SoC: Back and forth!


I don't think this community would necessarily disintegrate when we switch to x86. But I do think that there would be a significant pressure to somehow get Windows and Windows games to work on the device, legaly or illegaly. There would probably be lots of people coming to the boards with questions like "where can I find a torrent for the cracked Windows?" and "help! <insert Windows game name> does not work! please make it work NOW!". If that happens, I'll have to resign as a mod since I don't feel like modding that mess.
Separate forums, separate mods?
 
Last edited by a moderator:
The Snapdragon 805 is announced in Q2/2014, so that would throw us back about 6 - 8 months... not feasable, I think.
Aren't the first /products/ announced for Q2? Wouldn't that mean that producers can already get the chip?
 True, but only the chip. I'd still like to test all the hardware using a devboard upfront.

We don't have a high  R&D cash fund... (unless someone wants to sponsor us ;) )

I can wait.   :)   I can spend more time with my Pandoras :)
 My company has monthly costs.

You want to see it go bust before the Pyra is finished?
So - realistically we're looking at SoCs that you can easily obtain today in quantities of 1-5000 with complete data sheets?
 And Linux drivers available and - if possible - a design guide, because it makes it a lot cheaper and faster to design the PCB.
 
Last edited by a moderator:
Well, at least ED titled the thread appropriately :D


I think ED is right in weighing up the options in a balanced way. Im sure ED will be able to do this in an unbiased way. Weigh up what we want for the Pyra in terms of its goals against what each SOC offers.


The only thing I'll say about the OMAP5 is this, there won't be an OMAP6. TI are already cutting back that division.


What does that mean if we were to go with OMAP5 now for future generations of the Pyra ?


Well lets look at it. I have seen talk of the NEON thats implemented for OMAP5 (S800 NEON is slightly different apparently) and using it in emu's etc . Also using its 2D acceleration - which other SOCS like the Snapdragon 800 don't have apparently according to ED.


With the above in mind. Imagine if we go with OMAP5 today and in 5 years time we have a whole bunch of apps and emu's and games optimised for it. What happens when ED needs to upgarde the Pyra and launch a Pyra 2 , you will have to do it on a new non TI SOC. Any optimisation done for the particular flavour of NEON and 2D accelaration on the OMAP5 is now a problem. You will at that point(hopefully ) have lots of apps , emu's and games etc that you would then need to migrate to a new non TI SOC.


Doesn't it make sense to change now and avoid creating a bigger headache for ourselves down the track?
 
Last edited by a moderator:
NEON is a command set and therefore (mostly) the same and compatible for all ARMs that offer NEON.

However, the Snapdragon one is pretty slow.

Other SoCs don't have that issue.

The 2D accelerator is mostly part of the driver (i.e. we can implement it into the SDL library or X) so it's nothing that will be optmized in the program.

So it's no issue that there won't be an OMAP6
 
I can wait.   :)   I can spend more time with my Pandoras :)
 
My company has monthly costs.


You want to see it go bust before the Pyra is finished?
So - realistically we're looking at SoCs that you can easily obtain today in quantities of 1-5000 with complete data sheets?
 
And Linux drivers available and - if possible - a design guide, because it makes it a lot cheaper and faster to design the PCB.
From what I can tell, the z3770 family is the same as the N2820 - they're both Bay Trail (Silvermont) architecture.

http://www.cpu-world.com/Compare/767/Intel_Atom_Z3770_vs_Intel_Mobile_Celeron_N2820.html

Intel has a small footprint computer based on the N2820.

http://www.amazon.com/Single-Pack-DN2820FYK-NUC-BOXDN2820FYKH0/dp/B00HUYLZ7Q

$150 + RAM + HDD

I wonder if it would be worthwhile to pick one up to see how 64bit Debian Linux runs on Bay Trail.  From what I can tell, it should run right out of the box with a current kernel.  The driver base between the z3770 and the N2820 should be the same for most things - though the N2820 has 'more stuff' hanging off of it.  Still, they have the same CPU architecture and the same graphics hardware.  They both 'turbo boost' to the same clock rate on single threaded stuff.  Would there be any benefit in the analysis?

Edit:

Wow - just went out looking for how to install Linux on this thing - I'm contemplating getting one to fiddle with - and ran across this:  http://www.phoronix.com/scan.php?page=article&item=intel_baytrail_nuc&num=1

Looks good so far.  They installed kernel 3.14 and found joy.
 
Last edited by a moderator:
EvilDragon, on 07 Feb 2014 - 5:22 PM, said: A TI OMAP5 developer mentioned to me, that one of the design guides they had for the OMAP5 was 2W maximum power usage. Not sure if they really did that (we need to measure it), but 6W sounds like quite a bit, only for the core. 6W does sound like a lot - but if that is true, wouldn't it be 'peak' with all 4 cores tapped out? Does the SoC self-throttle under thermal protection - or would we be able to throttle it back and only -give- it a max 4W? I've seen the 4W number thrown around about the OMAP too - but we seem to have even less information about it's true thermal requirements.
Actually, if you look at the OMAP5 datasheet and check the table 3-2 (maximum current ratings), you'll see that the OMAP can even consume a lot more peak power than 6 W. For example, for the vdd_mpu domain a maximum current of 6460 mA has been specified, which at 1.5 V supply voltage is well above 10 W. If you would just stupidly add up the maximum currents for the other power domains, you would end up with even more than that.

Point being, those maximum ratings are theoretical values and don't tell you much about the actual power consumption. You'll need them for properly designing the circuit, but they are far from realistic numbers for average power consumption. Real peak power consumption for Cortex-A15 cores in typical SoCs is about 8 W, but this does not matter much for both battery runtime and cooling system requirements.

By the way, on page 112 the OMAP5 datasheet states a TDP of 4 W. I don't know where the 2 W figure comes from, but those 4 W seem more realistic given the TDP of other Cortex-A15 based SoC.
 
Last edited by a moderator:
And the big question is, how much does it consume overall, i suppose that is a scaling and powersave game, but base values, on doing something but not all out, i should think would favour the omap5
 
There's an interesting thing to consider... do any of the CPU's under consideration have a pin-compatible successor planned? i.e. something that would allow a future upgrade without more than a minor change in board layout, like with the DM3730 in the Pandora?

- Neelix
 
There's an interesting thing to consider... do any of the CPU's under consideration have a pin-compatible successor planned? i.e. something that would allow a future upgrade without more than a minor change in board layout, like with the DM3730 in the Pandora?- Neelix
yeah, i was going to ask that. But if the OMAP5 is available for 5 years it does not really make much sense to just do an upgrade later on. A renewed model would be better
 
There's an interesting thing to consider... do any of the CPU's under consideration have a pin-compatible successor planned? i.e. something that would allow a future upgrade without more than a minor change in board layout, like with the DM3730 in the Pandora?- Neelix
yeah, i was going to ask that. But if the OMAP5 is available for 5 years it does not really make much sense to just do an upgrade later on. A renewed model would be better
Doesn't mean we have to use it for 5 years.

OMAP 3 is also still available and we don't use it any more.
 
Doesn't mean we have to use it for 5 years.

OMAP 3 is also still available and we don't use it any more.
I see your point, but I don't think OMAP3 to that new chip power isn't as big asPandora to Pyra. Other words, there's enough power to warrant a whole new design.
 
From what I can tell, the z3770 family is the same as the N2820 - they're both Bay Trail (Silvermont) architecture. http://www.cpu-world...eron_N2820.html
Sort of, but there are some differences. You can follow z3xxx story on phoronix a little (the story about asus t100 netbook). It's not muck there except he couldn't install linux on it due to uefi issues.

There is a thread on xda-developers (http://forum.xda-developers.com/showthread.php?t=2500078) about putting linux on this device. In short there are problems with uefi that could be worked around, network and sound driver problems because everything is connected via sdio. It's not a problem for a device that would be designed for linux like pyra.

The biggest problem so far is the lack of any video hardware acceleration as intel driver refuses to work on a system. Guys at Intel are supposedly working on this problem but who knows when it'll be done and well tested. I think by the time next-gen atom is out driver problems will be sorted out but for now going with z3xxx SoC seems like a gamble.

So for pyra to be created as fast as possible OMAP5 is the best choice for now. It may change in a year or half a year but from my understanding of the situation the pyra design process starts now.

And after pyra there'll be much more reasons to go x86 as even on pyra we won't need as much optimized software as we need now and all emulators that we hopefully could run on a handheld in future like pcsx2 are developed and optimized for x86 and emulators for modern systems need much more manpower to optimize for some other architecture. Manpower I don't think we have in our community.
 
The biggest problem so far is the lack of any video hardware acceleration as intel driver refuses to work on a system. Guys at Intel are supposedly working on this problem but who knows when it'll be done and well tested.
If you would have followed the thread a little further, you would have noticed, that someone there actually has sucessfully used the Intel driver with the Z3770 GPU on X.

I think by the time next-gen atom is out driver problems will be sorted out
What makes you think that? In recent years, Intel has actually been pretty quick with releasing driver updates and supporting new hardware on Linux. Also, the GPU in that Z3770 is from the Ivy Bridge generation, so supporting that chip really should not be that difficult for them. Also, as I said above, it looks like it actually works okay.
 
Back
Top