Memory Cards Test - my 7...

[Paul Iddon]

So today - I ran 7 MicroSD memory cards through Crystal Disk 7.0 to measure their read/write speeds. Each card was blank and has been formatted - and the USB card reader is the G4 Mobile Lite USB3 card reader through a USB3 port. The seven images below show the results. These cards get used in a variety of dashcams running 4K, 1440p, and 1080p. Obviously, the higher spec dashcams carry the faster cards when in the car and in use. These are the only cards I have - I haven't got any High Endurance or High Endurance Pro memory cards. Manufacturers claims - the Delkin would be the fastest I suspect but in my test it wasn't quite there - however for maximum speeds I would need a USB3.1 port and card reader (designated USB 3.1 Rev 2,” “USB 3.1 Gen 2) which would take advantage of their UHS-II architecture (they have 2 rows of contacts which theoretically would allow up to 250MB/s write speed). One day, I'll put one of these ports into my computer!

I hope this helps in some way for you all - and fair to say - every card I have works, though the slowest I only ever used in 1080p cams.
The price of these cards go from about £7 for the slower smaller capacity cards - up to around £150+ (and I have 2 of these buggers!!!) for the most expensive card, the Delkin Power.

I make no recommendations for which you should use - the manufacturers may or may not cover their cards for using in dashcams to due the constant writing.

I have added them from best write speed through the worst.

Paul.

 

[DT MI]

Do you have a dedicated SD slot on your computer? In the past I've done the same testing and found that even a USB3 adapter will test a bit slower (about 2% or so) than the same card being tested in a dedicated slot.

 

[Paul Iddon]

I do have a slot, but it doesn't work. The ports are set up this y by the builder of the PC.

Paul.

 

[SawMaster]

I'm looking at the two Samsung EVO's, and wondering why one writes at 89.97 and the other writes at 35.45? The only difference is card size, so does the file system (fat32 versus EXFAT) play that large a role here? It just doesn't make sense to me

Phil

 

[AverageOlderGamer]

I’m guessing one is the U3 and the other is U1? Does that U number change the speeds that much?

 

[Nigel]

I'm looking at the two Samsung EVO's, and wondering why one writes at 89.97 and the other writes at 35.45? The only difference is card size, so does the file system (fat32 versus EXFAT) play that large a role here? It just doesn't make sense to me

Phil

The 128GB card is quite likely made up of four 32GB memory blocks that can write simultaneously giving 4x the size and nearly 4x the speed. However that only works when you have sequential data to write, which is why on the random write test the smaller card is actually faster! We don't need the very high sequential write speeds for dashcams, the random write speed at the bottom of the write column is probably the most important figure for comparisons.

"(fat32 versus EXFAT)" - most of our cameras still won't use exFAT, so it doesn't make much difference. exFAT only does better than FAT once you have a large number of files on the card, which can happen when there are lots of small parking mode files. Dashcams should be using exFAT, it was designed for flash memory and should give longer life then FAT which was designed for rotating mechanical disks.

Does that U number change the speeds that much?

The U number is the version of the communications hardware+protocol between card and reader, U3 is faster than U1 which is faster than the original with no U number at all, but I don't think any dashcams are actually using the UHS bus yet, so it doesn't make any difference until you put the card into a USB3 UHS1/3 card reader and plug it into your computer.

The A number, as in "A1", is of interest since it guarantees a minimum performance. The A2 gives even better performance but again the dashcams don't have any hardware/firmware to support A2 so it doesn't help. An "A" number of 1 or greater is good to have, but many cards without it will still do fine.

 

[Paul Iddon]

My Delkin USH-II (has 2 rows of pins) offer no advantage in dashcams (atm).

I have now put into my PC a USB3.1 port, and with my USB3.1 card reader, it performs much faster. The card is meant to be capable of up to 300MB/s read and up to 250MB/s write speeds.

This is what I got tonight:

 

[Paul Iddon]

Quick update for the SanDisk High Endurance 128Gb card:

 

[Nigel]

Quick update for the SanDisk High Endurance 128Gb card:

Seems a bit slow considering we have had 2 years of development since your first post!

Well, they improved on the sequential reads/writes, but I assume that it is the RND4K numbers that are the important ones...

 

[Lothar]

Seems a bit slow considering we have had 2 years of development since your first post!

Well, they improved on the sequential reads/writes, but I assume that it is the RND4K numbers that are the important ones...

My understanding is that sequential write is more important than random write when it comes to dashcams?
For a phone or computer HDD/SSD, then yes random will be more important than sequential.

 

[Nigel]

My understanding is that sequential write is more important than random write when it comes to dashcams?
For a phone or computer HDD/SSD, then yes random will be more important than sequential.

I would expect it to be the other way around, but I don't have any evidence either way!

When I copy/play a file from the memory card onto my PC, I normally play/copy it sequentially, from start to finish. This seems fairly clear to me, although I suspect the dashcam video files on the card are stored in scattered blocks of physical memory, which maybe makes this a random operation?

When a dashcam writes video, it is going to write a block to the front file followed by a block to the rear file, followed by a block to the interior file. This does not appear to be sequential to me, it appears interleaved, which probably means it uses the random writes benchmark figure, but I don't really know how the multiple files get written to physical memory, possibly by the time they get to the physical card memory, they are sequential blocks. However, even if they are sequential, the dashcam needs to update the file access tables fairly regularly, and that will not be a sequential operation, and it also needs to delete blocks fairly regularly, and that is definitely not sequential operation.

 

[DT MI]

When a dashcam writes video, it is going to write a block to the front file followed by a block to the rear file, followed by a block to the interior file. This does not appear to be sequential to me, it appears interleaved, which probably means it uses the random writes,...

Irrespective of what's being written to the card the actual writing of the data is being done sequentially.

For simplicity's sake let's assume we're dealing with a newly formatted card with no data on it, and we're writing files as you described from a front, rear and interior camera. The first block from the front camera writes to location 1 on the card, first block from the rear camera to location 2 on the card, first block from the interior camera to location 3 on the card, second block from the front camera to location 4 on the card, etc. While the files from each camera are interleaved the totality of the data being physically written is being done sequentially.

 

[Nigel]

Irrespective of what's being written to the card the actual writing of the data is being done sequentially.

So the dashcam always writes and deletes data in amounts that are multiples of the erase block size?

While the RND figures in the screenshots above are only for operations which write and erase amounts of data smaller than the erase block size non-sequentially, something the dashcam never does, except maybe for the FAT, which doesn't matter since the FAT is not stored in normal memory?

For simplicity's sake let's assume we're dealing with a newly formatted card with no data on it,

I'd rather assume that the card has looped 100 times since the last format, and that 100 files have been left as read only thus not erased, since that would be a more realistic scenario for most dashcams!

 

[Lothar]

I would expect it to be the other way around, but I don't have any evidence either way!

When I copy/play a file from the memory card onto my PC, I normally play/copy it sequentially, from start to finish. This seems fairly clear to me, although I suspect the dashcam video files on the card are stored in scattered blocks of physical memory, which maybe makes this a random operation?

When a dashcam writes video, it is going to write a block to the front file followed by a block to the rear file, followed by a block to the interior file. This does not appear to be sequential to me, it appears interleaved, which probably means it uses the random writes benchmark figure, but I don't really know how the multiple files get written to physical memory, possibly by the time they get to the physical card memory, they are sequential blocks. However, even if they are sequential, the dashcam needs to update the file access tables fairly regularly, and that will not be a sequential operation, and it also needs to delete blocks fairly regularly, and that is definitely not sequential operation.

Yeah, I thought that way too but I was wrong.

Deciding on SD Size and Brand Option

Why, out of curiosity? It's $30 more for a slower U1 card vs U3 for less. What's the upside? The simple answer is that one is authorized and warrantied for dash cam use and the other is not. *Fine print on all Samsung Evo Select cards* "Warranty does not extend to dash cam, CCTV, surveillance...

dashcamtalk.com

Read my post above, plus the followup from @Deacon and even the legendary @jokiin himself, along with many others.

 

[DT MI]

So the dashcam always writes and deletes data in amounts that are multiples of the erase block size?

No, the dash cam will write and delete data in whatever increments it's programmed to do. The 'erase block size' can vary from card to card and the camera software/firmware has no knowledge of the specifics.

I'd rather assume that the card has looped 100 times since the last format, and that 100 files have been left as read only thus not erased, since that would be a more realistic scenario for most dashcams!

Irrelevant. My intentionally simplistic (for clarity) example was intended to demonstrate why your assertion that interleaved data streams from 3 different cameras (lenses) being recorded to the same card would result in random versus sequential writes was wrong. The camera puts together a 'hunk' of data and tells the card to write it after the previous 'hunk' it sent - that, by definition, is a sequential operation.

 

[Nigel]

that, by definition, is a sequential operation

And what is the definition of a random operation?

As I understand it, operations must always be sequential, it is the memory addresses of the reads and writes that are either random or sequential, where sequential means that the addresses of consecutive accesses are sequential addresses, and random means that the addresses jump all over the place. With interleaved files, a dashcam writing is going to better match the random definition, but I don't understand enough to know which of the benchmark figures are relevant to multi-channel dashcams.