340 256G514.2 300.0 27.4 72.1 232.2 217.3 0.055 0.049 789
Intel 730 240Gb      512.8 279.5 30.0 73.0 262.7 210.1 0.046 0.048 831
Kingmax Xvalue 240Gb 514.2 265.3 25.3 75.0 247.6 201.8 0.046 0.048 790
Samsung 840Pro 256Gb 511.0 439.0 30.8 73.1 255.1 224.4 0.059 0.047 859
Crucial M500 240Gb   493.1 273.7 24.2 75.9 241.7 208.1 0.047 0.066 786

AS SSD Copy Benchmark

ASSSD-Copy-Transcend340

For completeness, these scores are provided. This drive seems to do better for ISO than normal, average for the Program load and slightly worse than what I’d expect for Game.

AS SSD Compression Benchmark

ASSSD-Compression-Transcend340

The results of AS SSD’s compression benchmark revealed an unexpected “wrinkle”. The line is mostly straight, thus implying no compressions is being utilized in the drive itself, however, the dips in the graph on readback are unusual and reproducible. It may be due to the test workload, and integrated garbage collection schedules interfering with the benchmark.

ATTO Disk Benchmark

ATTO-Transcend340

The Atto results for this drive show very strong performance for small block writes. This may be due to DRAM caching helping with write coalescing and “deferring” writes. The read performance doesn’t scale up as quickly as the write does at small blocks, however, it does put in a good performance across the board, reaching peak read performance at 256kB transactions and above, and peak write performance at 16kB transactions and above. This is typically better than many SSDs which are not as small-block oriented.

Anvil’s Storage Utilities

TS256GSS D340 SATA Disk Device_256GB_1GB-20140825-1705

Anvil Pro is another one of those standard benchmarks, and it seems to like the Transcend SSD340 quite a bit, with a score that trounces the Kingmax Xvalue and beats the Crucial M500 by a hair.

Drive                Score
Transcend SSD340 256G3772.04
Intel 730 240Gb      3842.91
Kingmax Xvalue 240Gb 3381.02
Samsung 840Pro 256Gb 4150.13
Crucial M500 240Gb   3737.94

H2testw

H2Testw-Transcend340

I’m glad to report no corruption issues were experienced with the drive under H2testw, although the read performance is capped due to CPU-limitations in the program. The write speed seems to average 304Mb/s in H2testw, which is a bit higher than the HDTune result, but within the same ballpark.

Power Consumption

Testing for power consumption was performed on another system with only a SATAII port. As a result, the drive was only put under a “moderate” load for these measurements – full heavy workloads may see higher current consumption. The current was recorded at about 29 readings per second while the drive ran a full CrystalDiskMark run.

Transcend SSD340 - Current vs Time

The drive seems to show “spiky” behaviour in its current consumption. Once the test was over (near 369s), the drive seemed to consume more power at a periodic interval, which suggests possible scheduled garbage collection algorithm. After a while, the current returned to a flat-line reading just over 100mA.

The current consumption was averaged for the idle, sequential read and sequential write phases for comparison.

Drive                           Idle    Read   Write
Transcend SSD340 256Gb          109mA   255mA  519mA
Crucial M500 240Gb              186mA   289mA  538mA
Kingmax SME35 Xvalue 240Gb      60.5mA  216mA  513mA
Kingston V300 120Gb             119mA   372mA  590mA
Samsung 840 Pro 256Gb           59.7mA  300mA  386mA
Western Digital WD1600BEVS      220mA   680mA  700mA

From the results, the idle current of the Transcend drive is on the higher end of the scale, but not too badly so. The read current consumption is also pretty average, as is the write power consumption. In all, an unremarkable performance here.

Conclusion

While this drive represents the value segment of the market, its performance was extremely solid and slightly better than most value competitors when it came to reads. Sequential writes were quite good, although the write file benchmarks seem to allude to poor write performance or TRIM consistency issues. The construction itself is light and a hair flimsy, but sufficient.

For the price, it seems like it might be a good buy, however, it’s important to note that this performance was achieved with TRIM-enabled, and it’s likely TRIM-disabled systems will suffer from poor performance due to the small level of overprovisioning this drive has (which makes it a poor candidate for USB enclosures).

The drive itself also seems to have a fair amount of write amplification, which is surprising given the DRAM cache which should help with write coalescing. This will consume the flash memory quicker than drives with smaller levels of write amplification (i.e. those with effective compression or write coalescing strategies). This drive doesn’t seem to utilize compression, so its performance is consistent across workloads.

Unfortunately, the drive also seems to feature minimum power-loss protection (or none at all), and it seems possible that data can be lost upon unexpected power removal given the use of DRAM cache, making this a poor candidate for USB enclosures and servers or mission-critical systems.

Overall, this drive seems to represent the normal set of compromises you would expect from a value line SSD and should satisfy mainstream users.

Posted in Computing, Flash Memory | Tagged , , , , , | Leave a comment

Further Investigation: Power Bank Endurance Issues and Modification

Through many tests of power banks, there seems to be many metrics by which they can be judged and their performance determined. While I have focused on capacity as the first metric, and determined some power banks to be outright lies, other issues remain. One of them is the ripple voltages they develop, which can interfere with devices and stress/shorten their lifetimes.

Luckily for us, these tests are reasonably straightforward to perform. However, there are still other important metrics which can affect the usability of a power bank – for example, endurance to charge/discharge cycles and signalling methods to the connected devices.

In producing a newer, more versatile testing rig, I had inadvertently set the wheels in motion in regards to determining whether there were any significant endurance issues with the power banks already in my possession.

Unfortunately, in that article, it was determined that the Orzly Slim 4050mAh power bank was not performing correctly only after a very few number of cycles. Despite a glowing initial review, I was forced to withdraw my support for this product based upon the performance anomaly.

However, given that the power bank had some evidence of degradation, I wanted to further investigate this. It was at this time, a flood of power banks arrived for testing, which delayed this investigation. But now, I’ve managed to make it happen.

Motivation and Methodology

The motivation for this experiment was to try and catch degradation “in progress” for a lithium-polymer power bank. Another motivation was to confirm the degradation as real, as opposed to an anomaly. In order to do this, a large number of cycles would be required.

Unfortunately, this makes it prohibitive to test many power banks due to the manual labour required and time required to run the cycles. However, as the power bank was degrading, it meant that cycle times got shorter and it was possible to make a reasonable number of cycles.

The testing methodology utilized the new test rig, and ran for a total of fourty (40) cycles! Discharge was performed at the 1A rate to shorten test times, however, this results in slightly lower effective capacity readings compared to 500mA runs, thus the data for 500mA runs were excluded.

Results

Orzly-Endurance-Story

The data from the old rig when the power bank was new and the five runs at 1A has been plotted starting at -40. After that, the power bank was subjected to 500mA runs which aren’t comparable and approximately 10 runs usage. Then, the five runs tested with the new rig (when designed) was plotted starting at -20. In-between, another five runs at 500mA and about ten runs of usage occurred. The endurance investigation runs were run sequentially, one after another, and are plotted as numbers 0 to 39.

From the graph, it seems that the degradation was happening all along and manifested itself subtly even in the first five runs of the power bank – noting the last of the runs had a dip. The degrading capacity, first discovered when the new rig was built, is indeed real.

Since the endurance investigation began, it seems we have missed the bulk of the degradation, and it has slowed significantly, resulting in a residual capacity of about 750mAh (so about 1000mAh cell capacity from an initial 4050mAh).

The results, while in some way influenced with noise (as all real data is), seem clear enough to distinguish a clear degradation from when the investigation started, to the end, however, the degradation appears to have settled down. It does, however, provide me confidence in the consistency of the test rig results itself.

Brand name Lithium-ion and Lithium-polymer cells are typically rated for cycle lives of 300 to 500 cycles from 100% to 80% remaining capacity. By the looks of this graph, this cell was down to about 25% of its original capacity within 50 cycles!

Modifications and Further Testing

This particular power bank was actually quite a desirable one, as the circuitry turned in good results for ripple and excellent results for efficiency. It was desirable not to waste the circuit, and given I had spare lithium-ion cells lying around, I could probably re-use it.

By doing so, it would also allow us to determine whether the PCB was at fault (unlikely) and identify whether there were any issues. As a matter of fact, upon disassembling the power bank, it was observed that the battery was slightly swollen like a pillow, but still “airtight”. This suggests either a circuit which may be over-charging the cells (and thus extracting a little more capacity but damaging the cells in the process) or poor material quality or manufacturing practices which results in chemical reactions that can degrade the cell’s capacity and lifetime.

It was obvious to me, at least, that the cell itself had failed. It is important to remember that in all these tests, that statistics come into play. Testing very few units, in this case, one, doesn’t reliably predict what the other units may be like. It may be possible that my unit is an anomaly – however, I prefer to err on the side of caution and withdraw my recommendations upon seeing this happen.

Anyway, the unit was modified by removing the original lithium-polymer cell, and disposing of it. The front plastic cover was cut off to provide more space to mount new cells. In its place, a pair of two Samsung ICR18650-30B 3000mAh Lithium-Ion cells salvaged from the last damaged bank were installed. The power bank was then taped up with clear plastic tape, and then subjected to the regular power bank testing regime.

DSC_7977

After the modification, the capacity testing at 500mA and 1A resulted in capacities as follows:

Load (mA) Run Capacity (mAh)
500 1 4741.833212
500 2 4760.694378
500 3 4761.023768
500 4 4747.902328
500 5 4730.434977
Mean 4748.377733
Range 30.58879051
StDev 13.0059716
Load (mA) Run Capacity (mAh)
1000 1 4463.097205
1000 2 4385.292853
1000 3 4407.627859
1000 4 4391.637501
1000 5 4375.14465
Mean 4404.560014
Range 87.95255467
StDev 34.78323178

The efficiency based on assuming a source capacity of 6000mAh is 79% at 500mA and 73% at 1000mA. It is noted that in the initial review, based on the 4050mAh battery capacity, the originally determined efficiencies were 93.5% and 88.7%. The difference in results likely comes down to the fact that cells have a tolerance. A cell that’s rated for a given capacity can vary by as much as 100mAh per cell due to manufacturing variations and it also comes down to the charge voltage termination and discharge voltage termination. Some cells are designed for 4.3v charging, rather than 4.2v (i.e. the ICR18650-30B), and using the lower voltage can reduce the capacity (but increase the lifetime). Likewise, the ICR18650-30B is designed for a discharge down to 2.75v, whereas others may be terminated at 2.8-3.0v.

As a result, the efficiency is likely to be higher than calculated as the circuitry probably isn’t optimized to take the maximum advantage of the voltage range (2.75v to 4.3v) of the new Samsung ICR18650-30B cells.

Using the previous efficiencies, we can determine the usable capacity into the circuit of the cells. It comes out to be 2482mAh to 2539mAh when referencing the original batteries. However, even this calculation is flawed as it assumes the original Lithium-Polymer cell was 4050mAh new, at the loads we applied. As a result, the truth is likely that we initially over-estimated the efficiency as the original cell may have been over capacity, and the usable capacity of the 3000mAh cells is probably about 2700-2800mAh.

The unit retained its performance when it came to its ripple performance, charge and discharge termination. In fact, the range of the capacity results is 31mAh and 88mAh at the two rates, which is the best that I’ve seen in a while. I didn’t feel a need to post the output voltage profiles as they weren’t very interesting – they were as stable as before.

Conclusion

Some caution is definitely warranted when dealing with batteries from unknown manufacturers, as depending on the quality of materials and methods used in manufacturing, the lifetime of the batteries can vary quite significantly. In this case, this power bank was likely down to 80% of its original capacity in just 25 uses.

By performing modifications and testing with new cells, it was determined that the reduced capacity isn’t a result of a fault in the circuitry. However, we should acknowledge the crucial role of the circuitry, as it can directly have a bearing on the life of the cells. If it fails to terminate charge correctly, or allow over-discharge, then it is likely that the circuitry may cause damage to the cells.

As a result, analyzing the charge and discharge profile of the cells when used with the paired battery management system would be important to verify that the circuitry isn’t harming the cells themselves.

It seems that testing isn’t as simple as I thought … and yet others are publishing positive reviews based on “it works” …

Posted in Electronics, Power Bank | Tagged , , , , | Leave a comment