Potential X86 (Non ARM) SoC Alternatives


I would still like us to consider having Pandora successor use an Intel z3770 SoC.

So far it appears viable. I'm still curious if ED had a chance to check availability and feasibility from his end.

Perf/watt: check

Overall power: check

Overall heat produced: check

Unit cost: $37 - check

Available from suppliers other than the manufacturer: check

Likely to be available over a multi-year lifecycle: check

Linux FOSS drivers, including GPU: check

Android drivers: check

So, what's missing to make this a viable candidate?
 
Okay.  To be fair it looks like they use it for the ODROID-X2, and as a means to sell such things to second parties based on their FAQ.
Just because they put in their FAQ questions about using it in third party designs doesn't mean that was the sole motivation for why they did it. Putting the cards in different products would have been at least another one. But it doesn't really matter why Hardkernel did it, the only reason I mentioned them is to give an illustration of the sort of board I was talking about. Not to use them as a business example.

However they haven't elected to do that with their new product line based on the Exynos 5, and the ODROID-U2 SBC is less expensive then said module is by itself.  In terms of the SBC industry that module doesn't make a compelling argument, and the fact they didn't update that part of their product line speaks to how successful their business related to being a middleman selling integrated SoC modules has been.
It makes less sense for SBCs because they've gotten small to the point where they're not that much larger than the central board would be.

Pandora isn't an SBC. It has a much bigger PCB so it can accommodate things like dual full-size SD slots, a keyboard, gaming controls, high profile connectors like full USB and so on. The cost/analysis tradeoff for this is different.

Proposing that you can do the equivalent of drop an i7 into a POWER7's Blade and all you'll need is a little daughterboard with no issues with chipsets and power rails is a rather different proposal then being able to drop that i7 into a board that's designed for it.
If that's what you think this is equivalent to then you don't understand what it is at all. In this scenario the "chipset" is part of the SoC. The PMIC is on the daughter board and no high voltage sources are needed, you will be able to run any ARM SoC + RAM or Intel SoC + RAM off of the same power rail - the battery (~3.7-3.8V) or USB/wall (5V). Everything conceivably platform specific is constrained to the board itself. It is NOT a board that just breaks out the SoC connections blindly.

This is more like separating the computer part from a set of I/O over a standard set of interfaces that anything would be able to provide. slaeshjag already described the interfaces involved. And don't ask me to explain the advantages after I already did so - I'm sorry if this is contradicting ED but I don't agree with him that there are zero advantages. And please spare me your crap about moving goalposts, you sound like a forum argument bot. I didn't move anything.
 
Last edited by a moderator:
I would still like us to consider having Pandora successor use an Intel z3770 SoC.


So far it appears viable. I'm still curious if ED had a chance to check availability and feasibility from his end.


Perf/watt: check


Overall power: check


Overall heat produced: check


Unit cost: $37 - check


Available from suppliers other than the manufacturer: check


Likely to be available over a multi-year lifecycle: check


Linux FOSS drivers, including GPU: check


Android drivers: check


So, what's missing to make this a viable candidate?
arm isa  :rolleyes:
 
I've seen some people complain that the decoding units of x86 CPUs are extremely large and complex due to that, eating up a lot of space on the die and therefore burning quite a lot of energy, which is gaining more significance with such ultra mobile targets as you don't have a lot of possibilities to simplify it like the rest of the CPU.
People complain a lot of things that are either out of context or were never really correct to begin with. No offense, but someone could quote you and pretty much use the same justification - but that's not very meaningful o nits own. If you look at the comments of actual CPU architects who have worked on x86 as well as other things (for example Mitch Alsup) you'll hear something pretty different from this.

Yes, x86 decoders are more expensive than decoders for other ISAs are, but this is a very relative cost. Back in the days of the 8086 it was a huge deal and why you got a much more efficient ARM2 processor in a similar number of transistors. But those transistor counts were thousands of times smaller than what they are today. Decoders grew in individual complexity and replication (not just for x86, ARM decoding is substantially more complex than it was in the ARM2 days too) but not at the same rate as everything else - when you add all of the other plumbing necessary to support a wide out-of-order execution engine, branch prediction and other speculation, deep cache/memory hierarchies with aggressive prefetching, fat SIMD units, and so on the more complex ISA at the front-end is less and less significant.

It's still a big deal for something like microcontrollers, which is one reason why Quark is not the least bit impressive. Moving up to Cortex-A5 or A7 level, it may still be a small deal. At the BayTrail vs Cortex-A15 level it's only a minor expense. You have to balance it out with everything else, which means considering Intel's other advantages like access to more advanced manufacturing and a lot of experience hand-optimizing CPU layouts and resources. This can easily outweigh the ISA disadvantage.

The obvious result of having a x86 CPU would be that many users would just bring over binaries from other distros and playing proprietary games that are available for x86 Linux or even via Wine - and those binaries will suffer from that fact, as they are usually optimized for ancient x86 CPUs to gain max compatibility. Maybe an alcohol-induced fuzzy thought got the better of me at that statement, though.
The non-x86 alternative here would mean not having access to those binaries at all. How would that be better?
 
In this scenario the "chipset" is part of the SoC. The PMIC is on the daughter board and no high voltage sources are needed, you will be able to run any ARM SoC + RAM or Intel SoC + RAM off of the same power rail - the battery (~3.7-3.8V) or USB/wall (5V). Everything conceivably platform specific is constrained to the board itself. It is NOT a board that just breaks out the SoC connections blindly.

This is more like separating the computer part from a set of I/O over a standard set of interfaces that anything would be able to provide. slaeshjag already described the interfaces involved. And don't ask me to explain the advantages after I already did so - I'm sorry if this is contradicting ED but I don't agree with him that there are zero advantages. And please spare me your crap about moving goalposts, you sound like a forum argument bot. I didn't move anything.
Lets give this some thought from a physical device perspective. I'm going to be making some very broad assumptions - please excuse me if these are not exact.

The daughter board with the SoC and miscellaneous bits to make it all work and get to the pin-outs is going to have a relatively modest footprint. Is 25mm X 25mm enough?

The daughter board will need to have a corresponding socket on the motherboard. Would a notebook DIMM module connector have enough pins? Using the existing Pandora as an example... The keyboard connections all reside on the top side of the motherboard. That side cannot be made thicker or have 'stuff' on it. This socket would have to reside on the bottom side of the motherboard - resulting in the device thickness issue that ED brought up.

I have a rather radical idea to solve both this and the thermal issues.

Put the SoC daughter board in the case lid with a metal heat sink in the shape of the Pandora logo (or fins or whatever) exposed through it. In this scenario, the display is driven directly from the daughter board. Or is it the motherboard now since it has the SoC on it?

Connectors then go through the hinge (or next to it in a flat ribbon) to the lower portion of the unit containing the keyboard, ports, battery, etc...

The lid would need to be somewhat thicker in the middle. Device edge thickness could remain about the same. Newer screens are thinner and designed to have the SoC riding right behind them - so that may save some.

It would have a 'hot spot' on the center of the lid - which would be the best place to have a hot spot for heat dissipation.

This should keep the main motherboard in the base relatively simple. Or is it a daughter board in this scenario? It gets the traces for keyboard, nubs, ports, LEDs, radios, etc... that all get run via a relatively simplified/generic ribbon cable to the SoC board in the lid.

Thoughts?
 
The daughter board with the SoC and miscellaneous bits to make it all work and get to the pin-outs is going to have a relatively modest footprint. Is 25mm X 25mm enough?
It'd probably have to be substantially wider than that, at the very least.

The daughter board will need to have a corresponding socket on the motherboard. Would a notebook DIMM module connector have enough pins? Using the existing Pandora as an example... The keyboard connections all reside on the top side of the motherboard. That side cannot be made thicker or have 'stuff' on it. This socket would have to reside on the bottom side of the motherboard - resulting in the device thickness issue that ED brought up.
A 72-pin SO-DIMM connector strikes me as too limiting. ED's thickness issues were with particular modules he looked at, not a custom made one. The added height from using something like Hardkernel's form factor is very minimal, for example. They didn't just use them ODROID-X2, they also used it with this for example: http://www.hardkernel.com/renewal_2011/products/prdt_info.php?g_code=G132988773094 http://www.hardkernel.com/renewal_2011/products/prdt_info.php?g_code=G129689935155&tab_idx=3 If the thickness is acceptable for those form factors doing it in something like Pandora shouldn't be a problem - you most likely need to clear more for the USB connector.

I can't think of anything that'd cause the thermal situation to be worse from this.

I have a rather radical idea to solve both this and the thermal issues.

Put the SoC daughter board in the case lid with a metal heat sink in the shape of the Pandora logo (or fins or whatever) exposed through it. In this scenario, the display is driven directly from the daughter board. Or is it the motherboard now since it has the SoC on it?

Connectors then go through the hinge (or next to it in a flat ribbon) to the lower portion of the unit containing the keyboard, ports, battery, etc...

The lid would need to be somewhat thicker in the middle. Device edge thickness could remain about the same. Newer screens are thinner and designed to have the SoC riding right behind them - so that may save some.

It would have a 'hot spot' on the center of the lid - which would be the best place to have a hot spot for heat dissipation.

This should keep the main motherboard in the base relatively simple. Or is it a daughter board in this scenario? It gets the traces for keyboard, nubs, ports, LEDs, radios, etc... that all get run via a relatively simplified/generic ribbon cable to the SoC board in the lid.

Thoughts?
One of the thing ED rightfully wants to get away from is a big fat ribbon connector going through the hinge, hence why he'd prefer a high speed serial display connection that just needs a few wires. What you're saying goes quite a bit in the opposite direction. No matter what it'd make the screen part at least a little thicker too, and if the minimum thickness of the base is dictated by the connectors you'll be looking at a thicker device..
 
Last edited by a moderator:
No, you're not getting it. I've mentioned Hardkernel's module approach several times, rohezal linked to it.. here it is again.. http://www.hardkernel.com/renewal_2011/products/prdt_info.php?g_code=G135270682824

You can see the connectors are in the bottom, it isn't an edge connector socket. It's actually not uncommon to see phones have little boards stacked like this, containing things like wireless or SIM PCBs.
 
OK - thank you. I get it now. It doesn't add a -lot- of thickness, but it would add some when compared to using a surface mounted SoC.

I don't think that the full sized ports are the limiting factor in general feature height on the motherboard.

Thinking in 3 dimensions... The side-on cross section of the Pandora motherboard is essentially L shaped with the stem on the L being the thickness required for the full sized ports on the back. The battery and stylus live in the gap/block defined by the L. Adding a mm or two there could push into the battery and stylus space.

Don't get me wrong - I'd love to see this work.

Here's a thought - I'm not sure if it is feasible.

During motherboard design, mark off a rectangular section in the front/middle as 'SoC'. Run all of the traces for the rest of the features on the board that interact with the SoC to the edge of that rectangle. Keep everything outside of that rectangle, 'generic'. Re-design just the area inside the rectangle to match whichever SoC is being used in that product run so that it matches out to the other features.

Modular based architecture on a single board - just change out the contents & construction of the SoC rectangle.

It would not be as efficient on board space. But would it increase flexibility enough to be worth it?
 
Okay, ignore the full size connectors on that side then, look at the SD cages on the other side. The edges not covered by the battery definitely doesn't clear those. A very low profile daughter board could be made at worse slightly taller than those cages, I don't think it'd add a mm or two.

This isn't about simplifying PCB design, making some partition in it doesn't buy you anything, it'd just pointlessly restrict your layout.. without the ability to actually physically replace the SoC there's no point.
 
Last edited by a moderator:
Thinking in 3 dimensions... The side-on cross section of the Pandora motherboard is essentially L shaped with the stem on the L being the thickness required for the full sized ports on the back. The battery and stylus live in the gap/block defined by the L. Adding a mm or two there could push into the battery and stylus space.
your assuming the Pandora case or parts of the case are going to be used for the successor? I don't suppose the case needs to house a stylus anymore (assuming they settle for a capacitive screen). There are plenty of design decisions to make the module fit. The "it makes it bigger" is no valid argument, really. For instance you could make the screen lid and the keyboard smaller to save some hight or use a wider but more flat battery (using the obsolete stylus space) etc.
 
Last edited by a moderator:
Bay Trail in action - or at least reported on.

http://www.pcgamer.com/2013/09/11/i-played-team-fortress-2-on-a-tablet-and-it-didnt-suck/

Link is a few months old, but it was new to me so I thought I would share.

The z3770 can apparently run Team Fortress 2 well enough to play at 2560x1440. It should do all of the HL1&2 derived titles easily at 1080p.

Everything in the HL1&2 series, with the exception of Portal 2, is available on Steam for Linux. All play well in multi-player mode.

I know it looks like ED is likely to go ARM for a Pandora successor. Frankly, if someone came out with a Pandora form factor (nubs & all) z3770 powered handheld clamshell with 4GB RAM and a 1080p screen, I'd buy that instead - even if it's double the price. It would open up content that I'd really like to have.
 
Bay Trail in action - or at least reported on.

http://www.pcgamer.com/2013/09/11/i-played-team-fortress-2-on-a-tablet-and-it-didnt-suck/


Link is a few months old, but it was new to me so I thought I would share.


The z3770 can apparently run Team Fortress 2 well enough to play at 2560x1440. It should do all of the HL1&2 derived titles easily at 1080p.


Everything in the HL1&2 series, with the exception of Portal 2, is available on Steam for Linux. All play well in multi-player mode.


I know it looks like ED is likely to go ARM for a Pandora successor. Frankly, if someone came out with a Pandora form factor (nubs & all) z3770 powered handheld clamshell with 4GB RAM and a 1080p screen, I'd buy that instead - even if it's double the price. It would open up content that I'd really like to have.
impossble to render at 1440p sir, 720p 20-30 at best for source games

http://www.youtube.com/watch?v=eBlv29R_0so

gameplay videos here

http://www.youtube.com/watch?v=mNWXPo-e_gc

here are some videos

http://www.youtube.com/user/th3drow/videos

http://www.youtube.com/user/BrueComputing/videos

if one wants to sell a boat load of P2s just market it capable for playing league of legends at playable fps OTG!

http://www.youtube.com/watch?v=MX6LwVDf5XI

dell venue 8 pro is very small...

http://www.youtube.com/watch?v=1yNbXzWUgDI
 
Last edited by a moderator:
Am i correct in assuming that Asus T100 doesnt run linux? http://www.phoronix.com/scan.php?page=news_item&px=MTQ5NzE

32bit EFI isnt implemented on Linux, and wont be, is that the problem?
 It can be run. http://liliputing.com/2013/10/booting-ubuntu-asus-transformer-book-t100.html
Another one:

http://www.happyassassin.net/2013/11/24/the-fedlet-revived-or-fedora-linux-on-a-dell-venue-8-pro-bay-trail/

The 32bit EFI does seem to be a major annoyance though. The chip is 64bit. They've announced that 64bit Windows versions are 'on the way'. That doesn't mean that they'll change the UEFI to 64bit though. You should be able to boot a 64bit OS from a 32bit UEFI.

I'm not sure why there isn't Linux support for 32bit UEFI. There will probably have to be, eventually. It looks like there are viable work-arounds though.
 
Back
Top