low key derivation function performance

[beroal]

I measured that my tablet is 20 times slower than my desktop when computing a key derivation function. I've done this by creating a LUKS container with cryptsetup --type luks1 --iter-time 10000 --hash sha512 luksFormat $CONTAINER and measuring the time for opening the container with a watch. The clock frequencies of processors are not that different:

• 2.36 GHz Cortex-A53 in the tablet
• and AMD Ryzen 3 2200G 3.7 GHz in the desktop.

I suppose that the computation is single-threaded. Maybe, edslite uses an inefficient cryptographic library?

[emmenlau]

Isn't this speed difference what you would expect between a RISC and a CISC type CPU? I'm not sure if some kind of intrinsics or other complex instructions play a role in computing a key derivation function. But it would be very surprising if the many times more transistors in the desktop CPU can not aid in this intense computation?

[beroal]

But it would be very surprising if the many times more transistors in the desktop CPU can not aid in this intense computation?

Actually, transistor counts are comparable:

• 4.95e9 in AMD Ryzen 3 2200G,
• 4e9 in HiSilicon Kirin 960.

These numbers include GPUs. Actually, I measured a slower HiSilicon Kirin 659, but I do not have data on it.

Of course, mobile CPUs are slower than desktop ones, but 20 times seems too much.