<markvdb>
I'm looking for an arm device with 3 or ideally 4 usb ports, ethernet, 2Gb ram, some flash (at least 4Gb, 8Gb would be better), and a powerful cpu where one core could use all of the 2Gb of ram if needed
<markvdb>
that's a long list, and I think I found some Chinese set top boxes that do exactly that
<markvdb>
where I could flash Debian
<markvdb>
I just wonder about the cpu addressing memory
<markvdb>
for the rest it seems rk3188 devices are good candidates
<Astralix>
I am not sure if you find a device that is ready for you. Looks more like you need to design your own PCB.
<markvdb>
I've seen these CS968 devices
<markvdb>
they seem to do exactly that
<markvdb>
and they claim to be RK3188
<markvdb>
even found a source that claims he flashed debian onto one
<markvdb>
don't see them in the wiki, and haven't actually fiddled with one yet
<Astralix>
rotation and cropping can be done by the GPU
<markvdb>
...if the support for the gpu is there
<markvdb>
that's a big question mark for me
<Astralix>
There is additional picture handling support in the SOC
<markvdb>
how well supported is it in software?
<Astralix>
It surely is, but how to get the information from RK...
<markvdb>
I mean, I see the jpeg enc/dec hardware mentioned
<Astralix>
You need to find out which IP they bought from whom and then you can ask for a license to get the sourcecode
<markvdb>
but that doesn't buy me anything, does it
<markvdb>
Astralix: we're a small non-profit running FOSS exclusively
<Astralix>
yes it does. The more work you can hand over to a specialized hardware, the more time you have for your own software and the less CPU power you neeed
<markvdb>
so no chance of doing that
<markvdb>
from a technical perspective, you're right
<markvdb>
from a legal/human perspective, we don't even want to go there
<Astralix>
I am not sure if there is "no chance"
<markvdb>
we can't afford to spend much time on it in any case
<Astralix>
I did some support for mainlining the RK3xxx devices and just by reading original open source kernels code, legally available, we found out that RK did it just the normal way
<markvdb>
now that's inspiring of course
<Astralix>
They just bought an ARM core and arranged deveral other IPs aound it, they bought from certain companies
<Astralix>
several
<markvdb>
I was just referring to "having to ask them nicely" as a big no-no
<Astralix>
So ther is some IP from Designware, some SiliconImage
<markvdb>
if the information is already there in the wild, that's a whole different story
<Astralix>
Yes, but as you see there is no magic inside this RockChips even the company tries to keep up the myth.
<Astralix>
It is assembled just by usual things you can buy, in this case, just well known IPs assembled to a SOC.
<markvdb>
yeah
<Astralix>
So just find the matching kernel driver and you're in
<markvdb>
first step for us would probably be to benchmark existing postprocessing code on the cpu
<Astralix>
So why don't you support this community and buy a radxa rock board?
<Astralix>
It already has 2GB/8 or 16GB, 2x Host and 1x OTG USB, Ethernet and much more
<markvdb>
you think that board has the most community buyin?
<markvdb>
then that might be a good first step indeed
<Astralix>
I don't know the numbers but I can ask. All in know is, that it was created by ideas collected from this chat
<markvdb>
I see
<markvdb>
I was thinking of either odroid-u3 or radxa rock
<markvdb>
for testing hardware performance - they should be in the same ballpark
<Astralix>
And in this chat are some guys having good hardware knowledge, some have good software knowledge and a few even have access to datasheets RK never would give to them
<markvdb>
:-)
<markvdb>
and you lot generally run the radxa rock right?
<Astralix>
There is even some connections to the people who ported picuntu to the board
<Astralix>
I personally program anything that is not intel and not windows
<Astralix>
But yes, I do have a radxa and it's serial number is probably very short :)
<markvdb>
then it's obvious to me which one we should play with
<markvdb>
thank you for your hints
<markvdb>
Astralix: btw, can you confirm to me that one core can address all of the memory available?
<Astralix>
I did not read any limitations on that, but I can double check
<markvdb>
sounds good
<Astralix>
So I still cannot find any limitation per core. Sure the SOC inherits a System Coherency Unit (SCU) that coordinates caching and data access between all memories and the four cores. It si needed to keep data coherency for the four cores
<markvdb>
cool!
<Astralix>
The registers available for the SCU look like allowing to lock in any core to a certain memory and you can add a cpu to the coherency team or keep it outsinde
<Astralix>
-n
<markvdb>
cool
<Astralix>
What is the target application on that Parallella board?
<RaYmAn>
"supercomputing on the cheap" is a pretty good tagline for it, lol
<RaYmAn>
honestly though, I think the parallela board is 100% a way to get publicity and funding by getting hackers interested => getting noticed
<Astralix>
Yes it looks really interesting. Keep that in my mind, if I am getting asked on something like this.
<Astralix>
I personally look for a 4-Core ARM that has up to 4 USB3.0 or even better SATA.
markvdb has quit [Ping timeout: 272 seconds]
akatik has joined #linux-rockchip
markvdb has joined #linux-rockchip
markvdb has quit [Quit: Page closed]
VargaD_ has joined #linux-rockchip
VargaD has quit [Ping timeout: 248 seconds]
VargaD_ is now known as VargaD
markvdb has joined #linux-rockchip
FreezingCold has joined #linux-rockchip
fefifofum has joined #linux-rockchip
<fefifofum>
hi, any idea why enabling CONFIG_ENABLE_DEFAULT_TRACERS in defconfig completely breaks the kernel?
<fefifofum>
i'm trying to build 3.0.36 for rk3066
<fefifofum>
without that option, kernel boots but cannot load mali module
ganbold_ has quit [Remote host closed the connection]