Hacker News Reader

Comment by tischler | Hacker News Reader

Why is that significantly slower than the specs of these SD cards (100-80 MB/s read/write)?

As @written says, there is a huge limitation on the microSD card reader on the Pi itself; but also, in my testing on a UHS-II USB 3.0 card reader on my MacBook Pro, most of the cards can't sustain more than 30-40 MB/s write even if the specs say they can.

Large block read speeds are usually pretty accurate, but manufacturers take quite a bit of liberty with their performance claims. And random I/O is pretty terrible in almost every case.

Remember that these types of cards are _usually_ optimized for large file I/O since they're used in dashcams, GoPros, and the like—use cases that are vastly different from a general computing device running Linux!

The RPi is more then likely the limiting factor... kind of makes the tests useless if you want to figure out which cards are faster

The fastest cards in these tests were also fastest when writing the entire image on my USB 3.0 UHS-II card reader on my Mac. Large file writes are where most of these cards shine, and some can do 40+ MB/s when writing larger blocks of data.

If I'm remembering correctly, the Pi's SD reader runs at 50MHz, doing 4-bit transfers, and not using any of the UHS signaling methods, because those use 1.8V signals that the Pi isn't set up to use for SD.

> kind of makes the tests useless if you want to figure out which cards are faster

But...which card is faster in a USB3 UHS-III transfer isn't useful information for a Raspberry Pi benchmark. It would certainly tell you which cards are faster, but the info wouldn't be directly applicable to what the tests are trying to measure.

I'm sure the interface is slower than what you would find in a laptop, but I'm sure the blocksize of 4k/8k isn't helping on those tests.

Host side interface limitation.

Reply on news.ycombinator.com