memory bandwidth (was Re: JZ4750 driving large displays?)

Kristoffer Ericson kristoffer.ericson at gmail.com
Fri Aug 20 04:07:07 EDT 2010


On Fri, Aug 20, 2010 at 01:25:29AM -0300, Werner Almesberger wrote:
> I wrote:
> > These are theoretical guesstimates. When I run a loop that does
> > memsets on the Ben, I only get about 38 MB/s with display refresh,
> > 39 MB/s without. This suggests that memset only uses a fraction of
> > the available bus bandwidth, and so would anything based on memset.
> 
> This was on a Ben still running the system it was shipped with (I
> killed gmenu2x, which is a CPU hog in that old distro but didn't
> touch anything else). I compiled the benchmark with the toolchain
> from openwrt-xburst checked out yesterday.
> 
> Rafa ran the same benchmark on jlime and got about 57 MB/s memset
> bandwidth. That's 150% of my 38 MB/s. Still a far cry from the
> ~200 MB/s I think the hardware should be capable of, but an
> interesting difference.

Hmm, did rafa copy your environment (processes running) somewhat?
Off the top of my head I cant remember any special optimizations
that we put inside the compilation. OE binaries are usually stripped
down somewhat so whatever runs compaired to openwrt tends to give lesser
memprint/runtime. But it shouldnt reflect that hard I think.

Really interesting benchmark btw! Says that we are nowhere near
the actual capabilities. 

> 
> My crude benchmark is here: http://pastebin.ca/1919777
> 
> - Werner
> 
> _______________________________________________
> Qi Hardware Discussion List
> Mail to list (members only): discussion at lists.en.qi-hardware.com
> Subscribe or Unsubscribe: http://lists.en.qi-hardware.com/mailman/listinfo/discussion




More information about the discussion mailing list


interactive