What is the write speed (to SD* cards)?

Nikon Rumors Forum » Nikon DSLR » [D7000]

What is the write speed (to SD* cards)?

(40 posts) (12 voices)

Tags:

12 Next »

rbid

preferred member
Joined: Jan '11
Posts: 344
offline

I'm looking to get this animal (Nikon D7000) while organizing my budget, I would like to know what is the write speed of the camera..

The motivation is the same as the one I had few years ago with my D70s that in the time I got it, the market was already populated with faster-than-the-camera CF cards (that in one way were overkill for the camera), and instead of getting one expensive overkill memory chip, I could get few for the same price that are more suitable cards for the camera.

What is the protocol the camera uses? UHS-I?.. (I do not think the camera is capable of UHS-II)

I guess that the "class" number represents a multiple of 8 Mb/s (1 MB/s) ?

Thanks in advance..

Posted 2 years ago #
TaoTeJared

preferred member
Joined: Apr '10
Posts: 2,422
offline

Search the site - there are a ton of threads on this.

BTW - "..."class" number represents a multiple of 8 Mb/s (1 MB/s)" NOPE.
UHS? - never heard of it.

Get a class 6 or above and top name brand and you will be good.

Posted 2 years ago #
jonnyapple

Goldfingers
Joined: May '09
Posts: 3,400
offline

Yes, there are two threads that have been posted to in the last day. Here's a link for you:
http://sportsphotoguy.com/nikon-d7000-raw-burst-test/
bottom line: about 25 MB/s for the cards he tested.

Posted 2 years ago #
SportsPhotoGuy

member
Joined: Sep '10
Posts: 10
offline

Speed classes defined:

http://www.sdcard.org/developers/tech/speed_class/

No UHS-II (yet)...jury's out as to whether UHS-I provides a useful speed difference in the D7000 until I've had a chance to test more of them.

And yes, Class 6 means sustained transfer rate of 6 megaBYTES/sec which is 48 megaBITS/sec...etc. But much like manufacturers claims of so many MB/sec, the class rating performance, while generally more conservative, doesn't necessarily bear out in a particular device or application. (As my tests, and those of many others like Rob Galbraith's - http://www.robgalbraith.com, bear out.)

Thanks for the referral jonnyapple...but I have to chuckle...you guys have "Goldmembers" here? ;-)

Posted 2 years ago #
heartyfisher

preferred member
Joined: Apr '09
Posts: 1,701
offline

LOL good ol GoldMember is special.. but I am preferred.. ;-)

Thanks for the great review!

I have 1 question though. IF you used Lossless COMPRESSED NEF which card would be optimal?
well maybe let me ask it another way.. what are the number of frames before it starts to slow?
I think you said it was about 11 frames for Uncompressed before it started to slow? and would it slow down at all if you used lossy compressed nef?

What are the File size differences between compressed and uncompressed nef?

Posted 2 years ago #
Paperman

preferred member
Joined: Jul '10
Posts: 297
offline

jonnyapple said:
Yes, there are two threads that have been posted to in the last day. Here's a link for you:
http://sportsphotoguy.com/nikon-d7000-raw-burst-test/
bottom line: about 25 MB/s for the cards he tested.

I'm no expert but the method he chose to determine the speed of the cards seems strange to me - apologies if that's the industry standard .

He takes 15 shots / 11 from buffer at full 6fps speed and the rest (4) as the camera clears the buffer slowly . He then calculates the average write speed. It would be different if he chose to shoot 20 shots , 25 shots, 30 shots etc ... The write speed he determines depends also on the clearance speed of the D7000 buffer .

I guess it should be OK to determine which card writes fastest/faster this way but the numbers/speeds indicated are meaningless ( unless 15 shots in a row is the accepted sports photography shooting standard )

Wouldn't it have made more sense to limit the test to the buffer amount ( 11 shots ) for all cards so he could time how long it takes to totally transfer the files to the card . It would have given the true maximum write speed of the card .

One more thing - we are all discussing the write speed of the cards . Does anyone know the write speed of the "camera " ? If we had an infinitely fast card , at what speed could the DSLR transfer the files ??

Posted 2 years ago #
heartyfisher

preferred member
Joined: Apr '09
Posts: 1,701
offline

i think his logic is sound. and it should have the same numbers for 15,20,25,30 frames. but I guess he needs to prove it by doing a test :-)

Posted 2 years ago #
Paperman

preferred member
Joined: Jul '10
Posts: 297
offline

heartyfisher said:
i think his logic is sound. and it should have the same numbers for 15,20,25,30 frames. but I guess he needs to prove it by doing a test :-)

How ?

15 shots - 11 shots in 2 seconds and let's say the last 4 in another 8 seconds . So it is like 16MBx15 = 240 MB / 10 sec = 24 MB/s ( +/- the figure he reached in the fastest card )

25 shots - 11 in 2 seconds , and probably the last 14 takes 28 seconds or so ( same rate as above when buffer is full ) . 25x16MB = 400 MB / 30 sec = 13 MB/s

???

Posted 2 years ago #
jonnyapple

Goldfingers
Joined: May '09
Posts: 3,400
offline

It is sound as long as the card write is the bottleneck in the process, Paperman. If you shoot 15 shots at 10MB/shot and you assume the camera has been writing the entire time from the first shot until the write light turns off, then the time between first shot and light off tells you how long it took the camera to write 150 MB to the card, regardless of whether the buffer filled in the process.

Even if the buffer fills completely, you get the same result as long as the bottleneck was always the card write. (I'm off to test it myself like a good scientist.)

Posted 2 years ago #
Paperman

preferred member
Joined: Jul '10
Posts: 297
offline

You're probably right . I wasn't aware he measured the time till the COMPLETION of writing to card /till he got a clean buffer . I was just assuming he measured the time he was able to shoot the 15 shots .

Now what about my second question - the write speed of the camera if there had been an infinitely fast card ??

Posted 2 years ago #
jonnyapple

Goldfingers
Joined: May '09
Posts: 3,400
offline

I agree with the "problem" of having an infinitely fast card (and interface). Then the buffer would never fill and you wouldn't be able to measure the card write speed with this method. In fact, you can't measure anything faster than 7 shots/s (105MB/s). I'll let you know when I have a card that fast. ;-)
EDIT: (I had my numbers above wrong—sorry.) I guess what you're saying is you can't tell if the card or the interface is the bottleneck for the fastest card you measure. I agree with that.

Results are in. I list them as number of shots, size of all shots, time to write, apparent write speed. The buffer was 10 shots, so I took two points before and after the buffer filling. Time precision is to about 0.5 seconds, so take the 2-shot burst with a grain of salt.

2, 29.4MB, 4s, 7.4MB/s
5, 75.1MB, 9s, 8.3MB/s
14, 209MB, 24s, 8.7MB/s
36, 536MB, 59s, 9.08MB/s

I suspect there's a systematic error in the finish time because I'm not certain the light off time is right when it's done writing. For example, subtracting 0.75 s from each of my reported times puts all four measured values (9.04, 9.10, 8.99, 9.20 MB/s) within 2% of each other.

EDIT: I should mention that this is a Transcend 16GB class 10 card, just so you can compare with his results for a similar card.
EDIT: I just noticed why you thought his numbers were bogus. His first number (effective frames per second) actually is bogus and depends on number of shots taken. He mentions that the second number (write speed) was calculated the way I described in my post above.

Posted 2 years ago #
Drab

preferred member
Joined: Aug '10
Posts: 540
offline

jonnyapple said:
I suspect there's a systematic error in the finish time because I'm not certain the light off time is right when it's done writing. For example, subtracting 0.75 s from each of my reported times puts all four measured values (9.04, 9.10, 8.99, 9.20 MB/s) within 2% of each other.

A very astute observation and one of the key classes of variables I haven't seen controlled for in most supposed card tests.
What sort of write speeds do you see on the card when using a PC interface (not the camera) and sequential writes?

Posted 2 years ago #
jonnyapple

Goldfingers
Joined: May '09
Posts: 3,400
offline

I got 13.8 MB/s transferring a 1 GB file to the same card with my card reader. For those who are counting, in both this and the previous numbers I reported 1073741824 Bytes—2^30—as 1 GB, not 1E9 Bytes and the same goes for MB above as 2^20 bytes.

Posted 2 years ago #
Paperman

preferred member
Joined: Jul '10
Posts: 297
offline

Thanks for all the explanations . It is interesting to see you are not getting the "promised" minimum 10 MB/s with your class 10 card .

By the way , my question was actually a question on its own ; I wasn't relating it to the test .

What is the write speed of a DSLR like D7000 if the card has infinite speed ? ( In other words , what should the limit be ( today ) for card producers if they don't want any redundant performance )

Posted 2 years ago #
heartyfisher

preferred member
Joined: Apr '09
Posts: 1,701
offline

interesting test and results JA. The key things I found interesting were.
1) Your class 10 can do almost 14MB/s
2) Yet in the camera its only getting to 9 MB/s
3) your 0.75 seconds on top is also interesting.

From this I conclude for best performance a class 10 or better is essential. However, because a class 10 can actually perform better than 10MP, for best value maybe a class 6 would be sufficient as if we can assume that it may have the same 40% performance above its rated MBPS it would only be a bit lower than the 9MBps, This being if you shoot Nef Uncompressed. If however(as I do) shoot Nef Compressed then a class 6 should be more than enough through put that it wont be the bottleneck. JA would you be so kind as to try one quick test with NEF Compressed?

Posted 2 years ago #
Drab

preferred member
Joined: Aug '10
Posts: 540
offline

heartyfisher said:for best value maybe a class 6 would be sufficient as if we can assume that it may have the same 40% performance above its rated MBPS

Unsafe conclusion.
Class 10 is the highest recognized non UHS speed class. Therefore 40% faster than 10 should be sold as 10s, while those 40% faster than 6 could be sold as 8s.

Posted 2 years ago #
jonnyapple

Goldfingers
Joined: May '09
Posts: 3,400
offline

Also, hearty, don't confuse write speed with file size. Remember that the D7000 can take 6 (or 7? I forget) frames per second so even with a file size of 10MB, that works out to one shot per second cleared from the buffer once the buffer fills.

Here are the results you wanted. The ones from above were taken in 14-bit lossless compressed—the highest raw quality settings on the D7000.

(the buffer was 11 shots for both of these)
14-bit compressed:
36 shots, 469MB, 54s, 8.7MB/s

12-bit compressed:
36 shots, 367MB, 42s, 8.7MB/s

Posted 2 years ago #
jonnyapple

Goldfingers
Joined: May '09
Posts: 3,400
offline

Hearty, if you're interested, I think I've worked out the total time you can shoot without filling the buffer. To start, you set the number of frames remaining in the buffer to zero:

-B - R*t + W*t/S = 0

Here, B is the buffer size in frames, R is the frame rate, W is the write speed in MB/s, and S is the image size in MB/frame. This equation is generally true for any camera, file type, and write speed.

Then, solving for t gives

t = -B/((W/S) - R)

If you prefer to know the number of consecutive shots that can be taken before the buffer fills, just multiply by the frame rate R

N = R*t = -B*R/((W/S) - R)

After the buffer fills, the frame rate is now limited by write speed and the equation isn't as useful anymore (well, it just says you can keep shooting at that rate for infinite time, which is true if you have a card with infinite capacity).

Rfull = W/S

Plugging in some numbers from the D7000 and my card: B=10f, W=9MB/s, S=15MB, R=6f/s I get...negative time: -1.9s. Brilliant. ;-) Does anyone see where I went wrong?
If I had a card with a 25MB/s write speed, the number would be 2.3s. Is it worth it? Probably to some people.

EDIT: I found it. How I interpreted B I needed a negative sign.

All right. Enough math—I'm off to have a swim.

Posted 2 years ago #
Drab

preferred member
Joined: Aug '10
Posts: 540
offline

And determining the buffer size would be best accomplished by shooting with the lowest jpg compression quality, smallest resolution, lens cap on, max shutter speed, minimum (largest number) aperture, lowest ISO, very slow SD card (of a known maximum write speed tested on a computer), all custom picture controls off.

This should give incredibly small jpegs (essentially uniformly black) which require very little CPU to compress, removing as many variables from the equation as possible except actual buffer size.

IF anyone cares. ;)

Posted 2 years ago #
rbid

preferred member
Joined: Jan '11
Posts: 344
offline

I guess that the bottle-neck is in the interface the camera has for writing the data to the SD card, therefore a faster card than that write-speed is not needed. and this is what I'm looking for :)

From other side, a faster card, has also a faster "read" speed, that is required when you download the photos to your computer (preferable with a good card reader and not via the camera)

Posted 2 years ago #
SportsPhotoGuy

member
Joined: Sep '10
Posts: 10
offline

Wow. Lots of discussion here!

My test results are described as "effective" rates - both in terms of frames per second and megabytes/second. Neither is intended as any sort of absolute measure, but a means of comparing cards under similar conditions.

There's really no such thing as a "standard" here. I probably never shoot 15, and only 15, frames in a row at top speed. I chose 15 because it is more than the buffer will hold at my selected settings (which are typical for many sports situations). It is not unusual for long bursts - or multiple shorter bursts fired in quick succession - to fill the buffer, so having an idea of how quickly the camera can clear the buffer to allow additional shots is paramount.

I'm actually measuring two different times here - the time to fire all 15 shots, and then the total time to write all of the shots to the card.

Ultimately, it's important to view the results in comparison to those of other cards, rather than as an absolute measure of what the camera or cards can do.

Posted 2 years ago #
jonnyapple

Goldfingers
Joined: May '09
Posts: 3,400
offline

Oh, I think your results are great, sportsphotoguy. And I wouldn't criticize you if you were 10% off in your results (prices fluctuate by more than 10% so write speed doesn't need to be known better than that, IMO). So thanks for doing the tests so that everyone including me can make more informed decisions when shopping for cards.

I think what threw paperman off is something that might confuse a lot of your readers—the "effective frame rate". That shouldn't even be included because it depends on how many shots you take. The write speed, which you also list, is the important number because it is not a (strong?) function of file type, frame rate, number of shots, or buffer size.

Re: my user role, it's the mods' retribution on me because I used to close all my posts saying, "Yeah, baby!"

Posted 2 years ago #
Drab

preferred member
Joined: Aug '10
Posts: 540
offline

SportsPhotoGuy said:
My test results are described as "effective" rates - both in terms of frames per second and megabytes/second. Neither is intended as any sort of absolute measure, but a means of comparing cards under similar conditions.

The following is likely reflective of a difference in our backgrounds, rather than a treatise on right vs wrong:

IF one were to determine the buffer size of a camera (expanding on my thoughts above) AND one were to determine the native write speed of a camera (possibly two different speeds on the D7000 due to the differing buses between SD and UHS) one could model card performance and accurately predict how fast any given card would be on the camera based on a simple write test performed with a card on a computer.

This would eliminate the vast majority of variables, tighten up the error bars, and allow one test (the card-on-computer one) to answer the question for all modeled cameras.

OR:
One could walk down the street, catch a show, and drink a few beers.
;)

EDIT:

jonnyapple said:
Re: my user role, it's the mods' retribution on me because I used to close all my posts saying, "Yeah, baby!"

Took me about 30 minutes to catch that.
;)

Posted 2 years ago #
heartyfisher

preferred member
Joined: Apr '09
Posts: 1,701
offline

jonnyapple said:

Re: my user role, it's the mods' retribution on me because I used to close all my posts saying, "Yeah, baby!"

LOL! really? @ SportsPhotoGuy : I am sure you believe him! heh!

@ JA : Thanks for the math.. (I checked out early).. Thanks for doing the test though. I am surprised that both took 11 frames to fill the buffer. It probably shows that the buffer is used only for PURE RAW before any compression occurs. It also probably means that clearing out the buffer is a lower priority task ie the clear buffer function is not run unless you are no longer pressing the shutter OR if it is full so it clears the one space and fill it again in the next shot. so from this we can predict the following.
1) Once the buffer is full, it will take the same amount of time to flush the buffer after you stop pressing the shutter release.
2) using higher compression rates should allow for higher FPS rates after the buffer is filled.
3) since initially the buffer is not cleared while the buffer is filling up(ie no write to SD occurring). It would probably account for the curious 0.75 sec you needed to add/subtract to your results to get a uniform write rate.

I would say not the best implementation for buffer clearing but good to know. looks like the cpu /process is single tasking. wonder if its the same for the other nikon cameras. I think I heard somewhere that the high end cameras have multiple cpu's but I guess they need to upgrade the OS to be multi threading :-) heh! If they open sourced the thing we could hack android into it huh !

Cool! this was fun.

Posted 2 years ago #
jonnyapple

Goldfingers
Joined: May '09
Posts: 3,400
offline

I think you checked out too early to see my bottom line, hearty, so your third conclusion isn't right. My fault for mistaking my audience (non-nerds), most likely—Drab is probably the only one who read it. ;-)

If you plug in the numbers, you get more than 11 frames before the buffer is full with the 10MB files and a fast card (it's the equation that starts N = above). For example, it works out to 18 or 19 before the buffer fills on 12-bit compressed with a 25MB/s card. Even my lowly 9MB/s card predicts almost 13 before the buffer fills at that file size. I actually get 14 at those settings before it slows down, so my model is too simple. The number of frames that fit in the buffer for each file size probably includes some wiggle room on how long it will take to process the files, and it's probably worst case so people are only pleasantly surprised.

A couple of notes that most people wouldn't be interested in (the kind of thing I'd write in the margins of a lab notebook):
-Once the buffer fills, it's possible to find the time to process and write each file by pressing the shutter button halfway and watching the remaining frames in the buffer tick back up to full (r00 to r11 or whatever).
-At 14-bit lossless compressed, the buffer clearing is not at a uniform rate. Two frames get freed up in quick succession and then there's a pause before the next set of two are freed up—I estimate about 0.5 s between the close frames and then 2 s between sets. This can be confirmed using the r00 number mentioned above, as well.

Posted 2 years ago #