XO battery/performance

[Yioryos Asprobounitis]

--- On Mon, 5/28/12, Jon Nettleton <jon.nettleton at gmail.com> wrote:

> From: Jon Nettleton <jon.nettleton at gmail.com>
> Subject: Re: XO battery/performance
> To: "Yioryos Asprobounitis" <mavrothal at yahoo.com>
> Cc: "OLPC Devel" <devel at lists.laptop.org>
> Date: Monday, May 28, 2012, 3:45 AM
> *snip*
> http://lists.laptop.org/pipermail/devel/2012-May/035198.html
> ) I also
> tried to test the XO-1(os880), XO-1.5(os883) and -1.75(os10)
> battery
> life and thought to report it. It might be of help to
> someone.
> 
> os10 is very much a development release and not great for
> testing.  In
> particular graphics acceleration was not working properly,
> so this
> test is strictly testing the cpu.
> 
> *snip*
> 
> > I can appreciate that this is far from a CPU/computer
> performance test but taking counts just as an *indication*
> of CPU/FPU performance would imply that the XO-1.75 is 65%
> faster than the XO-1 but 40% slower(?) than the XO-1.5.
> 
> see above reasoning as to why this is an incorrect
> assumption.
> Although the XO 1.75 will be slightly slower than an XO 1.5
> in this
> test as it is primarily doing glyph compositing in the
> terminal.  The
> XO 1.5 hardware supports compositing the A8 format via the
> 3d engine,
> while the XO 1.75 only supports rgb565 and argb8888 in
> hardware so a
> lot of the compositing functions will fall back to
> software.  We are
> helping to speed up this fallback by enabling iwmmxt
> optimizations in
> pixman.
> 

Tried to take this out of the equation modifying the script as such
`#!/bin/bash
let count=0
while [ $count -le 1000000 ]
do
((count++))
done'
and then run `time script'
The XO-1.75 gave:
real	2m28.625s
user	2m22.540s
sys	0m1.490s

and the XO-1.5:
real	1m45.648s
user	1m41.430s
sys	0m1.760s

So XO-1.75 does take ~40% more than the XO-1.5 in this particular CPU test.
I really do not know if this has to do with the ARM libs or the CPU itself but the difference appears to be the same as with the previous "scrolling" version of the test.

> > Considering counts/discharge however, the XO-1.75
> appears 33% more efficient than the XO-1.5 and 320%(!) more
> efficient than the XO-1
> >
> > One thing I noticed is that while on the XO-1 and
> XO-1.5 the output was a constant stream of new lines without
> and "hesitations" and the cursor barely visible, on the
> XO-1.75 were "hesitations" and the cursor was appearing at
> the end of the line during these "hiccups".
> 
> These hiccups are probably due to the lack  of hardware
> acceleration.
> However because  the SOC does do more rendering I have
> also been
> experimenting with increasing the buffer for libxcb to help
> prevent
> hiccups between the Xclients and Xserver.
> 
> OS12 should be a better image to test on the XO 1.75.

Tried it with os12 and got identical results and behavior.

> 
> -Jon
>