pecos wrote:
Close, except the lower frame rate is only for the remainder of the time not the full 30 seconds.
The fastest card (SanDisk Extreme Pro 95MB/s 32GB) averaged something like this:
36.7MB RAW (14-bit lossless compressed):
14 shots at ~6 fps for 2.33 seconds
52 shots at 1.88 fps for 27.67 seconds
Total: 66 shots in 30 seconds
I ran the test using the same settings but different test scene and the numbers increased due to smaller file size. Except this was a 15 second run.
29.5MB RAW (14-bit lossless compressed):
16 shots at ~6fps for 2.70 seconds
27 shots at 2.2 fps for 12.3 seconds
Total: 43 shots in 15 seconds; but it would be ~76 in 30 seconds.
So just changing the scene slightly is going to affect the results. SD cards are also not perfectly consistent in write speed. On longer runs this can cause a frame or two lower result....Show more →
Yes, things map out a bit nicer given the correction on the D600.
Your test may be unorthodox, but as someone who shoots wedding processions and reception entrances a 30second to 1 minute window is critical for me. It is one of the only times I will ever come close to hitting a buffer limit. Thanks for putting together the list.
as for the data...
Are the numbers you posted above what you measured via timestamps/video or is that just how things add up? Given the throughput numbers in the speed charts I would expect better performance after the buffer fills. You are counting the shooting time in the overall write time for the throughput test, correct?
mdbrown9999 wrote:
The Samsung Electronics 32GB PRO SDHC Upto 90MB/s Class 10 Memory Card (MB-SG32D/AM) was the best value at close to half the price of the sandisk. Thx for the effort, appreciate it.
I have a Samsung 32GB card that's rated for 80MB/s writes, and it keeps up properly with my Sandisk Extreme Pro 16GB, rated for 90MB/s writes itself, in my 6D. The 45MB/s reads Sandisk could not. And the Samsung was indeed cheaper.
But as the OP said in summary, 'just buy the Sandisk Extreme Pro', because you will be set with any current camera.
SloPhoto wrote:
Are the numbers you posted above what you measured via timestamps/video or is that just how things add up? Given the throughput numbers in the speed charts I would expect better performance after the buffer fills. You are counting the shooting time in the overall write time for the throughput test, correct?
Those are actual measurements. I use video to determine # of frames at full fps (from audio track) and write time by the card access light (recorded at 60fps so roughly accurate to +/-0.03 seconds). The continuous shooting test is the number of shots taken in 30 seconds (remote timer set for 30 seconds on continuous release). The camera continues writing after the 30 seconds to clear the buffer.
Maybe you are calculating differently but I'm not seeing why you expect better performance after the buffer fills. When I calculate the throughput on the continuous runs, the RAW write speed after the buffer fills are within 1% of the results of the write speed test. The JPEG are 92-99%, the lower write speed is due to inefficiency at writing more/smaller files (Except the Samsung EVO which is an odd duck which wrote faster with small files, it is extremely good at small file handling/random writes). Memory cards aren't optimized for handling random/small file writes compared with a solid state hard drive. They are most efficient at file size 64KB and larger. Cards that are optimized for the fastest speed are typically worse at small file handling.
pecos wrote:
Those are actual measurements. I use video to determine # of frames at full fps (from audio track) and write time by the card access light (recorded at 60fps so roughly accurate to +/-0.03 seconds). The continuous shooting test is the number of shots taken in 30 seconds (remote timer set for 30 seconds on continuous release). The camera continues writing after the 30 seconds to clear the buffer.
Maybe you are calculating differently but I'm not seeing why you expect better performance after the buffer fills. When I calculate the throughput on the continuous runs, the RAW write speed after the buffer fills are within 1% of the results of the write speed test. The JPEG are 92-99%, the lower write speed is due to inefficiency at writing more/smaller files (Except the Samsung EVO which is an odd duck which wrote faster with small files, it is extremely good at small file handling/random writes). Memory cards aren't optimized for handling random/small file writes compared with a solid state hard drive. They are most efficient at file size 64KB and larger. Cards that are optimized for the fastest speed are typically worse at small file handling....Show more →
I was using 35mb for the frame size. 1.88 fps at 35mb per frame is 65.8 mb/s which is already slightly less than average throughput. So I had some miscalculation there.
The place where I expect better performance is not only there though. It is that I would have expected it to take longer to fill the buffer. If you assume the buffer is 15 deep and you can write at 1.88 frames per second, I would expect to hit frame 17 before the buffer filled (over the ~2.3 seconds it takes to shoot 15 frames I would expect the camera to flush 2 full frames to disk). I phrased things poorly in my first response, sorry.
Possibly the buffer is not really 15 frames deep? Do you have an incredibly slow SD card that you could use as a true buffer depth baseline?
SloPhoto wrote:
I was using 35mb for the frame size. 1.88 fps at 35mb per frame is 65.8 mb/s which is already slightly less than average throughput. So I had some miscalculation there.
Yes, when you use 36.7MB file average it comes out a lot closer ~68.996MB/s write.
The place where I expect better performance is not only there though. It is that I would have expected it to take longer to fill the buffer. If you assume the buffer is 15 deep and you can write at 1.88 frames per second, I would expect to hit frame 17 before the buffer filled (over the ~2.3 seconds it takes to shoot 15 frames I would expect the camera to flush 2 full frames to disk). I phrased things poorly in my first response, sorry.
Possibly the buffer is not really 15 frames deep? Do you have an incredibly slow SD card that you could use as a true buffer depth baseline?
The buffer isn't a set number of frames deep it is a size in bytes. The amount of data allowed in the buffer changes depending on the image mode and other settings. For the D750 set to 14-bit lossless compressed I estimated the buffer size limit somewhere around 400MB. This means you can't take another shot until the buffer is less than that number. The buffer is actually larger, but the camera won't let you completely fill it. This is ~11 shots deep with ~37MB files. If you took pictures of a blank wall the buffer could fit more images in the buffer because they are smaller files (more compressible files = less bytes). If you shoot JPEG then the limit is higher, something like 475MB. Because the JPEG file is going to be smaller.
This is oversimplifying because I'm estimating buffer based on outflow. In continuous shooting the camera still has data in the pipe. I don't know what the actual internal size would be for the data as it comes off the sensor. That's really where the camera has a limit. If you shoot extended ISO and enable a bunch of features the buffer capacity (shots remaining) will drop. It needs room for the data it has yet to process.
pecos wrote:
Yes, when you use 36.7MB file average it comes out a lot closer ~68.996MB/s write.
The buffer isn't a set number of frames deep it is a size in bytes. The amount of data allowed in the buffer changes depending on the image mode and other settings. For the D750 set to 14-bit lossless compressed I estimated the buffer size limit somewhere around 400MB. This means you can't take another shot until the buffer is less than that number. The buffer is actually larger, but the camera won't let you completely fill it. This is ~11 shots deep with ~37MB files. If you took pictures of a blank wall the buffer could fit more images in the buffer because they are smaller files (more compressible files = less bytes). If you shoot JPEG then the limit is higher, something like 475MB. Because the JPEG file is going to be smaller.
This is oversimplifying because I'm estimating buffer based on outflow. In continuous shooting the camera still has data in the pipe. I don't know what the actual internal size would be for the data as it comes off the sensor. That's really where the camera has a limit. If you shoot extended ISO and enable a bunch of features the buffer capacity (shots remaining) will drop. It needs room for the data it has yet to process....Show more →
If you had an old card laying around that wrote at <5mb/s you could easily test buffer depth more conclusively.
Everyone keeps posting a buffer depth of 15 shots, but I am starting to agree with you. Nikon must be assuming a certain transfer speed during capture or utilizing a very optimistic compressed file size.
SloPhoto wrote:
If you had an old card laying around that wrote at <5mb/s you could easily test buffer depth more conclusively.
Everyone keeps posting a buffer depth of 15 shots, but I am starting to agree with you. Nikon must be assuming a certain transfer speed during capture or utilizing a very optimistic compressed file size.
I would think quoting a buffer depth requires several assumptions: file size (changes with image subject), camera settings, and card speed for the outflow.
Knowing card write speed in the camera is will tell you the outflow, and you know the buffer is considered full when it prevents another shot. The unknown of data in the pipe can be estimated. This is basically how I estimated the buffer size, averaging several cards.
I'll see if I can crunch some more numbers here and come up with something more exact for you.
You can tell something of a worst case from Nikon's r## number, which reads 12 when shooting RAW 14-bit lossless compressed.
SanDisk Extreme 16GB SDHC UHS-1 Flash Memory Card Speed Up To 45MB/s
In my d750, picked them up for about $15 each on Amazon, the only difference I noticed after switching to this card from my older slower card is when I review RAW image, they are pretty quick / no lag.
It's a little off topic, but when you tested the D800, which firmware version was in use? I've heard the latest firmware has caused issues with certain cards. I'm very intrigued by the Kingston cards as their performance seems to be within a whisker of the top flight Sandisk cards at around half of the cost.
gfinlayson wrote:
It's a little off topic, but when you tested the D800, which firmware version was in use? I've heard the latest firmware has caused issues with certain cards. I'm very intrigued by the Kingston cards as their performance seems to be within a whisker of the top flight Sandisk cards at around half of the cost.
The D800 test used the latest firmware A:1.10 / B: 1.10
I'm using Sandisk 32GB Xtreme Pro 95MB/S as primary and Sandisk 32GB Xtreme Pro 45MB/S as backup. I can definitely tell higher read/write speeds with the 95MB/S card.
