snapsy wrote:
Nice work Dave! Just an FYI, you can automate the process of calculating buffer depths, clearing times, and card throughput by utilizing the sub-second timestamp within the EXIF of images. For example, in exiftool you can specify the -SubSecDateTimeOriginal tag to extract only that field, along with -q to suppress the banner. You can run it against a directory of files and ingest the output into a spreadsheet to automatically calculate the intershot times, from which you can calculate the buffer depths and clearing times.
Here is an analysis I did of XQD vs CFE on the Nikon Z cameras which used that technique. That conclusion post also describes the process of how to determine where the bottleneck is, between the image processor (Expeed on Nikon, BIonz Z on Sony) and the card throughput. One technique I used was to use alternate file sizes, in particular TIFF output which the Z's support - you can alter the processing time vs output size ratio to determine if the bottleneck (ie, a TIFF is much larger but shares almost the same imaging pipeline, so if the total throughout of TIFFs is higher (sustained frame rate x average file size) than raw/jpg then that indicates the bottleneck is in the image processor. You can view my post here on automating the collection of file sizes for the total throughput calculations....Show more →
Thanks for the information. I took a quick look but it is pretty long. I will take a more careful look when I have some time to look at it carefully. Some automation would be helpful.
snapsy wrote:
Nice work Dave! Just an FYI, you can automate the process of calculating buffer depths, clearing times, and card throughput by utilizing the sub-second timestamp within the EXIF of images. For example, in exiftool you can specify the -SubSecDateTimeOriginal tag to extract only that field, along with -q to suppress the banner. You can run it against a directory of files and ingest the output into a spreadsheet to automatically calculate the intershot times, from which you can calculate the buffer depths and clearing times.
Here is an analysis I did of XQD vs CFE on the Nikon Z cameras which used that technique. That conclusion post also describes the process of how to determine where the bottleneck is, between the image processor (Expeed on Nikon, BIonz Z on Sony) and the card throughput. One technique I used was to use alternate file sizes, in particular TIFF output which the Z's support - you can alter the processing time vs output size ratio to determine if the bottleneck (ie, a TIFF is much larger but shares almost the same imaging pipeline, so if the total throughout of TIFFs is higher (sustained frame rate x average file size) than raw/jpg then that indicates the bottleneck is in the image processor. You can view my post here on automating the collection of file sizes for the total throughput calculations....Show more →
Thanks @snapsy@! Using your method to check the delay between new images creation when the buffer is full I see some fluctuation (0,12s, 0,22s, 0,32s delta values with the mean being 0,2s). I need to run this a couple more times but it seems like a fast way to test memory card write speed in camera.
j4nu wrote:
Thanks @snapsy@@! Using your method to check the delay between new images creation when the buffer is full I see some fluctuation (0,12s, 0,22s, 0,32s delta values with the mean being 0,2s). I need to run this a couple more times but it seems like a fast way to test memory card write speed in camera.
Thanks. This probably wont be all that helpful but here is a link to the Excel spreadsheet I used to organized and graph all the data - I'll keep it up long enough for you and @dclark to download:
I'm happy to say that after running a few more bursts, my cards seem to maintain about 5 fps when clearing the buffer (based on EXIF data). With the average file being about 51.3 mb this gives around 260 MB/sec, which is the rated write speed of the card (though tests in card reader show a higher sequential write value of around 270 MB/sec).
I did run into one "glitch" tough, where one of the cards slowed down to around 85 MB/sec (looks like UHS-I speed, so maybe not all contacts were aligned). Not sure why, but it was like this even before I started burst testing. Swapping the cards in the slots and returning them to their original slots helped (formatting alone did not), but I'm still not sure what caused it. The cards are Kingston Canvas React Plus 256 and 128 gb (they come with a fast usb card reader, yay).
I need to try to enter the data into Snapsy's xls for a better presentation ...
...