FAQ FAQ  Forum Search   Events   Register Register  Login Login

Memory Card Benchmark Thread

Page  12>
Author
Kaishi View Drop Down
Senior Member
Senior Member

Joined: 28 December 2007
Country: United States
Location: United States
Status: Offline
Posts: 322
Post Options Post Options   Quote Kaishi Quote  Post ReplyReply Direct Link To This Post Topic: Memory Card Benchmark Thread
    Posted: 16 July 2014 at 15:55
There are tons of different memory card configurations available to our cameras, especially when considering adapters. Here's some of the results I've found, including the testing method I used to find them.

NOTES
I will include the name of the card listed exactly as written on the top, along with an average read and average write speed across the tested ranges. I'll put write speeds in red, and read speeds in green, to match ATTO's colors. I'll put a bold blue asterisk (*) next to any cards that are particularly good value for money. Any prices listed will be from time of writing (e.g. not what I originally paid, but what I would have to pay right now).

I'm including some very old media just for the sake of comparison. I'll try to mark these.

PLEASE FOLLOW THESE GUIDELINES AND INCLUDE YOUR RESULTS! THANKS!

TESTING METHOD
ATTO Disk Benchmark, v3.05 (latest at time of writing)
Transcend USB 3.0 Card Reader (connected to a USB 3.0 port)
Windows 10 x64
Transfer Size Range: 32-1024KB
Total Length: 64MB
Direct I/O: Overlapped I/O
Queue Depth: 4

RESULTS
COMPACT FLASH
pqi Hi-Speed 40 2GB CompactFlash: 6MB/s 8MB/s (really old, really awful)
Lexar High-Speed CompactFlash Write-Acceleration Technology 40x 1GB: 9MB/s 7MB/s (I find the accelerated writes adorable and uncommon)
SanDisk Extreme III 2.0GB CompactFlash: 19MB/s 19MB/s (card from 2007, but still pretty decent!)
Transcend CompactFlash 133x 16GB: 15MB/s 28MB/s (note, this is a really old card, but still pretty acceptable)
* Transcend CompactFlash UDMA7 400x 32GB: 40MB/s 83MB/s $32
Transcend CF170 Industrial 32GB CompactFlash: 50MB/s 95MB/s $50
Transcend 64GB Compactflash Card - TS64GCF1000 (160/120 MB/s): 130MB/s 130MB/s $135 (WOW this thing is fast, seriously. It smokes any other card I've tested)

SECURE DIGITAL
* PNY 64GB SDXC Elite Performance UHS-1 90MB/sec 65MB/s 90MB/s $35 (note this one hit its advertised performance! I've now bought 3 of these and they all work the same, peaking at 67MB/s write speeds. Fantastic.)
Micro Center 16GB SDHC Class 10 Flash Memory Card 12MB/s 75MB/s $6 (or free with a coupon). This is pretty impressive for reads but awful for writes.

MEMORY STICK
SANOXY MS Pro Duo ← MicroSDHC Dual Slot Adapter w/ 2x SanDisk Ultra 16GB MicroSDHC: 14MB/s 19MB/s $5+$11+$11=$27
PhotoFast CR-5400 Duo Adapter w/ 2x SanDisk Ultra 16GB MicroSDHC: 13MB/s 18MB/s $9+$11+11=$31
PhotoFast CR-5400 Duo Adapter w/ 2x Class 4 SDHC 8GB: 15MB/s 18MB/s

REFERENCE COMPARISONS (HDD & SSD)
2x Western Digital Velociraptor 600GB, RAID-0 200MB/s 250MB/s
2x Crucial B100 500GB, RAID-0 800MB/s 900+MB/s (this is basically infinite compared to what other devices can offer. Yes, SSDs can be even faster, but it's a whole different order of magnitude.)

RECOMMENDATIONS
NAND storage has wildly improved in the past few years. If you are using legacy memory cards, anything rated less than 133x, you should probably invest in something newer. For MemoryStick, consider adapters or the HX-series.

Take your RAW file size (10-40MB), and multiply that by your max framerate (1.5-10 FPS) to determine just how much write speed you need to achieve a buffer that should never overflow. If your camera ever stops letting you shoot while it writes to the storage, your storage isn't fast enough.

Edited by Kaishi - 19 May 2016 at 19:47
α900/VG, α77ii/VG, A7rII/VG, AF 28-70 F2.8 G, 16-50 F2.8 SSM, Σ 105 F2.8 EX DG OS HSM MACRO, SP 150-600 F5-6.3 Di USD;Gallery
 



Back to Top
Kaishi View Drop Down
Senior Member
Senior Member

Joined: 28 December 2007
Country: United States
Location: United States
Status: Offline
Posts: 322
Post Options Post Options   Quote Kaishi Quote  Post ReplyReply Direct Link To This Post Posted: 18 July 2014 at 23:03
I'm trying to keep this thread current with each of my test as I run it. I'd love to see contributions from you guys. Please post any test results you have.
α900/VG, α77ii/VG, A7rII/VG, AF 28-70 F2.8 G, 16-50 F2.8 SSM, Σ 105 F2.8 EX DG OS HSM MACRO, SP 150-600 F5-6.3 Di USD;Gallery
Back to Top
neilt3 View Drop Down
Senior Member
Senior Member

Joined: 13 September 2010
Country: United Kingdom
Location: Manchester.U.K
Status: Offline
Posts: 2840
Post Options Post Options   Quote neilt3 Quote  Post ReplyReply Direct Link To This Post Posted: 19 July 2014 at 00:53
Not shure what tests your looking to do , but for a lot of information that Bob Janes started a thread on , look here.
see my photostream on flickr;
http://www.flickr.com/photos/neilt3/
C & C welcome.
Back to Top
Kaishi View Drop Down
Senior Member
Senior Member

Joined: 28 December 2007
Country: United States
Location: United States
Status: Offline
Posts: 322
Post Options Post Options   Quote Kaishi Quote  Post ReplyReply Direct Link To This Post Posted: 21 July 2014 at 15:10
Yes, I saw that thread, but it's very model-specific. I'm trying to provide card-specific information instead. I wanted to make a thread which would come up immediately upon searching, so users would be able to see the data easily, without digging through pages.

I've tested all the decent cards I presently own. I have a couple old SD cards I might try, just for comparison. I'd love to see some data from other users, though.

ATTO Disk Benchmark is really easy to use, and is free. It only takes a minute or two to test a card. Just make sure that you're using a fast card reader on USB3 if you want to see performance numbers above 32MB/s (which is as much as USB2 can do).
α900/VG, α77ii/VG, A7rII/VG, AF 28-70 F2.8 G, 16-50 F2.8 SSM, Σ 105 F2.8 EX DG OS HSM MACRO, SP 150-600 F5-6.3 Di USD;Gallery
Back to Top
Kaishi View Drop Down
Senior Member
Senior Member

Joined: 28 December 2007
Country: United States
Location: United States
Status: Offline
Posts: 322
Post Options Post Options   Quote Kaishi Quote  Post ReplyReply Direct Link To This Post Posted: 23 July 2014 at 17:01
I've picked up a couple of SD cards I can test tonight, including a nice one:

* PNY 64GB SDXC Elite Performance UHS-1 90MB/sec $35
α900/VG, α77ii/VG, A7rII/VG, AF 28-70 F2.8 G, 16-50 F2.8 SSM, Σ 105 F2.8 EX DG OS HSM MACRO, SP 150-600 F5-6.3 Di USD;Gallery
Back to Top
addy landzaat View Drop Down
Senior Member
Senior Member

Joined: 22 April 2006
Country: Netherlands
Location: Netherlands
Status: Offline
Posts: 12441
Post Options Post Options   Quote addy landzaat Quote  Post ReplyReply Direct Link To This Post Posted: 23 July 2014 at 17:39
Yes, but this probably is card-reader/computer specific and these results might not reflect actual in-camera use. So, I doubt the reliability of these tests, unless a lot of people chime in and I doubt that will happen.

For me, this thread and the other thread are hardly useful, but I think some people will find it useful. So, keep calm and carry on
Why not follow me on Instagram? @Addy_101
 



Back to Top
analytical View Drop Down
Senior Member
Senior Member

Joined: 30 January 2007
Country: United States
Location: Texas
Status: Offline
Posts: 3069
Post Options Post Options   Quote analytical Quote  Post ReplyReply Direct Link To This Post Posted: 24 July 2014 at 18:37
Originally posted by Kaishi Kaishi wrote:


TESTING METHOD
Transfer Size Range: 32-1024KB
Why that range? No file produced by any Sony/Minolta SLR/SLT is as small even as 1024kb. (My old 6 MP Minolta 5D jpegs are at least twice that. Recent ARWs from 24MP cameras are more like 24,000kb.)

Including 32kb is irrelevant to these cameras and will lower the speeds measured in your results. With cards I have tested, 32kb is at least 30% less than 64kb, for some much more.    

For a more extreme example, a Transcend SDHC 32GB UHS-I at 32kb tested W/R at 8.5/18.7. But at 64kb it got 60.7/87.6 and best was at 2048kb with 62.7/90.1. (From 128kb to 8192kb essentially the same results.)   

For the cards I tested, including 64K would not make so much difference as 32kb, but it might on others. In any case considering file sizes that the camera produces, anything less than 1024kb makes no sense.

All this assumes that the point of testing is for
1. Camera writing files to card.
2. Unloading the card to computer using a high speed card reader.

Edited by analytical - 24 July 2014 at 18:52
12mi SW of Glen Rose,TX
photos
equipment
Back to Top
analytical View Drop Down
Senior Member
Senior Member

Joined: 30 January 2007
Country: United States
Location: Texas
Status: Offline
Posts: 3069
Post Options Post Options   Quote analytical Quote  Post ReplyReply Direct Link To This Post Posted: 24 July 2014 at 18:49
Originally posted by addy landzaat addy landzaat wrote:

Yes, but this probably is card-reader/computer specific and these results might not reflect actual in-camera use. So, I doubt the reliability of these tests, unless a lot of people chime in and I doubt that will happen.

For me, this thread and the other thread are hardly useful, but I think some people will find it useful. So, keep calm and carry on
It is certainly relevant for the read side unloading card to computer on fast card reader - which for me is the the main benefit of faster cards.

For the write side, I agree that relevance to camera performance is less clear. Once the card can keep up with the camera write rate, a faster write rate card will not help. Camera vs. camera, the file size and in camera buffer size will result in different performance from the same card.       
12mi SW of Glen Rose,TX
photos
equipment
Back to Top
Kaishi View Drop Down
Senior Member
Senior Member

Joined: 28 December 2007
Country: United States
Location: United States
Status: Offline
Posts: 322
Post Options Post Options   Quote Kaishi Quote  Post ReplyReply Direct Link To This Post Posted: 25 July 2014 at 14:07
Originally posted by analytical analytical wrote:

Why that range? No file produced by any Sony/Minolta SLR/SLT is as small even as 1024kb. (My old 6 MP Minolta 5D jpegs are at least twice that. Recent ARWs from 24MP cameras are more like 24,000kb.)

Let me clarify the difference between the write size and the file size. Write size is more a measure of bandwidth, and the efficiency of switching operations. Smaller writes are more taxing as they create more overhead. They indicate the performance of the underlying NAND rather than the complete memory card package. A good card will have very similar performance characteristics across all the tested transfer sizes, perhaps slightly slower at the low end and slightly faster at the high end, but not a huge range between them.

The cameras internally may write to the memory card in varying sizes, which isn't exposed to the user in any way. We don't know if they write only complete files, or in segments. It is possible that when taking different exposure lengths, the characteristics of a given camera's write operations change. Also, video recording is universally small repeated writes to disk rather than as a complete file, but how much at a time?

Additionally, above a transfer size of 1024KB, almost all storage media will exhibit the same characteristics: my cards perform the same at 1024KB as they do at 2048KB, 4096KB, and 8192KB, with only the most subtle variation in speeds which we can drop as a rounding error. Testing below 32KB causes extremely large penalties to any storage media as the filesystem overhead is usually greater than the actual writes themselves. It would artificially skew results in the same way as testing the upper range.

The "Total Length" being 64MB is to replicate the on-disk sizes you'd expect to see, when writing RAW+JPG or TIFF, or similar. It should be a little larger than the largest expected file size.

Originally posted by analytical analytical wrote:

For a more extreme example, a Transcend SDHC 32GB UHS-I at 32kb tested W/R at 8.5/18.7. But at 64kb it got 60.7/87.6 and best was at 2048kb with 62.7/90.1. (From 128kb to 8192kb essentially the same results.)

For the cards I tested, including 64K would not make so much difference as 32kb, but it might on others. In any case considering file sizes that the camera produces, anything less than 1024kb makes no sense.

All this assumes that the point of testing is for
1. Camera writing files to card.
2. Unloading the card to computer using a high speed card reader.

Thanks for the data! I'd like to include your findings, can you give me any more information?

The performance characteristics (namely the variance) indicates an issue with the filesystem on the card not necessarily being optimized for the NAND and the controllers on the card. If the filesystem isn't aligned to the NAND, writing different sizes can cause weird performance drops.

Your two assumptions are correct, though, we should be optimizing testing to replicate camera performance, as well as card-reader transfers to a computer. Both of these tasks matter, in my opinion. Most of the cameras out there only do UDMA5, but it is ideal to have the bottleneck be the camera and not the NAND, so that the buffers will flush unrestricted.

Edited by Kaishi - 25 July 2014 at 14:21
α900/VG, α77ii/VG, A7rII/VG, AF 28-70 F2.8 G, 16-50 F2.8 SSM, Σ 105 F2.8 EX DG OS HSM MACRO, SP 150-600 F5-6.3 Di USD;Gallery
Back to Top
Kaishi View Drop Down
Senior Member
Senior Member

Joined: 28 December 2007
Country: United States
Location: United States
Status: Offline
Posts: 322
Post Options Post Options   Quote Kaishi Quote  Post ReplyReply Direct Link To This Post Posted: 25 July 2014 at 14:15
Note: all this really only helps if we're using very fast card readers over USB3. If we test over USB2, the fastest we can possibly see is ~31MB/s R/W. USB3 raises this cap to ~375MB/s, well above the fastest cards that will be available for a couple years.

The point of this thread is to demonstrate that decade-old cards don't match modern cards in performance metrics. New cards do better at small writes, large writes, and everything in between. They allow the cameras to flush their buffers very quickly.

The biggest flaw in the results will be that we're only testing one instance/copy of a given model of card in most cases. The manufacturers are not obligated to use the same NAND for each copy of a card, obviously, so the results may vary considerably from batch to batch. However, some card manufacturers fab their own NAND, and others source it externally. Lexar, for example, is owned by Micron, who also owns Crucial. They are a major NAND and D-RAM manufacturer, who supply chips to many other companies (dell, apple, ASUS, etc.) Samsung makes a very large amount of the NAND found in memory cards, as well.
α900/VG, α77ii/VG, A7rII/VG, AF 28-70 F2.8 G, 16-50 F2.8 SSM, Σ 105 F2.8 EX DG OS HSM MACRO, SP 150-600 F5-6.3 Di USD;Gallery
Back to Top
Kaishi View Drop Down
Senior Member
Senior Member

Joined: 28 December 2007
Country: United States
Location: United States
Status: Offline
Posts: 322
Post Options Post Options   Quote Kaishi Quote  Post ReplyReply Direct Link To This Post Posted: 25 July 2014 at 14:50
Another note on card design:

Cards that use more NAND chips in parallel will both last longer, and achieve greater performance. This comes at the cost of added complexity, miniaturization, etc. which raises their price to assemble, and ultimately their retail price.

NAND comes in 2 flavors: SLC and MLC. I'm not going to get too technical on the difference, but suffice to say, SLC is faster and much more durable. MLC is slower and less durable, but yields double the capacity (per chip) of SLC. Most memory cards are MLC, unless they advertise being SLC (this is very rare, and very cool).

Industrial-grade cards have much faster and smarter chipsets/controllers. package along with the NAND. SLC cards are usually intended for industrial usage and will have similarly powerful chipsets.

CompactFlash is able to have such "staying power" because it is very flexible in its design. Each card has to have a dedicated controller on board (meaning the memory card isn't just a bunch of NAND stuffed in a square of metal and PCB). These controllers are about as good as the ones you'd find in a commercial SSD or HDD, and are tuned for the NAND used in the card. Cards have extra pages, a controller that can permanently mark a page as "bad" and replace it with one of the extras (just like an HDD or SSD), and control which blocks are next to be erased.

SD is much more simple, basically just a pile of NAND all crammed into a package. It uses a serial 4-bit bus. This makes it cheap and easy to design devices, but, SD relies on the device to manage the NAND. Most devices don't bother with advanced NAND management, so SD cards are pretty easy to corrupt or otherwise wreck the NAND. Most of these failures are recoverable, even if it seems like the card is just dead, not being recognized at all.

MemoryStick isn't all that different from SD, other than Sony and SanDisk agreeing on some extra DRM layers that are optional ("MagicGate" as an example). It uses a parallel 8-bit bus. That's why it's possible to stack a pair of microSD cards with a small controller chipset and get a reasonable result (4-bit serial + 4-bit serial + tiny RAID0 controller = 8-bit parallel)
α900/VG, α77ii/VG, A7rII/VG, AF 28-70 F2.8 G, 16-50 F2.8 SSM, Σ 105 F2.8 EX DG OS HSM MACRO, SP 150-600 F5-6.3 Di USD;Gallery
Back to Top
analytical View Drop Down
Senior Member
Senior Member

Joined: 30 January 2007
Country: United States
Location: Texas
Status: Offline
Posts: 3069
Post Options Post Options   Quote analytical Quote  Post ReplyReply Direct Link To This Post Posted: 25 July 2014 at 22:25
Whatever Bench32 settings actually mean and whatever it actually does, there is no question that transferring many small files is slower than transferring larger files, even though the total size is the same. And large files will transfer from card to computer at the same maximum rate that Bench32 measures. And typically this is the specification read rate of the card.

It is true we don't have specifications for how the camera writes files, but it seems pretty likely that for stills that they write whole files. Since all bodies can buffer many images it would make no sense to waste time with partials and reprocessing and re-compressing many revisions. Pretty much the purpose of the fast buffer.      

As you suggest the only certain data for write performance would have to come from running tests in camera, for each model, for each image type and quality setting. Seems like there was a thread/db for that at one time.
12mi SW of Glen Rose,TX
photos
equipment
Back to Top
Kaishi View Drop Down
Senior Member
Senior Member

Joined: 28 December 2007
Country: United States
Location: United States
Status: Offline
Posts: 322
Post Options Post Options   Quote Kaishi Quote  Post ReplyReply Direct Link To This Post Posted: 29 July 2014 at 20:02
Originally posted by analytical analytical wrote:

Whatever Bench32 settings actually mean and whatever it actually does, there is no question that transferring many small files is slower than transferring larger files, even though the total size is the same. And large files will transfer from card to computer at the same maximum rate that Bench32 measures. And typically this is the specification read rate of the card.

It is true we don't have specifications for how the camera writes files, but it seems pretty likely that for stills that they write whole files. Since all bodies can buffer many images it would make no sense to waste time with partials and reprocessing and re-compressing many revisions. Pretty much the purpose of the fast buffer.      

As you suggest the only certain data for write performance would have to come from running tests in camera, for each model, for each image type and quality setting. Seems like there was a thread/db for that at one time.


Let me rephrase what I meant about how the cameras write to the cards.

disk operations consist of filesystem-level commands being passed to a controller of some kind, and the controller interpreting those commands to produce block-level operations, which ultimately are executed against the NAND. Depending on the interface, the actual level of control that the camera has over the NAND will vary. Also, different protocols: UMDA allows a host(camera/computer) to have lower-level control over the NAND. So depending on the controller configuration, firmware, protocol support, and filesystem, the read and write operations may execute in different sizes.


It is extremely rare to simply pass a single 24+MB file from a cache directly to NAND. Normally there is a degree of hashing, metadata generation, block-level writing, verification, and buffer clearing that has to come along with that operation. So at a high level, yes, the camera wants to simply write a frame from the buffer to the storage, but lots of other steps need to occur before, during, and after that write operation. Any of those may vary in size. The block-level commands passed could be in 512KB chunks but at an extremely high rate, or could be in 2MB chunks. There's really no way to know for sure.

Hence the wide range of sampled rates. We want to provide data that will reflect real-world performance based on real-world unknown (and likely untestable) variables that necessarily have to exist.
α900/VG, α77ii/VG, A7rII/VG, AF 28-70 F2.8 G, 16-50 F2.8 SSM, Σ 105 F2.8 EX DG OS HSM MACRO, SP 150-600 F5-6.3 Di USD;Gallery
Back to Top
Idyllic Pics View Drop Down
Senior Member
Senior Member

Joined: 07 May 2014
Country: Australia
Location: Melbourne
Status: Offline
Posts: 415
Post Options Post Options   Quote Idyllic Pics Quote  Post ReplyReply Direct Link To This Post Posted: 28 November 2014 at 03:37
Great thread.

Any thoughts or experience in using 2 x 32GB micro SD cards in a CR-5400 dual MS Duo Pro adaptor?

I have seen people posting videos of this combination and using it successfully in PSP - curious if anyone has done so with a camera, like an A850 or A900?
Back to Top
Dyxum main page >  Forum Home > Equipment forums > Camera Talk Page  12>

Forum Jump Forum Permissions View Drop Down



This page was generated in 0.078 seconds.

Monitor calibration strip

Dyxum.com - Home of the alpha system photographer

In memory of Cameron Hill - brettania

Feel free to contact us if needed.