Yummy new CPU/GPU combo

Carsten Haitzler (The Rasterman) raster at openmoko.org
Fri Jun 6 04:36:44 CEST 2008


On Thu, 5 Jun 2008 12:56:00 +0200 Tom Cooksey <thomas.cooksey at trolltech.com>
babbled:

> Well... yeah, that's kindof the reason for the initial post! I must have
> misunderstood previous posts as I was convinced the glamo was itself attached
> to the CPU's SDIO interface. You can understand where I was coming from now!

:)

> Right. Shame the cpu doesn't have more cache so it has a better change of
> getting on with something else while the DMA occurs.
> 
> Thanks for clearing things up. It's good to hear that the problem isn't
> because of any particularly bad decision on OpenMoko's part. Renews my faith
> a little. :-)

glamo is a fairly old graphics cpu and design. frankly it was designed for qvga
- at best. it CAN do vga, but that is beyond its intended use. we have pushed
it much farther than smedia intended it to be - even smedia is amazed at the
bus speeds we get to the glamo - they have never seen it so fast.

glamo turned up primarily for 2 reasons.

1. we added wifi - and we just didn't have any other (good) interfaces (we can
sit and discuss spi all u like right now - but lets hold that thought). so we
needed to use up our sdio port for... wifi - but where will we stick the actual
sd card? glamo came with its own sd interface - bonus, so we get that back
again.

2. refresh. we run a vga screen. 640x480x70x2 - 41mb/sec ... JUST to refresh
the screen. we lost 41mb/sec of memory bandwidth just to keep the screen
displaying on the gta01. vga was by far pushing what the soc can do. literally
that is a huge amount of the bandwidth we had left (as its reads only and we
clock @ 100mhz for the bus clock on the gta02, it's be about 20% of memory
bandwidth gone). the glamo removes that drain and gives u back bandwidth... but
at another cost - the glamo's own bus and the extra work to write acceleration
to keep as much on the glamo as possible. we definitely don't do everything we
can on the glamo - but there is only so much time and development effort to go
around. as such we only really knew what the glamo was like once we had it
going reasonably, and by then - it's too late. glamo is in. it's not going away
as there just is no alternative.

so solutions could be: ditch vga, go to qvga, thus dropping mem bandwidth loss
to 10mb/sec (1/4 of what it was), and now have to drive only 1/2 the pixels in
software. end result would be much lower dpi, but much smoother and nicer
performance. but then go all the cries of "i must have vga!!!" (though i wonder
just how many people have good enough eyesight to really tell the difference).

if we want to live in the realms of such high resolution, imho, we need to look
very carefully at a high-end soc with good on-board graphics acceleration, but
that would have put the freerunner on hold as a product for a looong time. so
you would all still be waiting and waiting. no one would be talking of
production runs right now! :)

-- 
Carsten Haitzler (The Rasterman) <raster at openmoko.org>




More information about the community mailing list