Out-of-the-Box Performance Evaluation

Silicon Motion claims read speeds of up to 895 MBps and write speeds in excess of 700 MBps in their marketing material for the SM2708. However, these speeds are heavily dependent on the NAND flash in the card. For the sampled reference design, Silicon Motion mentioned that they were seeing 895 MBps reads and around 420 MBps writes in terms of peak performance. Real-world speeds are bound to be much lower, depending on the particulars of the access trace.

CrystalDiskMark [Fresh]

CrystalDiskMark serves as a quick check to ensure that the card can meet the performance claims of the manufacturer. The workloads were processed for the SM2708 card in both SD Express and UHS-I modes.

CrystalDiskMark [Fresh] Benchmarks
TOP: BOTTOM:

In the out-of-the-box case, the read and write speeds in SD Express mode match Silicon Motion's numbers - coming in at 890 MBps and 418 MBps respectively. The bump in the random access IOPS in multi-threaded scenarios indicates that the SM2708 controller is pretty much a NVMe SSD controller with the legacy UHS-I controller functions addedin another form. The UHS-I numbers are around the 100 MBps mark, as expected. In any case, UHS-I operation is for backwards compatibility, and performance is not much of a concern in that mode.

Sequential Access - fio Workload

One key aspect to note with SD Express moving forward is that it is going to be pretty much impossible for the cards to maintain their peak speeds beyond the initial SLC cache region. In effect, the claimed speeds are going to be only for burst scenarios. For most applications, that really doesn't matter as long as the card is capable of sustaining the maximum possible rate at which the camera it is used in dumps data. We use fio workloads to emulate typical camera recording conditions. We first run the workload on a fresh card, and also after simulating extended usage (covered in a later section). Instantaneous bandwidth numbers are graphed.

fio Sequential Workload [Fresh]
TOP: BOTTOM:

The reference design has a SLC cache of around 5.5 GB up to which write speeds of up to 375 MBps can be sustained. Beyond that, we have a 75 MBps region, and further on, 30 MBps. Once the controller has been subject to this traffic, the reads start off around 200 MBps before moving higher and higher and ending up at around 700 MBps. On the other hand, the SLC caching effect is a bit more difficult to track in the UHS-I mode. Write speeds vary from as low as 10 MBps to peaks around 62 MBps. Reads are consistent at 72 MBps.

Next Gen SD Card Review: SM2708 Simulating Extended Usage
Comments Locked

43 Comments

View All Comments

  • eastcoast_pete - Thursday, September 9, 2021 - link

    Those temperatures and power draws are way too high for what is probably the main use scenario: removable storage in high-resolution handheld cameras. They're also too high for potential use as removable storage, again mainly for video, in future Smartphones, once Android counterparts to Apple's upcoming iPhone Pro Max become available. That can reportedly use ProRes as file format to save videos, and ProRes eats storage space for breakfast.
  • bananaforscale - Thursday, September 9, 2021 - link

    Temps are certainly too high, but power draw not necessarily. That peak draw of 4.58W is ~1.1A at 4.2V, so you could power one off a single 18650 cell for a couple of hours assuming peak is constant, which it won't be.
  • Fulljack - Thursday, September 9, 2021 - link

    I think it's more intended for state-of-the-art digital camera that output high bitrate, high resolution video. one of the limiting factor are storage speed. having this on what seems like a laptop on your shoulders would help portability a lot.
  • at_clucks - Friday, September 10, 2021 - link

    At that point many users simply invest in cameras that output to an external SSD, at least with something like a CFast to SSD adapter.
  • schuckles - Friday, September 10, 2021 - link

    Whether these cards are usable as is really depends on your workload. I would have liked to see something like the anadrech light benchmarked and include the MBPS/watt to know how realistic that 4 watt workload is. Also curious what the idle/standby power of the card itself is.
  • PaulHoule - Thursday, September 9, 2021 - link

    How much heat could you remove from that kind of thing if you made the exterior out of metal and pressed it between two cold plates with a strong spring? I'd like to give an SDExpress card the same treatment as the "CPU" on an IBM 3090.
  • shelbystripes - Thursday, September 9, 2021 - link

    So… CFExpress cards are stupid fast, but get hot just doing sustained read (copying a full 256GB card’s contents to a computer all at once). And this is putting the same thing in an even smaller form factor. Of course the end result is inevitable.

    For photography this is still valuable, the cards are fast enough even in UHS-I mode for sustained burst in most cameras, and then having PCIe speeds for clearing the contents quickly to a PC is quite handy. But for video… these things will get hot, and compact camera thermals are already a nightmare. The Canon EOS R5 can technically shoot 8K video, but do it for very long and the damn thing overheats. The more you add awful thermal properties inside a compact environment like that, the worse off you are…
  • spaceship9876 - Thursday, September 9, 2021 - link

    you forgot to mention ufs cards.
  • Einy0 - Thursday, September 9, 2021 - link

    Wow, 96C that will leave a mark!
  • Kamen Rider Blade - Thursday, September 9, 2021 - link

    When are regular sized SD cards going to go up in size?

    1TB Micro SD Cards already exists.

    I calculate that if you make optimal use of the physical space of a SD card, you can get 4.5 TiB of storage using existing 512 GiB Nand Flash cells.

    Wouldn't you want an entire regular sized SD card with 4.5 TiB of storage?

Log in

Don't have an account? Sign up now