I ran some tests to try and determine how fast SD cards operate in the GH1 the way it uses them.
I tested a Transcend class 10 card claiming 20MB/s speed. When I test it as a physical device using 512k blocks it comes very close to the claimed speed (20MB/s) when reading - 19.8MB/s. Writes are slower - about 17MB/sec. This is not how the GH1 uses the devices, however. The GH1 uses a cluster size (for these cards at least) of 32K. If I test the card as a logical device using 32K blocks the read speed becomes 17.3MB/s and the write speed becomes 14.5MB/s. That means the write speed I can count on is 116Mb/s (bits, not bytes) at best. This is much better performance than is required for class 10 (which requires 80Mb/s), but somewhat less than they claim.
Class 6 cards from Transcend claim 6MB/s speed. Using the above test method the read speed becomes 15MB/s and the write becomes 8.8MB/s resulting in a maximum bitrate of 70.4 Mb/s. I know from testing that I can cause occasional failures with these cards using a 60Mb/s peak data rate ("C" settings), so I know I am not always getting the maximum 70.4Mb/s throughput. This is significantly better performance than Transcend claims.
If we factor in the whole "rogue frame" issue - which temporarily doubles bitrates - that means that if you have used the "C" settings (60Mb/s peak), rogue frames will cause occasional failures because 60Mb/s x 2 = 120Mb/sec, which is higher than the 116Mb/s best case maximum for the class 10 card. It seems that things aren't so mysterious after all.
My conclusion is that the Transcend class 6 cards can be counted on to reliably operate at 120% of their rated speed, and the Transcend class 10 cards can be relied upon at 70% of their claimed speed. Note that this factors in the camera's behavior, which significantly affects performance - so it isn't a comment on Transcend's marketing claims. Also, Transcend only claims 48Mb/s for class 6 but 160Mb/s for class 10. I think the big throughput claim in the class 10 card is for reads only - whereas conformance to a class requires that the lesser of read or write performance be used.
Chris
I tested a Transcend class 10 card claiming 20MB/s speed. When I test it as a physical device using 512k blocks it comes very close to the claimed speed (20MB/s) when reading - 19.8MB/s. Writes are slower - about 17MB/sec. This is not how the GH1 uses the devices, however. The GH1 uses a cluster size (for these cards at least) of 32K. If I test the card as a logical device using 32K blocks the read speed becomes 17.3MB/s and the write speed becomes 14.5MB/s. That means the write speed I can count on is 116Mb/s (bits, not bytes) at best. This is much better performance than is required for class 10 (which requires 80Mb/s), but somewhat less than they claim.
Class 6 cards from Transcend claim 6MB/s speed. Using the above test method the read speed becomes 15MB/s and the write becomes 8.8MB/s resulting in a maximum bitrate of 70.4 Mb/s. I know from testing that I can cause occasional failures with these cards using a 60Mb/s peak data rate ("C" settings), so I know I am not always getting the maximum 70.4Mb/s throughput. This is significantly better performance than Transcend claims.
If we factor in the whole "rogue frame" issue - which temporarily doubles bitrates - that means that if you have used the "C" settings (60Mb/s peak), rogue frames will cause occasional failures because 60Mb/s x 2 = 120Mb/sec, which is higher than the 116Mb/s best case maximum for the class 10 card. It seems that things aren't so mysterious after all.
My conclusion is that the Transcend class 6 cards can be counted on to reliably operate at 120% of their rated speed, and the Transcend class 10 cards can be relied upon at 70% of their claimed speed. Note that this factors in the camera's behavior, which significantly affects performance - so it isn't a comment on Transcend's marketing claims. Also, Transcend only claims 48Mb/s for class 6 but 160Mb/s for class 10. I think the big throughput claim in the class 10 card is for reads only - whereas conformance to a class requires that the lesser of read or write performance be used.
Chris
Last edited: