Re: Who is working with the SAM9263?


"rickman" <gnuarm@xxxxxxxxx> skrev i meddelandet
news:1175474691.077254.15870@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
On Apr 1, 7:16 pm, "Ulf Samuelsson" <u...@xxxxxxxxxxxxx> wrote:
You can get the real thing from
www.mechtronicbrick.dk ==>www.mechatronicbrick.dk
www.ronetix.ch ==>http://www.ronetix.at/
www.cogcomp.com
www.iotech.dk
www.liab.dk

I could not get some of these links to work. But the others gave some
good results. Cogent has a 520 MHz PXA270 module for only $50 more
than the SAM9260 module. Way better deal and a *HUGH* increase in
speed! I have been looking for something like this.

Funny, I remember communicating with a guy recently which
balked at $600 price for a development kit with Ethernet
and $975 for DevKit with Ethernet and LCD screen.

Yes, that is $50 more than *their* price, not yours. It is $600 less
than your board.


The PXA270 DIMM is $399 DIMM in single qty.
If you want to get a single DIMM with baseboard, it is $1499.
With QVGA LCD = $1799..

Your LCD is a tiny thing that is of no value to me. I can connect
whatever I want to their PXA module with a starting price of $400!
That's $600 less than your 9263 board.

What can you do with a DIMM board without connectors?
The equivalent board is the $1799.
If you skip the LCD, and buy your own, then it is still $1499.
If you are prepared to design your own baseboard, then you
can get a chip for free.
If you burn a PXA module as part of the baseboard design,
then you are close to the cost of the 9263EK
Time to market can't be a concern if you are prepared
to wait for a working baseboard before you start.

I can get lots of good LCDs
for that price, and they fit *my* needs, not a arbitrary choice. If
the LCD is what is driving up the price of your board, maybe you
should consider selling the LCD separately.


If you need VGA or better resolution, then a dual bus system
like thewww.iotech.dkmodule and the AT91RM9200 is
probably faster than the PXA, since you do not steal bandwidth
from the CPU and you have several interesting accelerators.
The VGA controller, they have implemented in the FPGA
collects windows on the fly, so no heavy bitblt in the processor
just to move windows, or handle overlapping windows.

I am not sure what application you are thinking about, but I don't
expect this to be a general purpose computer that has to solve spread
sheets while managing a full size display. I seem to remember 386 AT
computers managing Windows very adequately as long as you didn't ask
it to swap out memory. So I don't think I will need special FPGA
hardware to manage a display in a handheld device.


IIRC, the 386 class processors only visibly moved the outline of a window
and not the complete window in itself, maybe even with WinCE, the
performance needs have increased since then.
I know for sure a number of guys starting with PXA, and
ending up with an external display controller when going past the QVGA.

If QVGA is OK, then SAM9261 will excel, due to its internal 160 kB SRAM.

Since the AT91RM9200 does not have a display controller, it
needs something outside and the FPGA will support large screens
like SVGA/XVGA, something I doubt the PXA will handle.



The AT91SAM9263, which is likely to soon be available from
a number of module companies, also has a dual bus structure
so you can have pretty large screens without affecting the CPU
performance.

I would like to consider that. But how long will I have to wait for
an eval board in a reasonable price range?


You can't do zilch with the DIMM module, unless you have
a baseboard, so you will have to develop your own board.
If you dont value time, why not start with a chip?

Since the part is just about to enter production, you will not
find anything, but I know there are DIMM modules coming
as soon as they can get production volumes.
My best guess is May/June.

How fast you can turn around a baseboard for the PXA?
What will be the long term cost of using the Cogcomp module?
How will that compare to doing your own design?


The Ronetix module has a complete frame buffer inside, and
if you make all redraws in this for a QVGA screen,
you have up to 800 MB/s bandwidth to the memory.
The effect is about 2 x performance compared to
doing draws to SDRAM.

MHz is not everything.

No, but it is very important. What happens with the 9263 when I go to
a full VGA?

The part was optimized for a VGA screen 24 bit per pixel.
A typical design would use one 32 bit bus (EBI0) for O/S and application
and a 16 bit bus (EBI1) for display refresh.

IMHO, The way a decent driver should work is that the 9263
will nmaintain a single buffer to draw the screen in a 32 bit SDRAM on EBI0,
with the help of the 2D accelerator (Line drawing, Fill etc.)

Once a screen is ready, it can be copied from EBI0 to EBI1
by the scather/gather memory to memory DMA.
In EBI1 it should be double buffered.

If you use SDRAM for EBI1 at 100 MHz /16 bit, your max throughput is 200
MB/s
Don't know sustainable throughput but I hope 120 MB/s is not overly
optimistic.
VGA at 24 bit per pixel is 921600 bytes per frame @ 60 Hz < 60 MB,
so you are using less than half the bandwidth to maintain screen refresh
and leaving half the bandwidth for DMA.
With 16 bit VGA, then you need about 40 MB/s and it becomes realistic
to use a PSRAM, with which you get more I/O pins, some of them real useable.

Typically you would not need to update the screen at the same rate
as display refresh, and every time this happens, you gain bandwidth.

VGA or no QVGA, the display refresh does not affect the performance of the
chip,
except for the cost of DMA; and the small latency coming from the need to
copy before
you display.
You could of course at the cost of bandwidth always do display refesh from
the
32 bit SDRAM the until the frame is in the EBI1.


--
Best Regards,
Ulf Samuelsson
This is intended to be my personal opinion which may,
or may not be shared by my employer Atmel Nordic AB


.