Add three more microSD cards to the benchmarks, test A2 performance

[geerlingguy]

After the last round of benchmarks, I purchased two new microSD cards that I would like to test:

1. SanDisk Extreme 64GB U3 V30 A2
2. Samsung Pro Endurance 32GB U1

For the former, some people suggested A2 cards may have better random I/O characteristics. For the latter, many suggested these cards can handle multiple write cycles better (though I wonder if it's more optimized for the 'write entire card then write over entire card' use case of a dashcam, than for random small file writes...).

Indeed, the SD Association raves about how much better Application Performance Class 2 cards promise to be:

The Application Performance Class 2 (A2) is defined by SD Physical 6.0 specification. It makes SD memory card much higher performance than A1 performance by using functions of Command Queuing and Cache. Efficient flash memory management can be supported by Maintenance functions. Application Performance Class can be applied to UHS SDHC/SDXC Memory Card product family.

And:

However, Android Police (and other sites which have benchmarked available A2 cards) have noticed that the cards tested don't really meet advertised minimum IOPS ratings. So... let's find out what happens with these cards in a Pi! And on my Mac!

[geerlingguy]

SanDisk Extreme 64GB U3 V30 A2

• hdparm: 43.73 MB/sec
• dd write: 36.3 MB/sec
• 4K read: 7.35 MB/sec (1882 IOPS)
• 4K write: 3.70 MB/sec (947 IOPS)

Samsung Pro Endurance 32GB U1

• hdparm: 43.52 MB/sec
• dd write: 37.4 MB/sec
• 4K read: 10.21 MB/sec (2613 IOPS)
• 4K write: 2.83 MB/sec (724 IOPS)

(IOPS calculated using formula: IOPS = ($MB_PER_SECOND / 4) * 1024)

[geerlingguy]

I'm kind of amazed that the A2 is getting worse random I/O / IOPS on the Pi than the 'super old and totally outdated' A1/non-A-class Sandisk Extreme... sounds like either I have a terrible card, the Pi has a terrible microSD interface, or it's marketing wank.

[geerlingguy]

To test the theory above, I used AmorphousDiskMark using a brand new UHS-II XQD/SD USB 3.0 reader on my Mac, with the A2 microSD card formatted as FAT32. I couldn't use iozone because it gets caught up in macOS's built-in filesystem caching, which makes the numbers go a few orders of magnitude off from reality.

Running 4K I/O benchmarks now.

[geerlingguy]

Using ADM:

• 4K read: 6.99 MB/sec
• 4K write: 1.05 MB/sec

[geerlingguy]

I'm not satisfied condemning the 'A2' designation based on one card, so I've also ordered a Lexar Professional 128GB microSDXC UHS-I/U3 A2 V30 card. I'm waiting for it to arrive, might take another day or two.

[geerlingguy]

Lexar Professional U3 A2 V30 128GB

• hdparm: 44.35 MB/sec
• dd write: 32.4 MB/sec
• 4K read: 9.65 MB/sec (2470 IOPS)
• 4K write: 2.55 MB/sec (653 IOPS)

[geerlingguy]

Yeah... so A2 means absolutely nothing. Completely made-up marketing BS to sell 2x more expensive cards. If one of the manufacturers, or the SD association would like to disagree, I invite them to show me any reproducible test which proves the IOPS are anywhere near their required minimum specifications. It's just like how air compressor manufacturers bend and break the rules to get their CFM ratings.

I also tested on another brand new microSD and SD card USB 3.1 Type C reader, which supports UHS-II for SD (doesn't mean anything for microSD AFAICT though), and it was giving similar results.

[geerlingguy]

Also taking a glance through the SD 6.0 Part I physical specification to see if there's something huge I'm missing here...

From the spec:

Note that unit of random performance [IOPS] indicate number of 4KB sized I/O Operations completed in one second, wherein each transaction was targeted to a random address generated for 256MB address range.

[geerlingguy]

There's nothing else indicating what kind of conditions would be different, however I have been testing with a 100MB file/block size in the past, so I'm going to run IOzone with a 256MB file to see if it makes any difference:

./iozone -e -I -a -s 256M -r 4k -i 0 -i 2

[geerlingguy]

Nope, still 9.58 MB/sec read, 2.55 MB/sec write. Weak. Results in all cases differ < 1%... at least the card is very reliable.

[geerlingguy]

A chart, to finish out this issue:

It saddens me the SD Association can get away with this.

[stale]

This issue has been marked 'stale' due to lack of recent activity. If there is no further activity, the issue will be closed in another 30 days. Thank you for your contribution!

Please read this blog post to see the reasons why I mark issues as stale.