Potential ARM (Non X86) SoC alternatives


Grench

Forum Addict!
Joined
Oct 3, 2008
Messages
6,629
There has been a lot of fear and controversy surrounding the recent presentation and consideration of the Intel X86 based Z3770 as a potential alternative for a Pandora successor.

Let's leave all of that for another thread(s).

For this thread, let's discuss part availability, capabilities, performance, performance per watt for the ARM alternatives.

Later, after one or more ARM champions are found, in another thread, we can discuss the relative strengths and weaknesses of all viable potential solutions.

I am not an expert on any SoC.  I've seen lots of product announcements over the last few months for Apple, Samsung and other ARM SoCs, but parts availability and datasheet type information and benchmarks seem to be a bit scarce.

So, ARM proponents, what's out there that we should be looking at?

Mods - if you have the time or inclination, please edit the thread title?

Maybe: Potential ARM (Non X86) SoC alternatives.  I should have put ARM in there, but can no longer edit the thread title.

Thank you.
 
Last edited by a moderator:
As far as on-topic, I would love the Exynos 5 Octa, but the previous thread about it showed that, while it is ARM, it spits out heat like an x86. So it too would require a fan or extensive heat sync, which, as I understand it, the low-power operation of an ARM chip is one of the main draws of ARM basically completely negating any appeal of the Exynos 5 Octa. Shame, really.
 
Mods - if you have the time or inclination, please edit the thread title?
You can edit the title yourself by clicking Edit next to your post, then clicking the "Use Full Editor" button. Then you can clarify whether you're talking just ARM or all Non-X86 SoCs
 
Here are the ones that aren't an option:

Samsung: Originally said they'd give ED Exynos 5250 (but not anything nicer/newer) then retracted it because they decided it's too much of a risk.

Qualcomm: Doesn't even talk to someone like ED

MediaTek: Said no way to Linux drivers happening

Ones that haven't been ruled out:

nVidia: Haven't been approached yet. Normally they require big orders too but there have been exceptions like OUYA (bear in mind that's something they're also directly sponsoring)

TI: Says they'll sell OMAP5 but without any technical support

Rockchip: Haven't been approached yet. They're in lots of cheap stuff, you can possibly get them through distributors.

I would classify anything not on a currentish manufacturing process (TSMC 28nm, Samsung 32/28nm, Global Foundries 28nm, or Intel 22nm) as something you shouldn't get. Meaning I'd avoid TSMC 40nm or Chinese fabbed stuff like SMIC.
 
I understand the reason for ruling out MediaTek, but I'm not sure ED has even tried to talk to Qualcomm yet. I know they normally don't express an interest in small orders, but I don't think it should be immediately out of the question to at least ask. As for Samsung, since they have been really swingy, is there not some chance that they may flip back into a more favorable position later?

As for the others:

Something from nVidia would be nice. Would they be likely to have open drivers? I'm sure with TI we would likely run into those problems we did before with the GPU closed drivers.

-God Ginrai
 
I'll let ED speak for himself on who he feels is ruled out then, or correct my list.. As for Samsung, do you really want to pursue a design around a product from a company that promised they'd give you one in the past, then changed their mind? Sounds like a pretty bad idea to me. I can see it now, you get samples and datasheets from them and make engineering boards, then 10 months later you're ready to start mass production and go to place an order for chips and suddenly they tell you nevermind.

No one in the mobile space has official open GPU drivers in a product right now. BayTrail will be the first (it's not quite in a released product yet). Even Intel's next phone SoC (Merrifield) will probably still use IMG and still be closed.
 
Last edited by a moderator:
I'll let ED speak for himself on who he feels is ruled out then, or correct my list.. As for Samsung, do you really want to pursue a design around a product from a company that promised they'd give you one in the past, then changed their mind? Sounds like a pretty bad idea to me. I can see it now, you get samples and datasheets from them and make engineering boards, then 10 months later you're ready to start mass production and go to place an order for chips and suddenly they tell you nevermind.
Everything I have seen EvilDragon say about Samsung has given me the impression that he talked to Samsung Germany, they said "Neat! We'd love to give you something. Let us talk to Korea.", then came back saying "Korea said no. :("

I've never seen any mention of being promised something and then it being taken away.

-God Ginrai
 
Mods - if you have the time or inclination, please edit the thread title?
You can edit the title yourself by clicking Edit next to your post, then clicking the "Use Full Editor" button. Then you can clarify whether you're talking just ARM or all Non-X86 SoCs
Thank you - I wasn't aware you could edit thread titles through the advanced editor.  Done.
 
Everything I have seen EvilDragon say about Samsung has given me the impression that he talked to Samsung Germany, they said "Neat! We'd love to give you something. Let us talk to Korea.", then came back saying "Korea said no. :("

I've never seen any mention of being promised something and then it being taken away.

-God Ginrai
I don't know, they were saying things kike "we can give you Exynos 5250 but not 5410" and "we'll be willing to provide an entire platform" which sounded like a process was already starting. I didn't get an impression of them expressing a need to get more permission, although I agree that's probably what really happened. They sounded like they sold on something they had no authority to sell on.

The explanation was that Samsung feels that not only are they risking ED coming back to them begging them for support help when things go wrong, but they even think that if the product has problems it'll make them look bad. And that could include a fear of not delivering preorders. Frankly it's kind of hard to see why Samsung or Qualcomm even would support this product, the volume just does nothing for them financially. They're just not setup for this because they lack distribution.

That doesn't really mean Intel will be any different, but at least the netbook variants of Atom go on sale at Mouser now. BayTrail-T (for tablets), M (netbooks), and D (nettops) stand a good chance of all being the same chip this time around, so that helps odds a bit. Getting the phone chip (Merrifield) but be less viable.
 
Last edited by a moderator:
In the interest of keeping the conversation polite and on topic, lets not even consider X86 Vs ARM, but discuss them each in their own threads until we have a better grasp of what the options are within each architecture type?

I.e. lets not send this thread down the hole the previous 4(+?) went?
 
Exo, do you think BayTrail or Merrifield could be superior (leaving the architecture question aside) compared to OMAP5, Rockchip, Tegra3 etc?
Well like I said, I don't even think Tegra 3 should be on the table. RK3188 is better anyway. But yes I think it's better. Compared to OMAP5 it has two more cores, probably slightly more peak single-threaded performance (but it depends on what it can be clocked at) and a lot better power efficiency. vs RK3188 it has much more single-threaded performance although the core count is the same. GPU should be stronger than both too. Tegra 4 could be comparable in both peak CPU and GPU performance but the power consumption is going to be a lot worse for CPU - chances are you'll be thermally limited before it can reach that peak, just like the current tablets and phones are vs Shield. So unless Pandora 2 plans on being that thick and sporting a fan it'll be the same scenario.

OMAP5 raises the really serious red flags of "why has no one announced a product using this"? Either it's not competitive or it's still not going to be ready for anything any time soon. For what it's worth, TI said it'd be out in products by the end of last year.

In a long run a Tegra SoC might benefit from nVidia's reluctant approach of open source: http://linux.slashdot.org/story/13/09/24/1818222/nvidia-begins-releasing-documentation-for-nouveau
But none of that transferred to their mobile processor GPUs. Maybe it'll be relevant by the time Tegra 5 rolls out but that's not going to be terribly soon.
 
* raises hand.  I'm confused.

From Exophase's explanation it seems like each of these SoCs might have different strengths.

Would it be possible, or beneficial to rank order the ARM contenders multiple ways?

High - low single thread performance

High - low multi thread performance

Low - high maximum W

High - low performance per W
 
Frankly it's kind of hard to see why Samsung or Qualcomm even would support this product, the volume just does nothing for them financially. They're just not setup for this because they lack distribution.
Frankly, I don't see why they wouldn't. (aside from the risk factors you mentioned) It may be smaller quantities than they normally sell to, but they mass produce these things. They could easily make a mass production run that they can split the quantities up to sell to smaller-run products. It's not like ED is trying to get a special chip made specifically for the Pandora, like Apple does for their iPhone.

-God Ginrai
 
Frankly, I don't see why they wouldn't. (aside from the risk factors you mentioned) It may be smaller quantities than they normally sell to, but they mass produce these things. They could easily make a mass production run that they can split the quantities up to sell to smaller-run products. It's not like ED is trying to get a special chip made specifically for the Pandora, like Apple does for their iPhone.

-God Ginrai
Have you read any older posts I made about this?

These companies are not structured for distribution. There's a serious cost overhead involved for each client they enter an agreement with (which is always a hand-tailored thing). They schedule to deliver some number of trays that have a large number of chips on them to customers. Everything about their business is likely to be structured around a large quantity of units with a one to one relationship with a particular order, including testing/verification. Breaking off some small number of chips from a bigger order, to set aside specially for some small player whom will bring them little revenue, is not trivial.

Most likely the only reason why Samsung was even offering Exynos 5250 (which has shown up in Chinese tablets) is because Samsung mobile (who makes the inhouse phones and tablets) didn't want to use it for more than one product (Nexus 10) and they have surplus sitting around that they need to get rid of. Think about why they would say upfront that they'd give Exynos 5250 but their superior 5410 was off the table.
 
Last edited by a moderator:
"RK3188 is better anyway. But yes I think it's better. Compared to OMAP5 it has two more cores"

no it has not. OMAP5 has two Cortex-A15 plus a DSP (which one? C64+, C67 (w/ FPU?)), plus a Cortex-M4 for standby / powersave mode. I like that design.

EDIT: seems like it's the C64+ according to http://www.ti.com/pdfs/wtbu/SWCT010.pdf.
 
Last edited by a moderator:
That makes sense that the SK Samsung would then say, 'No' if they knew that the overstock on those was going to get eaten up by another contract.

That does bring up another important factor to consider though.  Continuity of product line.  TI's SoC used the Pandora's SoC was in lots of embedded systems on long-shelf-life contracts.  Mobile SoCs dedicated for use in phones seem to be unlikely to be stocked past the generation of phone they were built for.  I.e. if the manufacturer of the SoC only intends it for use in one generation of phones, they're likely to create enough for that + fail rate and not stock 'replacements' or spare parts for order.  

It feels like there is a very short (in months) manufacturing run on mobile SoC types, then the stock is used up right as the next generation chip goes into production.  Is that a misconception?

Should we also have a requirement for the SoC to be likely to exist (or have a pin-compatible successor) and be purchasable 2 years out from system design?

I think we lucked out with the 1Ghz part being able to nearly drop in place of the original 600Mhz CC board design.  It gave what was essentially 2 generations of product with nearly the same MB design.  What are the odds that there would be a future SoC that is pin and process/program compatible with whatever our choice is?  So many of these seem to be there then gone in 6 months.

Does a Tegra 4 have the same pin-out as a Tegra 3?
 
Trying to help...  Looking them up on Wikipedia.

http://en.wikipedia.org/wiki/Rockchip

RK31xx series[SIZE=small][[/SIZE]edit]

Rockchip has announced the RK31xx series for 2nd quarter of 2013. The RK3188 is based on ARM's Cortex A9 design and features a quad core CPU clocked at 1.8-2.0 GHz frequency on a 28 nm process.[12] In April 2013, Rockchip showed the RK3168, a very low power dual core Cortex A9 based CPU, also manufactured using the 28 nm process.[13]

RK3188 28 nm HKMG ARMv7-A Up to 1.8 GHz Quad-coreARM Cortex-A9 Mali-400 MP(Quad core, 533 MHz) LPDDR2, DDR3, DDR3L support, up to 2 GB 2013 GoTab GTQ97,[25] Cube Pea II,[26] Cube U30GT2,[27] CloudnetGo CR9,[28] iMito QX1,[29] PIPO M8pro,[30] PIPO M9,Rikomagic MK802 IV,[31] Ugoos UG802B, Ugoos UG007B, Ugoos MK809 III,[32]Ugoos QC802,[33] Measy U4B,[34] Minix X7,[35] Tronsmart MK908,[36] Tronsmart T428,[28] Measy U4B,[34] Freelander PD800,[37] FNF iFive x2 [38] Vido Mini One, JXD S7800b[39]

http://en.wikipedia.org/wiki/OMAP

OMAP 5[SIZE=small][[/SIZE]edit]

The 5th generation OMAP, OMAP 5 SoC uses a dual-core ARM Cortex-A15 CPU with two additional Cortex-M4 cores to offload the A15s in less computationally intensive tasks to increase power efficiency, two PowerVR SGX544MP graphics cores and a dedicated TI 2D BitBlt graphics accelerator, a multi-pipe display sub-system and a signal processor.[40] They respectively support 24 and 20 megapixel cameras for front and rear 3D HD video recording. The chip also supports up to 8 GB of dual channel LPDDR2/DDR3 memory, output to four HD 3D displays and 3D HDMI 1.4 video output. OMAP 5 also includes three USB 2.0 ports, one USB 3.0 OTG port and a SATA 2.0 controller.

Model number Semiconductor technology CPU instruction set CPU GPU Memory technology Availability Utilizing devices OMAP5430 28 nm ARMv7 1.5, 1.7 GHz[41] dual-core ARM Cortex-A15 (and two Cortex-M4microcontrollers) PowerVR SGX544MP2 @ 532 MHz[42] + dedicated TI 2D BitBlt graphics accelerator 32-bit dual-channel 532 MHz LPDDR2 (8.5 GB/sec)[43] Q2 2013 Jorjin APM-5

The first difference that jumps out at me is the difference in the GPU.  Mali-40 MP Vs. PowerVR SGX544MP2.  Are there Linux drivers for either?  Both?  FOSS?

They're both LPDDR2.  The latest Samsung and Apple ARM devices are LPDDR3 - which is apparently much faster and more efficient?  Do we care?

Measuring or finding perf/W for the two might get interesting - they have significantly different multi-core structures.  Would we be able to take advantage of the Cortex-M4 low power cores in the OMAP5430?  Does the A9 in the RK3188 ratchet/scale down to a low power standby type mode?

Edit:  Argh - the fancy HTML formatting that I stole from Wikipedia for those tables got eaten on submit.  Reference the links to find them in a prettier format.
 
Last edited by a moderator:
Back
Top