[Arm-netbook] http://lateral.netmanagers.com.ar/weblog/posts/the-raspberry-pi-suck s.html

[lkcl luke]

On Wed, Jun 20, 2012 at 11:16 AM, Gordan Bobic <gordan at bobich.net> wrote:
> On 06/20/2012 11:09 AM, Benjamin Henrion wrote:
>> On Wed, Jun 20, 2012 at 11:58 AM, Tzafrir Cohen<tzafrir at cohens.org.il>  wrote:
>>> On Wed, Jun 20, 2012 at 09:48:16AM +0200, Vladimir Pantelic wrote:
>>>> http://lateral.netmanagers.com.ar/weblog/posts/the-raspberry-pi-sucks.html
>>>
>>> Summary: The Mele a1000 costs way more than a Raspberry PI, but it's
>>> worth it. Now go buy it. Here's a link to DealExtreme.
>>>
>>>
>>> No news here if you followed this list.
>>
>> A10 suffers from the same problems as the BCM chip of the Rpi, namely
>> closed source binary blobs for the GPU.
>
> If that is such a big issue (and it is certainly not a trivial issue),
> then maybe pursuing an ARM solution is wrong in the first place and we
> should be pursuing a solution based on Loongson MIPS, a-la Leemote
> Yeeloong or similar.

 we're talking to a company that has a CPU design which also has
built-in GPU and VPU instructions.  they're using open64 (not gcc!)
and the *entire* toolchain *and* the full instruction set will be
fully 100% available as software libre, and documented.  they already
have the linux kernel up and running and ported, and are keeping some
engineers happily employed working on android (bless 'em).

 they have a number of key people on board.  one is a compiler expert
with over 20 years experience: he's the man behind the open64 port.
another used to work for SGI and was i believe on the team that did
the ATI graphics engine.

 the only problem is that their technology has taken them so long to
get up and running that they missed the release window for their first
design, by a number of years.  i.e. if they had released silicon back
in 2006 based on what they designed it would be - would have been -
absolutely amazing!  but they didn't, so a dual-core 700mhz CPU even
with 8-way hyperthreading looks pretty... slow, in today's market.

 so, we've advised them to go all out with a 28nm quad-core version,
which they should be able to put out in something like 15 to 18 months
time.  we're also attempting to persuade them to do demo boards with
their revision 1 silicon so that software (libre) developers can
actually work with it _before_ then, but each one will cost them a
king's ransom, relative to standard engineering boards.

 anyway - just so you know, there really is a long-term plan spanning
ahead about 3-5 years for this project.

l.