Generally I use the 10gbps (~950 MB/sec.) Type A USB-C card readers, but am now expected to be using the Sony a7rVI more frequently and with a higher volume of data. I have 3x480GB and 1x640GB different cards with CFe Type 4 logos, so assume they will download faster. I'm trying to find one that runs at full speeds and is of reasonable weight, since will only used in foreign zones with a laptop. Is the Lexar working for anyone at full speed? https://americas.lexar.com/product/lexar-professional-workflow-cfexpress-4-0-type-a-card-reader/
I purchased the Lexar Professional CFExpress 4.0 Type A Reader and twin 640GB Lexar Gold CFExpress Type A. All still in boxes, but no reason to think that these won’t be fast and somewhat future proof.
I use the Lexar card reader with their 40 Gbps USB-C cable on my Thinkpad laptop. Using a Sony Tough 960GB CFE-A type 4, I see ~1100 MB/sec reading. The card is rated 1800 MB/sec. If I do long transfers it will slow down, but I have not made measurements of rate vs data transferred. My guess is that the slowdown is the result of the card heating up and throttling the rate.
dclark wrote:
I use the Lexar card reader with their 40 Gbps USB-C cable on my Thinkpad laptop. Using a Sony Tough 960GB CFE-A type 4, I see ~1100 MB/sec reading. The card is rated 1800 MB/sec. If I do long transfers it will slow down, but I have not made measurements of rate vs data transferred. My guess is that the slowdown is the result of the card heating up and throttling the rate.
I think it more likely that an internal cache has been filled and transfers are falling back to native storage speed.
Alan321 wrote:
I think it more likely that an internal cache has been filled and transfers are falling back to native storage speed.
If by 'native storage speed' you mean the speed of the computer's storage device that the card reader data is being written to then I agree.
One can measure the speed of the reader in isolation by only transferring the data to the computer's memory, which can accomplished by either using a benchmarking app that supports this or by constructing a command line that transfers the data to the OS's null device.
I would use Crystal Diskmark with run length 64GiB on the card+reader. My little laptop has old school 2x 8TB SN850X. Those typically run about 6400MB/sec. writes under pSLC then around 1500 writes at worst. The dynamic pSLC is fairly large depending on how full the drives are, but I generally leave home with SSDs nearly empty as a matter of principle. I have added thermal pads on the SSDs that transfer some heat to the shell, not that it is designed for that.
Alan321 wrote:
I think it more likely that an internal cache has been filled and transfers are falling back to native storage speed.
I was writing to a high speed SSD that Crystal Disk measured at >6500 MB/sec write speed.
The slow down I am referring to takes several minutes and the card is noticeably hot when I remove it.
EB-1 wrote:
I would use Crystal Diskmark with run length 64GiB on the card+reader. My little laptop has old school 2x 8TB SN850X. Those typically run about 6400MB/sec. writes under pSLC then around 1500 writes at worst. The dynamic pSLC is fairly large depending on how full the drives are, but I generally leave home with SSDs nearly empty as a matter of principle. I have added thermal pads on the SSDs that transfer some heat to the shell, not that it is designed for that.
Not sure what all the computer setups are, but all this probably depends on the cable, the USB port, and the SSD speed and design (cacheless or not). To get the 40Gbps you need a USB4 or TB3/4 port, a fast enough data cable, and a Gen 3/4 SSD with a decent cache to maintain speeds and PCIe lanes clear enough to maintain the bandwidth. Even then an SSD might throttle because of heat or a full cache, depending on how many files are transferring.
wordfool wrote:
Not sure what computer setups are, but all this probably depends on the cable, the USB port, and the SSD speed and design (cacheless or not). To get the 40Gbps you need a USB4 or TB3/4 port, a fast enough data cable, and a Gen 3/4 SSD with a decent cache to maintain speeds
Type A single lane PCIe 4 is limited to under 2000MB/sec. but of course lower in practice. That corresponds to about the 20GBps external speeds. It seems that the 20Gbps USB 3.2 Gen 2x2 was never popular or widely supported, so there is really the choice between the the old 10Gbps USB 3.x or the 40Gbps USB 4 or TB.
The computer is an T&L 14' laptop released ~Q2 2025 with 255H (Arrow Lake-H) CPU so should not be the bottleneck. I've never used any USB 4 devices period, but supposedly the two USB-C ports are USB 4. The cable comes with the reader so I would use that. I have avoided USB 4 in any other of my computers due to PCIe lanes being needed for other things.
I replaced the stock 1TB SSD with two 8TB SSDs. The WD SN850X are a generation older than today, but PCIe 4.0 with TLC NAND and full DRAM for the mapping tables. I actually got those back in 2024 for an older laptop. The prices today are exhorbitant for a laptop with 16TB so I'm staying put for now.
And if you're working on a MAC, expect different speeds as well. While a TB4 and USB 4 should be able to generate 5k MB/s, actual peak speeds are around 3.1-3.5k due to caching, overhead, and controller issues. With my NVMe chips (which are faster, 5k-7k) in SSD's in USB 4, 40Gbps enclosures that's about what I achieve although CFE USB 4 card readers are slower due to the cards themselves. If your CFExpress B card is rated at 3000 MB/s, that's usually peak speed and with sustained file transfers, it will likely drop to as little as 800 MB/s or possibly less depending on the card.
My specific worst case is on travel where we are weight restricted on small propeller aircraft, so it is best to reduce the computer gear weight and allow for more photo gear. I have over 6TB of fast (mostly Delkin Blacks or Sony) CFe Type B cards so don't have to reuse them much if at all. Downloading Type B was not much concern, but I may reconsider that later.
I have far fewer Type A and really did not want to put in another $1500 or so in cards for the a7rVI yet. The a7rV was so slow at 10FPS that one did not shoot as many frames as 30FPS on more modern cameras like R5 II or upcoming a7rVI. Most of the Type A cards I had were only 900MB/sec. reads (CFe 2.0) so 10Gbps readers were already maxed out.
Of course the CFe cards are slower than the NVMe SSDs, partly due to the size/power/heat constraints and also the cut-down bus width. We are so used to SSDs reaching >14,000MB/sec. on PCIe 5 four lanes in the desktop/server computers for years already. Older standards like PCIe 4, cut down to 2 lanes for Type B or a paltry single lane for Type A are far more restricted. But the camera write speeds are nowhere near the limits.
TBH I have zero experience with USB 4 or TB anything. TB PCIe standards were a few years behind the times of what you got with internal PCIe slots, so I had no use case. An x8 PCIe 4.0 MegaRAID card from a few years ago is more than enough for me. And I can pop an M.2 SSD into an onboard slot to offload files onto whatever internal or external storage.
I finally received the lexmar and results are simlar to yours on the laptop depending on the Type A card (Delkins Black, OWC, Lexmar, SOny, etc.).
Unfortunately it's difficult to properly test the writes because every Type A card I have overheats and throttles down rather quickly. I know the Sony cameras are nowhere near the card limits yet, but the puny Type A form factor really could use some kind of heatsink for sustained CFe 4 mode writes. It's ridiculous and I have no idea how it would meet qualification at 50°C under load like we do with many electronic components and complete devices. The card slots need some kind of heatsink to come into play from the sides.
Actual ARW download speeds are less than theoretical, but about 1.53x faster than from a 10Gbps reader in the same machine so there is some value to it.
Good to hear the reader is providing good value to you. I agree the quicker downloads are useful.
Write speed in the card reader is not something I care about. I have no use case for that.
The puny size may be a contributing factor to heating and consequent slowdown, but in camera I have never noticed it or been able to measure it. I make measurements involving running the camera at 30fps for up to a minute. The frame rates are constant. In the field I shoot lots of long sequences, sometimes a few hundred frames. I have never been able to detect a thermal slowdown.
The puny size means I have two fast cards in the camera which is very valuable to me. I like the CFE-a form factor, and the Sony slots which accommodate either CFE-a or SD cards in both slots.
Rather than increasing the camera write speed to the cards, which will mean more heat, I would prefer to see the in-camera buffer size increase so that I get a full 10 seconds before any slowdown. It seems to me that should be an easier engineering problem than increasing the speed of clearing the buffer while acquiring frames at 30fps (I shoot with an A1m2).
Given the high cost of RAM, I'm not expecting huge buffers on the high MP cameras like a7rVI in my lifetime. RAM goes a lot further with 24MP. But if I didn't want high res I'd rather just use Nikon and Canon 45MP.
Yes, the cost of memory may inhibit putting larger buffers in cameras, at least for a while. I assume I'm older than you but I hope to see it. The A1m2 is 50MP, not 24, but as can be seen in my measurements, A1m2 CFE-a Measurements, the buffer is only ~71 frames. So even if the frames are not compressed prior to the buffer, the buffer is only about 7GB, and 4GB if frames are compressed prior to the buffer. Frames written to the card are compressed and of the 155 frames acquired in just over 5 seconds, 155-71=84 have been written to the card, which means about 830MB/sec write speed to the card, which matches the rate of the frame rate after the speed break. If the in-camera card write speed is increased to ~30*52=1560MB/sec (less than 2x) the buffer can be very small and it will never fill. I'll take either solution.
Whatever happens cameras will just keep generating more data. My laptop is no longer made and the newer ones in that class are worse, so I'm not making any changes there for a long time.
