This really isn't a Debian issue, but I'm hoping someone here might have some insight about why my kirkwood devices aren't as fast as I expected. (I have 2 dockstars and 1 sheevaplug)
I like running chess engines on all manner of embedded devices. One device is a
TRENDnet TEW-652BRP. It has a 400 Mhz MIPS 24Kc V7.4 CPU, 32 MB RAM, 4 MB flash, and runs OpenWrt. [I should mention that only R1.0 will run OpenWrt as the newer models have a different chipset.]
I was surprised to discover that these 1.2 Ghz Kirkwood devices are only about twice as fast as the router. I really expected at least 3x, possibly a lot more. I figured with a 3x clock speed advantage and newer architecture, the sheevaplug and dockstars would be well over 3x faster, yet 2x is all I got.
I get about the same speed on the Dockstar with OpenWrt as with Debian for the chess engines I'm testing. My primary motivation for trying OpenWrt was to see if the CodeSourcery enhancements made any appreciable speed difference. (-mtune=marvell-f compiler option).
So what gives? Why am I only getting double the performance of a similar MIPS device at 400 Mhz? Could it be the 800 Mhz memory speed that's the bottleneck?
Edited 1 time(s). Last edit at 10/15/2010 09:11PM by Nematocyst.