[Hound Tooth]

That can't be true, since the camera can process images (at 7fps) faster than it can write out - this is true for most cameras, where the camera can achieve a higher write speed in JPEG only compared to RAW, so the write speed is constrained by how fast the camera can write out the raw buffer.

That makes no sense at all. The camera isn't processing images at 7fps, it's just evaluating the scene at 7fps. The stream of data from the sensor just gets piled (unprocessed) into the buffer until the buffer fills up. Once the first image has been streamed into the buffer, the so-called "PRIME II" engine then starts processing this data as fast as it can, which is not necessarily anywhere near the theoretical write speed of the SD bus. By your standards, the K-7 (which used the exact same buffer and processing engine) should've been able to maintain 3fps indefinitely since it generated smaller RAW files. It could not. Not even close.

I just benchmarked my SanDisk Extreme Pro card in a USB 2.0 SDHC card reader. Real life write speed was 33.0 MB/s. This card reader is very old, it was "new" back when SDHC was "new", one of the first SDHC-capable card readers on the market. That's a far, far cry from both the 25 MB/s you quote, or the 17.5 MB/s claimed by the K-5. Clearly the card is capable of a lot more speed, even without UHS-1. Therefore the bottleneck lies elsewhere.

By the way, have you tried this experiment with just JPG files? I get 33 beautiful black pictures before the buffer fills up, totaling 129MB... and it takes 10 seconds to write them. That's barely 13 MB/s. Clearly the bottleneck is elsewhere inside the camera. It's not SD card write speed.

[Hound Tooth]

Since we can assume the K10D has a slower processor than the K-5 (much slower fps, despite smaller image size) clearly the image processing speed is not a factor in the K-5 write performance.

That's an odd assumption. The K-7 and K-5 both have exactly the same processor specs. There's no reason to assume that either of them are vastly superior to the K20D or K10D. It's more likely that the K10D produces smaller images which require less processing, thereby freeing up the processing engine to push more data to the SD card.

[Hound Tooth]

More proof that the bottleneck is the image processing engine:

I just took 36 RAW+ pictures (72 total files) at 1.5fps (that's how many shots it took to fill up the buffer). These were REAL pictures, no lens cap. I was using REAL exposure settings taking actual pictures of stuff around me as I held the shutter down. The SD card light was on the whole time. The camera wrote 944 MB to the SD card in 40 seconds. That's 23.6 MB/s.

...and this is why your testing methodology and conclusions make no sense.

edit: oops, make that 39 seconds. I forgot the camera doesn't start writing until 1 second into the burst. So make that 24MB/s.

[Christine Tham]

That makes no sense at all. The camera isn't processing images at 7fps, it's just evaluating the scene at 7fps. The stream of data from the sensor just gets piled (unprocessed) into the buffer until the buffer fills up. Once the first image has been streamed into the buffer, the so-called "PRIME II" engine then starts processing this data as fast as it can, which is not necessarily anywhere near the theoretical write speed of the SD bus. By your standards, the K-7 (which used the exact same buffer and processing engine) should've been able to maintain 3fps indefinitely since it generated smaller RAW files. It could not. Not even close.

I just benchmarked my SanDisk Extreme Pro card in a USB 2.0 SDHC card reader. Real life write speed was 33.0 MB/s. This card reader is very old, it was "new" back when SDHC was "new", one of the first SDHC-capable card readers on the market. That's a far, far cry from both the 25 MB/s you quote, or the 17.5 MB/s claimed by the K-5. Clearly the card is capable of a lot more speed, even without UHS-1. Therefore the bottleneck lies elsewhere.

Actually, if you check the long thread on dpreview, the K-7 can more or less maintain 5 fps indefinitely in JPEG (1 star, smallest file size). Also that thread has a fairly extensive discussion of why neither the K-5 nor the K-7 is constrained by image processing speed when writing out raw - it's more about the buffer layout and whether the raw image is compressed or not. It's of course up to you whether to believe any of this or not - just saying that not everyone agrees with your opinion.

As for your claim of 33 MB/s, I suspect you are confusing between instantaneous write speed (which is affected by caching) and sustained write rate. The USB 2 bus is capable of sustained rates of only 24-25 MB/s - regardless of how fast the card can or cannot do. Rob Galbraith's extensive benchmarks of SDHC readers show that there is no combination of card reader or card that exceeds this speed.

[Christine Tham]

I just took 36 RAW+ pictures (72 total files) at 1.5fps (that's how many shots it took to fill up the buffer). These were REAL pictures, no lens cap. I was using REAL exposure settings taking actual pictures of stuff around me as I held the shutter down. The SD card light was on the whole time. The camera wrote 944 MB to the SD card in 40 seconds. That's 23.6 MB/s.

That's interesting.

When I tried to repeat your test, I get different results (I was shooting in raw mode, not raw+).

First of all, my buffer never filled up - I was able to maintain 1.5 fps into 72 shots and the camera was showing no signs of slowing down.

The camera wrote out 906 MB in 56 seconds - effective write rate of 16 MB/s - not much different from the 7fps scenario.

Testing in raw+ mode vs raw is an apples vs oranges comparison. I would anticipate that the camera would have a a faster effective write rate for raw+.

PS - I'm not sure why you make a big deal of your pictures being REAL and shot under REAL conditions. All my tests are done like that (it's a specific scene - my living room - under controlled lighting conditions) - I consider testing with black pictures to be not representative of actual shooting conditions.

[Hound Tooth]

As for your claim of 33 MB/s, I suspect you are confusing between instantaneous write speed (which is affected by caching) and sustained write rate. The USB 2 bus is capable of sustained rates of only 24-25 MB/s - regardless of how fast the card can or cannot do. Rob Galbraith's extensive benchmarks of SDHC readers show that there is no combination of card reader or card that exceeds this speed.

It was a 1GB write using dd from the command line and timed that way. There can be no caching, the SD card wasn't mounted at the time. This is as accurate and unbiased a speed measurement as you can get. The data doesn't even exist until it gets written to the card. SanDisk has been bragging about 30MB/s for a long time, even before the UHS-1 speed class existed. This mythical 25MB/s barrier you refer to was broken a long time ago.

Testing in raw+ mode vs raw is an apples vs oranges comparison. I would anticipate that the camera would have a a faster effective write rate for raw+.

That makes no sense at all. When shooting in RAW+, the camera processes the image ONCE and then saves it TWICE. The embedded preview in the RAW file and the JPG file are essentially the same picture, except one of them is compressed more heavily. Therefore, RAW+ increases the write load without increasing the image processing load. Based on that, RAW+ gives a more accurate representation of write speed.

PS - I'm not sure why you make a big deal of your pictures being REAL and shot under REAL conditions. All my tests are done like that (it's a specific scene - my living room - under controlled lighting conditions) - I consider testing with black pictures to be not representative of actual shooting conditions.

Because a lot of the so-called tests in this (and other) threads use the lens-cap method for getting some mythical write-speed. All this does is generate the smallest possible RAW file, which minimizes the amount of data actually being written to the card. It does nothing to maximize actual write throughput.

[Christine Tham]

SanDisk has been bragging about 30MB/s for a long time, even before the UHS-1 speed class existed. This mythical 25MB/s barrier you refer to was broken a long time ago.

Not according to Rob Galbraith:
Rob Galbraith DPI: CF/SD Performance Database

That makes no sense at all. When shooting in RAW+, the camera processes the image ONCE and then saves it TWICE.

Actually, that potentially makes it up to 2 times faster.

In a typical camera, I think the real time OS has several threads running. When shooting multiple images, one thread copies the sensor readout to a raw buffer, another thread converts to JPEG, a third thread writes raw buffer to card.

The sustained write rate for raw depends on how fast this third thread can convert raw data into blocks to send to card.

When shooting raw+, potentially both the JPEG and the raw write thread can write to the card simultaneously, hence improving throughput (still subject to the max transfer rate of 25 MB/s though).

You of course don't have to believe the above. But it does explain why I get 16 MB/s writing raw in 1.5 fps and you get 23 in raw+.

---

Last edited by Christine Tham; 03-21-2011 at 11:16 PM.

[ozlizard]

I think this thread re-defines measurebating to a whole new level!

[MRRiley]

Well I just ordered 3 of the 16MB cards (B&H $69). My feeling is, I want a card that is capable of taking everything my camera can dish out and I believe the Extreme Pro's will fit the bill. Especially after slogging thru this thread. LOL

Mike

[RonHendriks1966]

Because a lot of the so-called tests in this (and other) threads use the lens-cap method for getting some mythical write-speed. All this does is generate the smallest possible RAW file, which minimizes the amount of data actually being written to the card. It does nothing to maximize actual write throughput.

One difference in the original test is to fill the buffer, in taking 15 seconds of reallife pictures and time them and meassure them. Then take 60 seconds of pictures and again time them and messure them. The difference between the two would be the actual writespeed of the card with some degree of error.

Pictures to be taken: RAW at 1/250th sec and 7 fps at iso5000 so you create large files

Well I just ordered 3 of the 16MB cards B&H $69). My feeling is, I want a card that is capable of taking everything my camera can dish out and I believe the Extreme Pro's will fit the bill. Especially after slogging thru this thread. LOL

Mike

I do agree with you on this. The K-5 is not keeping up with this card, so this is the best choice for getting the most out off our K-5.


I started out with this test to find out wheater it was smart to upgrade from a classe 10 card or just to keep going on as I'm doing at the time.

[Christine Tham]

Speaking of card readers, does anyone know of a good UHS-I or SDXC card reader?

Preferably something that sits on a faster bus, like USB3 or eSATA?

[Chex]

Is the difference in speed between a DSHC class 10 and UHS-1 or SDXC for your card reader.. does it really translate into enough saved time to warrant one? Even a 64GB card can be loaded onto your PC during an "good" bathroom break...
I do get trying to increase the controls for burst shooting.. that would be really sweet. But I think we're SOL unless there's a firmware update allowing it. Too bad you couldn't "overclock" the prime II processor and see if bumping up the bus speed would help speed up its transfer rates.

[bickrs]

I'll probably answer my own question here ...

I'm currently using a Transcend 8Gb Class 10 in my K-5 (1.03, image corrections off). I'm noticing that when review cycling through shots, whether to "chimp" or just look when not actively shooting, that it is fractionally not as quick as my K200D, but enough for me to notice. I'm reluctant to test my old San Disk Class 6 as it's still in use in my K200D.

I know the filesize is a lot bigger, hence it may take longer to read/dump from the buffer cache, so my question is, has anyone else noticed similar and does/will an Extreme Pro card help?

Cheers
David

[RonHendriks1966]

Well I just ordered 3 of the 16MB cards (B&H $69). My feeling is, I want a card that is capable of taking everything my camera can dish out and I believe the Extreme Pro's will fit the bill. Especially after slogging thru this thread. LOL

Mike

Such a good point. Just ordered the 32GB version. So on monday or tuesday I will run my own test to convince myself.

[ozlizard]

I have a Bugatti Veyron but it's just not fast enough, what can I get that will go that little bit faster?