Yummy new CPU/GPU combo

Carsten Haitzler (The Rasterman) raster at openmoko.org
Thu Jun 5 07:02:15 CEST 2008


On Wed, 4 Jun 2008 19:59:35 +0200 Tom Cooksey <thomas.cooksey at trolltech.com>
babbled:

> On Tuesday 03 June 2008 09:06:35 Carsten Haitzler wrote:
> > the day nvidia comes with open drivers for this... we can begin to take an
> > interest :)
> 
> Sorry, just how open is the current glamo driver exactly?

the driver is open. the chip specs so you can go write your own driver from
scratch are only available under nda though :(.

> IMO, OpenMoko's choice of using the the glamo was a big mistake (Connecting
> it to a shared, 4-bit bus was probably the _biggest_ mistake).

i definitely can't argue here. :) i know when i turned up glamo was already in.
my guess after some quick poking and a lot of spec doc reading was that given
full drivers - we'd give or take have about the same performance as pure
software only drivers (for 2d) in a dumb framebuffer - i.e. - what gta01 has.
the exception might be opengl (3d) there where software won't be able to render
and keep up with the glamo, but 3d comes with a lot of caveats on the glamo,
like max 256x256 textures. no render-to-texture (so intermediate buffers are...
well.. a pain), and the unhappiest is the max 3d buffer size (target for
rendering) of 511x511 - so as such we can't manage fullscreen 3d @vga res. we'd
always have to go down to something like qvga. 

> OpenMoko insisted on having open source drivers, and the only hardware vendor
> which would let them do it was SMedia. So we get this massively underpowered
> graphics processor - but that's ok because OpenMoko can release the source
> code to the driver! Except SMedia wont let OpenMoko realease any technical
> info about the chip. So, in fact, the glamo driver can only be developed by
> people employed by OpenMoko and only after they sign an NDA.

well we can't release the docs smedia gave us. but we are allowed to talk about
the technical details all we like in public. THOSE docs that smedia wrote are
theirs and we can't give them out. we are allowed to go write our own docs
based on it - but that won't be a small document, and right now internally no
one has the time to just sit down and write such a doc (as opposed to writing
code).

> The Xglamo may be open source, but it is not, nor can it ever be, a community
> project.

it can - partially, but only in as far as any chip specs that are apparent in
the existing code.

> Now look at the other options OpenMoko could have gone with. Well, _the_
> option of cource would be to have used a half-decent SoC, one with an
> integrated GPU such as a Freescale i.MX or TI OMAP, even an XScale would have
> been better. You'd then get a PowerVR GPU (same as used in the iPhone), which
> already has Linux drivers. What's more, OpenMoko would get the source code
> for the drivers, under NDA from Imagination Technologies. The only
> restriction would be that OpenMoko couldn't release the source. But that's no
> different to the glamo, given that only OpenMoko employees can work on it.

again - we could have chosen much better soc's indeed, but as it stands right
now that isn't happening. gta02 is basically a gta01 slightly improved. gta03
will at this stage be a gta02 minus glamo (which simplifies things a lot and at
least for 2d stuff should get us an improvement).

> We would also have had a decent processor, one with a more up-to-date
> instruction set than the nearly 10-year-old armv4t. So there may not have
> been Cortex A8s around 2 years ago when GTA02 development began, but there
> were plenty of options. Why on earth did OpenMoko stick with an aging CPU and
> an almost useless GPU when there were so much better options? And please
> don't say BOM, I refuse to believe the combined price of the 2442 and glamo
> is cheaper than e.g. an i.MX31 or OMAP2420.

because even before that the neo was inherited from an older project to do a
windows-mobile phone... the hardware is based on an ancient product design.

> Cheers,
> 
> Tom
> 
> PS: Very sorry for the rant, I just had such high hopes for OpenMoko and am
> just fustrated with the hardware design decisions.

understood. i can't blame you. i've done my share of ranting too. but even
internally the most i have managed to do, i think, is influence the glamo to
vanish from the gta03, but otherwise we have what we have now (just a new case,
different battery, added a camera, a different gsm module).

gta04 was looking good. the samsung 6400 as such is a nice soc. i've dug into
the 2d gfx portion of the docs. it's a bit tricky to use, but much better than
the glamo and much more powerful in the 2d department (u'll manage scaling
and compositing even in argb32 bit) nicely on that baby. it's memory bus was
way better than gta02, and as it's graphics is integrated on the soc software
cpu access to do fallbacks as well as mixing with hardware accel is sane and
performant (unlike accessing video memory across a puny bus from the cpu when
you need to fall back to using the cpu to do stuff).

but... gta03 is back from the dead, gta04 is off the map (for now) and
everything is going to be based on the existing 2442 arm v4 samsung soc for
gta03.

quick question - would you prefer a qvga lcd (save a bit of cost) since we'e
going to need to software-drive all graphics - the fewer pixels you have to
fill, the better for speed. i'm really tossing up if the speed of qvga is worth
the loss of resolution. i'm just not sure.

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




More information about the community mailing list