Why the Transcend TS128GUSD300S 300S was so bad in dashcams

[reverend]

I had a problem with a Transcend card I'd bought a few months ago in a Viofo A119 v3 and turned out it was the same card playing up in multiple Street Guardian cameras as advised by @jokiin and @OCD Tronic

I've done some digging this morning.

Basically for a dashcam you want a card that is able to exceed the video bitrate and get all the data to the card as fast as possible, otherwise you'll get problems like multiple frames dropped, missing video and general misbehaviour of the dashcam. Dashcams are fairly simple devices with low memory and so don't all have the advantage a PC running Windows, MacOS or Linux has where data is written to a very large buffer at high speed and then drained down to the slower storage beneath.

Some cards are able to run very quickly when doing simple sequential writes to free AU, but as soon as the card fills up and starts deleting old data and recycling things can slow down and cause problems at that point as it can't do a large sequential write but has to do more smaller chunks (IOPs).

Class 10 cards should be able to do 10MB/sec minimum (80 Megabits per second / Mbps) which is well above any non-4k dashcam.

I did some testing using h2testw which fills the card with data and then reads it all back to ensure it's valid.



Anyway, here is a graph showing the performance of a baseline card I use which has given me minimal issues - the Lexar 633x 64GB card from before they were sold off:



This graph shows write performance of a steady ~10.5MB/sec over a 90 minute period while the entire card was filled up and then a read performance of around 60MB/sec when the data was all read back.

This is how you'd normally expect microSD storage to work with a fairly flat performance baseline with slower writes than reads.

Now, here is the Transcend 128GB microSD UHS-I U3 A1 / TS128GUSD300S card formatted in the A119 v3.



At first glance it looks similar to the Lexar with similar performance - but check out the actual performance throughout:



There is an initial quick spike of high performance and then throughput plummets and takes over 90 minutes of continuous writes before it gets up to the 10MB/sec as called for by a Class 10 card - it's as though the controller is warming cache up or something bizarre which is all unlikely in a microSD card.

Lets zoom into the start a bit more



That shows the initial spike followed by a collapse in performance for over 90 minutes.

MB/sec and Megabits per second are very different and you'd normally divide megabits by 8 to get MegaBytes /sec - even with these figures above you can see the Transcend card just can't keep up with the bitrate of the Viofo A119 v3 camera in high bitrate mode.

I figured I'd give the card a chance and use the latest SD Formatter software to freshen it up - at this point the card was exFAT and the performance figures did change - here are the initial



Performance is again all over the place for the first 5 minutes - here's the full process.



Again the card takes around 20 minutes before it starts to perform more like you'd expect from a flash based product as though something needs time to warm up fully and it is not meeting class 10, never mind UHS-3 or V30 as labelled (30MB/sec sustained).

If the card is doing wear levelling maybe it's trying to do it during the initial power up but at least this indicates why it does not perform well in dashcams. I've not really paid much attention to microSD cards for the past couple of years and it may be that they don't do wear levelling like SSDs etc due to space constraints. It could even be trying to free some contiguous space up which is why the writes suffer.

I've logged an RMA for this particular card and will be interested to see if the replacement is the same (or whether I get a different card altogether).

If I was being paranoid I'd guess the burst of performance at the start is so that the card looks good in benchmarks such as CrystalDiskMark which generate a smaller subset of data initially and don't tend to last as long as something like h2testw but companies would never do that sort of thing would they

Other causes would be faulty NAND or controller code etc.

Some Ambarella cameras would be able to report slow memory speeds but with all the ways microSD cards can fail or misbehave there is no one size fits all solution - some cards go into read only mode and others will disappear - it's a real crap shoot what happens when you get a card failure.

All SD / microSD labelling should indicate the performace you can expect - more details here -https://www.sdcard.org/developers/overview/speed_class/



The Transcend card I notice does not have Class 10 markings, but does have U3 / V30 which should mean it can sustain a minimum of 30MB/sec writes.

All other cards I currently have here have the Class 10 marking as well as the newer markings other than this Transcend card so there's a chance they already knew there was an issue at time of production.

This post is mainly to highlight that cards do not always perform as you would expect, and to highlight one particular cause of issues.

Here's the blurb from the SD Card Association:

Best Combination between Speed Class Host and Card
Speed Class supported host can indicate Speed Class symbol somewhere on the product, package or manual. Consumers can find the best card for a host via Speed Class symbol match; choose the same or higher class symbol card than class symbol of the host indicated.

For example, if your host device requires a Speed Class 4 SD memory card, you can use Speed Class 4, 6 or 10 SD memory cards. If your host device requires a UHS Speed Class 1 SD memory card, you can use UHS Speed Class 1 or 3 SD memory cards. Video Speed Class is also the same. Note that expected write speed will not be available by a combination of different class symbols such as Class 10, U1 and V10 even those are indicated to the same 10MB/sec write speed.
Fragmentation and Speed
By repeating deletion and write of files, data area is gradually fragmented and it influences write speed. Generally, write speed to a fragmented area is slower than sequential write speed due to flash memory characteristics. In an era when memory capacity is not large enough, fragmented write needed to be considered. However, high capacity memory card is available at this time, Speed Class write is defined to perform sequential writes to a completely un-fragmented area (called "Free AU"). It makes Speed Class controls of host easy. On the other hand, even unused memory exists in total, there is a possibility that host cannot perform Speed Class recording. In that case, data arrangement to reduce fragmented area or move data to anther storage to re-format the card will be required. Video Speed Class supports "Suspend/Resume" function that can stop and retrieve sequential write. By using the function, it is possible to improve memory usage ratio considerably.

 

[Nigel]

I assume you used a card reader for that testing?

It would be interesting to test them using a dashcam as the card reader so that you are using the old "Class N" microSD interface rather than the newer UHS bus. The performance of the card could be quite different...

The U1/U3 specifications require buffering on the device, on the Class 6 specification the card has to keep up without buffering, not sure about Class 10, possibly your Transcend card doesn't have the Class 10 but does have the equivalent U1 because it can't keep up without the device providing buffering.

 

[Nigel]

Maybe we should only buy cards with a V10 or V30 spec? Then they should be able to cope with dashcam video!

 

[Rhody401]

Add the fact that many units now have dual cams writing at the same time, the requirements go up. (not illustrated on chart)

I've had no problems with Sandisk Ultra A1 256gb and 64gb chips in our vehicles. They have dual cams but only at hd. (A129)

Dual 4k cameras simultaneously writing to one card are here and new, too.

 

[DT MI]

...

Anyway, here is a graph showing the performance of a baseline card I use which has given me minimal issues - the Lexar 633x 64GB card from before they were sold off:

View attachment 47751

...

What program did you use to benchmark the cards? I'm experiencing some intermittent anomalies since I started using a 512GB card that I never saw with smaller cards. I could be related to what you're describing.

 

[Nigel]

What program did you use to benchmark the cards? I'm experiencing some intermittent anomalies since I started using a 512GB card that I never saw with smaller cards. I could be related to what you're describing.

The post says that he used h2testw.exe, which I think is sensible since it writes in the same way as a camera would.

Your question should probably have been, how did he draw the graphs?...

Windows Performance Monitor by the look of them , (watching the disk transfers.)

 

[DT MI]

The post says that he used h2testw.exe, which I think is sensible since it writes in the same way as a camera would.

Your question should probably have been, how did he draw the graphs?...

Windows Performance Monitor by the look of them , (watching the disk transfers.)

Correct, performance over time is what I'm interested in looking at. Didn't even consider real-time monitoring using the OS tools.

 

[reverend]

I'm just trying it on a few cameras in USB reader mode - initial figures (very slow!) aren't even seeing Class 10 figures being hit though when used in a camera so not sure if the card reader mode throughput is being constrained by the SoC speed or USB speed.

The tests were done on the SD reader in my old school brick Panasonic Toughbook - I'd assumed wrongly that it was old enough to not be UHS compliant but seems that came out in 2010!

Yes it's all just h2testw and then Windows perfmon to grab the data with 1 second intervals used to try and capture the drops to 0 that I kept seeing interactively with the Transcend card

This Transcend card I have here although it doesn't have Class 10 written on it they do list it as Class 10 in the specs - the marketing departments are going nuts though with the different versions of the cards:


Seeing as the high endurance card has a V suffix maybe the S suffix stands for s**t in dashcams

Where has that info come from @Nigel about buffering on the UHS cards as they don't mandate it in the specs for sequential throughput? The UHS-1 cards should be the same as Class 10 cards at 10MB / 80Mbps and the UHS-3 cards higher at 30MB / 240Mbps, albeit with the extra pins and higher clock rates for the new much higher bus speeds to support the higher throughput.

The only bit I can see in the specs says that the host has to be able to cope with a short delay when the card throws a busy signal when updating multiple FAT entries of up to 100ms per file entry which would be the same for all cards.

That's a good point though as suggests a dual camera with two video streams (or more), creating a GPS file and a debug log is going to see slightly higher delays there and need to handle pauses at every interval on slower cards better than one with just a single stream purely due to the FAT update delays and even if the other data streams are all interleaved together to match the block size underneath and likewise is going to have a harder time again with 1 minute clip intervals over 3 minute intervals.

The Application Performance classes have started to mandate host buffering though with Application Class 2 with A2 needing write caching to meet it's IOPS performance target for phones etc.

Some newer higher end cards will have their own cache in the card itself but the host just has the option to enable or disable the cache in card, with the risk that a developer may have to start worrying about clearing two caches before a device powers down if the device and card both have their own.

You'd have thought V speed rating would be good, but from having a good read of the specs would only be met if the device end supports that particular protocol and sends the right messages to the controller with the data in the right block sizes matching the card, if not it should fall back to whatever the Class x labelling is. It does mean if there's a mismatch in the block size used by the controller to the card that performance and longevity of the card will fall off a cliff.

It's bonkers with all the new logos as they're going to run out of space to fit them all on the card - low voltage for example would be an LV logo as well. It certainly looks like Transcend are dropping logos even though the cards meet the basic requirements in their spec sheets (although as we've seen over the years plenty of products don't actually meet certain specs).

I guess it does say a lot when the High Endurance card they offer is just labelled as Class 10 and nothing fancier than that - I guess they realise that CCTV and dashcams that need high endurance are usually writing data at a lower. but pretty constant speed.

I haven't heard back from Transcend yet so hopefully will get something today but from the sounds of it the replacement will be the same from what Transcend advised @jokiin as these cards seem to have been destined for phones.

Roll on what the higher spec industrial cards like Delkin use with SMART type parameters exposed to the host so you can start to warn when a card will need replacing due to imminent failure (HD Sentinel supports reporting on quite a few industrial cards now).

 

[reverend]

Here's the reply from Transcend Support - they make it very clear at the support level not to use these in dashcams with a similar line to what they told @jokiin and @OCD Tronic

In the past TLC cards would at least work in a dashcam albeit with a shorter lifespan - this is the first time personally I've seen a product behave so randomly but could very well be a design / cost choice.

Amazon have accepted a return so I'll get a refund in this case as the replacement is highly likely to be the same issue.

 

[TonyM]

Here's the reply from Transcend Support - they make it very clear at the support level not to use these in dashcams with a similar line to what they told @jokiin and @OCD Tronic

It's good to see that they took your issue seriously and provided a decent response