Some CF Express Card Tests

tiggy@mac.com

Pentax K-1000
Jan 20, 2014
603
410
Thetford, VT
www.ForestMetrix.com
I purchased CF Express cards for the coming R5, and they arrived over the past few days. Three different types. I did the Black Magic Disk Speed Test to each and thought people might appreciate the results. This test does repeated read/writes to the card to simulate real world video file use.

Delkin Devices CF Express Power (512GB)
Read: 1450 MB/sec
Write: 1120 MB/sec

Sony Tough CF Express G (512GB)
Read: 1050 MB/sec
Write: 1050 MB/sec

ProGrade (Gold Bar 256GB)
Read: 1250 MB/sec
Write: 950 MB/sec

The tests were performed about 10 times each. Card reader was a Sonnet reader with Thunderbolt 3.

The Sony card was abysmal, as it was the most expensive and claimed to be the fastest (1700/1480). The Delkin claimed 1730/1430 and wound up being the actual fastest by about 10 percent. The ProGrade with gold stripe is supposed to be 1600/1000. I have a ProGrade with black stripe (claimed 1600/1400) coming in and will add those results when it's tested.
 

tiggy@mac.com

Pentax K-1000
Jan 20, 2014
603
410
Thetford, VT
www.ForestMetrix.com
Thanks for sharing.
You're very welcome. Just a note: the implied maximum post-buffer rate of fire for the R5 would seem to be more than the 20 frames per second the camera is rated for. If we assume that a file is 55mb, you'd get 26 frames per second from the Delkin card, if it were laying those files down perfectly efficiently.

Skeptical of this, I took a couple of the cards and some very large database files on my computer and started throwing them around, making copies, transferring, etc. With this sort of informal test I came out with a roughly 300mb/sec rate. I suspect this'll be closer to actual camera performance in taking stills past the buffer. That would imply about a 6 fps rate of fire after the buffer, which isn't too too bad. So, using the fastest of these cards, your performance is likely to be somewhere between 6 fps after the buffer is full and no (card) limit whatsoever.

I don't believe we've been told how much buffer is in the camera yet. We can sort of deduce it a bit by looking at their estimates of how many frames you can take before the buffer is full, but those calculations by Canon have traditionally been very conservative, assuming average card speeds. With the R5, the most useful figures they've provided for these calculations are that at 20fps, you can get 83 full raws before the buffer is full. Of course, during those 4 seconds of firing, your camera is pulling from the cache and placing those files on the memory card. If the camera were doing this at 3 frames per second with an average card, it would imply that the buffer was actually about 70 raws big.

Extrapolating that to performance with the Delkin card, we might expect it (using conservative 300mb/s performance measured by moving files to/from card) to give us a cached performance of 100 frames rather than the 83 the camera is rated for:
FramesOffloadedTo Cache
Second 1
20​
6​
14​
Second 2
20​
6​
28​
Second 3
20​
6​
42​
Second 4
20​
6​
56​
Second 5
20​
6​
70​

If you shoot with the mechanical shutter at 12 fps, the extra time it takes to take those frames is well-used by the cache to off-load images, giving the camera a roughly 140 frames of cache - in the unlikely event all my assumptions are correct.

If you haven't rolled your eyes and stopped reading by this point, you might be interested in the calculation of just how fast a CF Express card would need to be in real world performance to effectively make the cache irrelevant. Assuming file size of 55mb, that would be throughput of 660 mb/sec for the mechanical shutter mode and 1100 mb/sec for the electronic shutter. The CF Express standard has a theoretical limit of 2000 mb/sec, and existing cards more often than not claim performance higher than the required figures. Actual performance as measured by the Black Magic tool shows some cards already exceed goal, and some don't in reality. My own file copying test shows that none of the cards are quite there. I suspect actual performance once we get an R5 to test will be somewhere between those two test ranges, making careful card choice among these options more important than has been traditionally the case.
 
  • Like
Reactions: HenryL and Click

tiggy@mac.com

Pentax K-1000
Jan 20, 2014
603
410
Thetford, VT
www.ForestMetrix.com
I purchased CF Express cards for the coming R5, and they arrived over the past few days. Three different types. I did the Black Magic Disk Speed Test to each and thought people might appreciate the results. This test does repeated read/writes to the card to simulate real world video file use.

Delkin Devices CF Express Power (512GB)
Read: 1450 MB/sec
Write: 1120 MB/sec

Sony Tough CF Express G (512GB)
Read: 1050 MB/sec
Write: 1050 MB/sec

ProGrade (Gold Bar 256GB)
Read: 1250 MB/sec
Write: 950 MB/sec

The tests were performed about 10 times each. Card reader was a Sonnet reader with Thunderbolt 3.

The Sony card was abysmal, as it was the most expensive and claimed to be the fastest (1700/1480). The Delkin claimed 1730/1430 and wound up being the actual fastest by about 10 percent. The ProGrade with gold stripe is supposed to be 1600/1000. I have a ProGrade with black stripe (claimed 1600/1400) coming in and will add those results when it's tested.

I got the ProGrade Onyx version in. The increased performance over the Gen 1 version of the ProGrade gold version was a bit disappointing...

ProGrade (Onyx [black/gray], 325GB)
Read: 1525 MB/sec
Write: 1020 MB/sec

That's a nice improvement on the read speed, but the all-important write speed was only 7 percent faster. The claimed speeds are 1600/1400, so the write speed is definitely suffering versus the claim.

The upshot is that the Delkin is - by far - the best performer, and the best $/GB. I'd love to try the Lexar and Sandisk versions just to be complete, but can't justify buying more cards. If someone else has those, I can show them how to conduct this test.

One interesting note. After doing so much burn-in testing of these cards, I'm noticing that they get very, very hot. Unlike CF cards or SD cards. There's just a ton of heat generated from this format it seems. This may have implications for the R5 heat limitations, and might be good news. It may mean that using an external recording could, indeed, matter in terms of the amount of time before you hit a heat limit.
 
  • Like
Reactions: TominNJ

davidhfe

EOS RP
Sep 9, 2015
302
443
I got the ProGrade Onyx version in. The increased performance over the Gen 1 version of the ProGrade gold version was a bit disappointing...

ProGrade (Onyx [black/gray], 325GB)
Read: 1525 MB/sec
Write: 1020 MB/sec

That's a nice improvement on the read speed, but the all-important write speed was only 7 percent faster. The claimed speeds are 1600/1400, so the write speed is definitely suffering versus the claim.

The upshot is that the Delkin is - by far - the best performer, and the best $/GB. I'd love to try the Lexar and Sandisk versions just to be complete, but can't justify buying more cards. If someone else has those, I can show them how to conduct this test.

One interesting note. After doing so much burn-in testing of these cards, I'm noticing that they get very, very hot. Unlike CF cards or SD cards. There's just a ton of heat generated from this format it seems. This may have implications for the R5 heat limitations, and might be good news. It may mean that using an external recording could, indeed, matter in terms of the amount of time before you hit a heat limit.
If you read/write a lot with the card hooked in to a PC, does it still generate heat? The whole body is designed to dissipate heat, so I wouldn't be surprised if the camera is heating up the card rather than vice versa. Though that said, heat sinks on M.2. drives on PCs pretty common. I'm super skeptical than an external recorder will help with overheating, but would love to be proven wrong. If an Atomos buys you unlimited 4K-fine, that's a huge deal.