An almost naked Pyra

[Swordfish II]

Well, as mentioned, Drastic runs fully playable with hires rendering on my OMAP4 phone (at least Okamiden and some other games I tried), so it would be weird if the vastly faster OMAP5 couldn't do it.

And ran at nearly full speed (99%) on the dev board...

 

[levi]

Basically, any electrical conductor is also a good thermal conductor.

Don't ask me why exactly.

It's because of all those free electrons zipping about over the surface. Heat one up and it'll speed off and knock into a few others, and speed (heat) them up too. It's not always true though; strong regular lattices are also effective at transferring heat vibrations, so electrically insulating diamond conducts heat better than electrically conductive metals like copper or silver.

 

[Exophase]

It's because of all those free electrons zipping about over the surface. Heat one up and it'll speed off and knock into a few others, and speed (heat) them up too. It's not always true though; strong regular lattices are also effective at transferring heat vibrations, so electrically insulating diamond conducts heat better than electrically conductive metals like copper or silver.

Guess you've solved the question of what to use in Pyra's thermal solution

[doublepost=1456286043,1456285896][/doublepost]

And ran at nearly full speed (99%) on the dev board...

It was over full speed, although we don't yet know what the real speed with a proper screen solution will be.

This is with a relatively demanding game and no frameskip. Since we're talking hires rendering a little bit of frameskip would make a huge difference.

A small number of pathological games like Golden Sun may still given issues but for the most part it should be okay so long as the OMAP5 can really sustain something not terribly far away from 2x1.5GHz.

 

[oskda]

Well, there were a few smartphones that used the Cortex-A15

I said "NEAR NOBODY" choosing Cortex-A15:

Searching phones with Cortex-A15 cores: only 15 found: http://www.gsmarena.com/results.php3?sFreeText=cortex-A15

Searching phones with Krait cores: 245 found: http://www.gsmarena.com/results.php3?sFreeText=krait

Something were wrong with Cortex-A15 cores and its SOC were delayed, and manufacturers had problems with heat and energy with Cortex-A15, as a result Qualcomm, with its custom Krait core, dominated the market and Cortex-A15 SOC (both were for high-end phones) where minimum, and in some cases only available with a next semiconductor iteration (with more advanced manufacturing allowing to be less power hungry).

As I said I have been following mobile phone industry for a very long time, and I have seen those problems with Cortex-A15. It was a FLOP.

As I said I understand why you choose this OMAP 5 (you tell us about this history) with Cortex-A15: you choose TI because other manufactures doesn't support Linux, doesn't supply (or even answer) for such small quantities, or doesn't offer good support at all. So we end with TI, and OMAP 5 was the highest TI SOC (TI is abandoning CellPhone industry so it has no better SOC for that matter).

Now I would choose Intel Atom x5 or x7 (Cherry Trail 14nm vs 28nm in OMAP5: x4 integration) as they are designed for smartphones (and tablets) and they had an Intel HD GPU: it means OPEN DRIVERS and best Linux support. Intel has excellent support for its products, support Linux, and has open drivers in this SOC (other Atom SOC uses PowerVR GPU so those have no open drivers); and Intel is desperate in mobile industry so they would be happy to show Pyra can do.

With those SOC you don't need rotartor chip (money, space, power and problems we read about initialization), as you can use graphic hard in Intel SOC.

I prefer ARM design (over x86 desing) but Intel has some GPUS with open driver, and that is more important point (I don't like closed in ARM SOC, like PowerVR case).

I understand they were late for Pyra as this Atom x5/x7 were released in Q1/Q2 2015. Althought I don't know if there are other older Atom SOC with Intel/open_drivers SOC (as most Atom SOC for smartphones used PowerVR GPU, with closed drivers).

Even today I don't understand why you need rotator chip in crrent OMAP 5: it has a 2D-Graphics Accelerator (Vivante GC320), so I suppose it can do image rotation, as this is a basic transformation granted for today hard (every smartphone or tablet has it).

I understand you wanted ARM so you can reuse a lot of work from Pandora (it is also ARM), but I think this is a minor plus, as Linux has more work done in x86.

Umm - that's exactly where these problems occur
The prototypes are the first real produced systems ...

Well, I don't see totally in the same way: I think you can't design a case and after that design interior circuit. I think you must test interior circuit before, because you can find there is too much hot and you need a bigger sink/radiator, or you need more power or diferents circuits to supply that bigger power, etc, etc.

The SoC has nothing to do with switching to a 720p display, btw, the reason was that I tried a 1080p display and everything was too small to be used. While Linux desktops have a dpi setting, not all programs respect those, so e.g. GIMP is unusable until you manually change some configuration files with a text editor.

A 1080p display with that size would have had too many drawbacks and wouldn't really have given us any advantages.

I know SOC has nothing to do with 720p switching: I read in this board it was rotator chip blame, as it doesn't support 1080p, but only up to 720p.

The worst part is that in a future time you can't find that rotator chip as it is withdrawn

I own a phone with 5 inches LCD IPS display: it is very nice, but I have checked it and I can see clearly edges/steps in lines (for that purpose I have created an image with pure straight lines in different angles an colours). So it is not so high resolution (my next phone will be QHD, and this has bigger resolution than my eyes can see: a 1080p is a bit bellow, and a 720p much much bellow).

I think 1080p would be very nice for some emulators, as with those resolution you can emulate scanlines, do better scaling and get a better result.

About it is too small for Linux desktop: well, 5 inches is too small for a standard Linux desktop, even if you choose 720p or if you choose 1080p: in both cases you have to tweak.

I'm pretty sure I could get a Rockchip or Allwinner somewhere, but they are pretty much fixed on Android whereas the TI SoCs even have mainline Linux kernel support.

I know, and because of that I prefer Intel with its own GPU: best support/documentation and OPEN DRIVERS.

We don't want to end up with outdated kernels at some time.

I suppose we don't want to end with closed drivers GPU, or drivers we can't update to next kernel version... but we may end with this in OMAP 5 with its PowerVR GPU.

Well, my guess is you didn't understand the post correctly.
This is not an issue with the SoC, it's an issue with the power circuit right now.

I understand, but I know Cortex-A15, in its own, were a flop and very problematic, so much that most cellphone industry jumped over it and Qualcomm ate most high market in that time.

In that way I was fearing problems with power or heat.

At least, Pyra has a very mice upgradable CPU board system And I sippose that a next time you can'y use TI, because TI doesn't make SOC for smartphones/etc (it withdrawn that market).

I know it is very very early to ask in this time, but I will ask What new CPU/SOC do you think you would use in CPU board upgrade (viewing the SOCs we have now in market and presented for next release?

 

[fusion_power]

As I said I have been following mobile phone industry for a very long time, and I have seen those problems with Cortex-A15. It was a FLOP.

Couldn't you have told us these informations a liiiiitle bit earlier, like 2-3 years ago?

Now I would choose Intel Atom x5 or x7 (Cherry Trail 14nm vs 28nm in OMAP5: x4 integration) as they are designed for smartphones (and tablets) and they had an Intel HD GPU: it means OPEN DRIVERS and best Linux support. Intel has excellent support for its products, support Linux, and has open drivers in this SOC (other Atom SOC uses PowerVR GPU so those have no open drivers); and Intel is desperate in mobile industry so they would be happy to show Pyra can do.

These recent Atoms sound good on paper, never dealt with these kind of chips sadly. I guess they need more power than the OMAP5, especialy x7 but I think the x3 is already pretty powerful and is the most power saving one from the series I guess.
No clue how good the Atom graphics unit is compared to the one ftom the OMAP5, hard to compare I guess (like everything between x86 and ARM).

 

[EvilDragon]

I said "NEAR NOBODY" choosing Cortex-A15:

Searching phones with Cortex-A15 cores: only 15 found: http://www.gsmarena.com/results.php3?sFreeText=cortex-A15

You should include big.little as well, as they use the A15 for higher processing as well, only fall back to the A7 when that power is not needed.

As I said I have been following mobile phone industry for a very long time, and I have seen those problems with Cortex-A15. It was a FLOP.

I know, but it won't help much. Our choices were a Cortex A9 or Cortex A15.
Period.
Everything else wasn't possible.

Now I would choose Intel Atom x5 or x7 (Cherry Trail 14nm vs 28nm in OMAP5: x4 integration) as they are designed for smartphones (and tablets) and they had an Intel HD GPU: it means OPEN DRIVERS and best Linux support. Intel has excellent support for its products, support Linux, and has open drivers in this SOC (other Atom SOC uses PowerVR GPU so those have no open drivers); and Intel is desperate in mobile industry so they would be happy to show Pyra can do.

Not yet - afaik Linux doesn't work well on these tablets right now.
But it would be an option for the future.

And that's exactly why we have an upgradable CPU Board.
As small company, we don't have the manpower to produce a completely new product within a short time.
But we don't have to redesign everything from scratch now - we simply can change the CPU Board.

With those SOC you don't need rotartor chip (money, space, power and problems we read about initialization), as you can use graphic hard in Intel SOC.

Space is not an issue, initialization has been solved and the price is 2 EUR afair.
But sure, would be nice to have something where it wouldn't be needed.

I prefer ARM design (over x86 desing) but Intel has some GPUS with open driver, and that is more important point (I don't like closed in ARM SOC, like PowerVR case).

Yes, though I hope ARM GPUs could get more open now if Intel steps into the mobile market.

I understand they were late for Pyra as this Atom x5/x7 were released in Q1/Q2 2015. Althought I don't know if there are other older Atom SOC with Intel/open_drivers SOC (as most Atom SOC for smartphones used PowerVR GPU, with closed drivers).

One problem was the BIOS, which you needed to design yourself with older SoCs. And that 32bit UEFI makes it almost impossible to use with Linux.

The new ones are worth to take a look at though, and they can be bought by anyone without any contracts.

Even today I don't understand why you need rotator chip in crrent OMAP 5: it has a 2D-Graphics Accelerator (Vivante GC320), so I suppose it can do image rotation, as this is a basic transformation granted for today hard (every smartphone or tablet has it).

Because it's not a dedicated one where you can rotate the monitor output alltogether.
You can use it within the system, but every program would need to be coded to use it, so you couldn't just use a standard Linux.
Not just one driver would need to be coded - but every program, etc.

Smartphones usually render using the API of the phone, which often uses OpenGLES to rotate. But Linux programs don't do that.

Linux would still have a 720x1280 display, and with the rotator chip, Linux sees a 1280x720 display and can use it out of the box.

Well, I don't see totally in the same way: I think you can't design a case and after that design interior circuit. I think you must test interior circuit before, because you can find there is too much hot and you need a bigger sink/radiator, or you need more power or diferents circuits to supply that bigger power, etc, etc.

It wasn't designed before, they were designed at the same time, where changes to the PCB had been made if the case needed some pegs and vice versa.

I think 1080p would be very nice for some emulators, as with those resolution you can emulate scanlines, do better scaling and get a better result.

That works well on 720p screens already, I've done that on the GPD XD.

About it is too small for Linux desktop: well, 5 inches is too small for a standard Linux desktop, even if you choose 720p or if you choose 1080p: in both cases you have to tweak.

I tried both and found the 720p to be usable out of the box.
You might want to increase some fonts, if it's too small for you, but you don't need to manually change configs.

In GIMP, with 1080p, even the biggest button preset was too small, so it needed config changes.

And 1080 needs a lot more processing power - so the disadvantages are higher than the advantages.

I know it is very very early to ask in this time, but I will ask What new CPU/SOC do you think you would use in CPU board upgrade (viewing the SOCs we have now in market and presented for next release?

I'm in contact with HiSilicone, but I also closely follow Rockchip and Intel ones.
So who knows... but let's first finish the main system.

 

[matzesu]

Lets talk about the CPU Upgrade,
My usecase for the Pyra will be the Pandora Emus i allready know to work great at my CC Unit (PSX, GBA, DS, SNES etc, plus some more CPU Hungry things like dreamcast..)
Also PC Ports like OpenMorrowind, and Doom 3..
Normal PC Tasks which the Pandora also runs well (Exel Word Powerpoint via Libre Office) Internet Browser, Youtube)..

So will i still be happy a long time whit Omap 5, or do i have to upgrade at some times??
And how will this work whit the Pyra Ports, if there are lot diverent CPU Boards in Future??

You would have to make lots of diverent ports, for every cpu board on market if you want all Pyra Users play your game..

 

[Exophase]

I said "NEAR NOBODY" choosing Cortex-A15:

Searching phones with Cortex-A15 cores: only 15 found: http://www.gsmarena.com/results.php3?sFreeText=cortex-A15

Searching phones with Krait cores: 245 found: http://www.gsmarena.com/results.php3?sFreeText=krait

Something were wrong with Cortex-A15 cores and its SOC were delayed, and manufacturers had problems with heat and energy with Cortex-A15, as a result Qualcomm, with its custom Krait core, dominated the market and Cortex-A15 SOC (both were for high-end phones) where minimum, and in some cases only available with a next semiconductor iteration (with more advanced manufacturing allowing to be less power hungry).

As I said I have been following mobile phone industry for a very long time, and I have seen those problems with Cortex-A15. It was a FLOP.

As I said I understand why you choose this OMAP 5 (you tell us about this history) with Cortex-A15: you choose TI because other manufactures doesn't support Linux, doesn't supply (or even answer) for such small quantities, or doesn't offer good support at all. So we end with TI, and OMAP 5 was the highest TI SOC (TI is abandoning CellPhone industry so it has no better SOC for that matter).

Now I would choose Intel Atom x5 or x7 (Cherry Trail 14nm vs 28nm in OMAP5: x4 integration) as they are designed for smartphones (and tablets) and they had an Intel HD GPU: it means OPEN DRIVERS and best Linux support. Intel has excellent support for its products, support Linux, and has open drivers in this SOC (other Atom SOC uses PowerVR GPU so those have no open drivers); and Intel is desperate in mobile industry so they would be happy to show Pyra can do.

If your argument is that it's bad because it wasn't used much in phones then how can you favor Atom x5 or x7 instead? Where does that show up on GSMArena? The last generation, Silvermont, also shows up in far fewer phones than Cortex-A15.

Similarly, Cortex-A17 shows up in only 2 phones there, certainly a major flop for phones but I think it'd work pretty well for something like Pyra too.

A15 did have efficiency problems and I don't think anyone really did a good job iterating those away in later designs except Samsung (with Exynos 5430). But the real issue, the real reason why Krait dominates the search on phones, is that at that time Qualcomm cornered the market. TI and ST-Ericsson were already squeezed out of the running entirely, nVidia was reduced to near irrelevance in phones, and Intel didn't manage to take much. This would have happened regardless of what CPU they used. At the time Samsung stuck with Qualcomm heavily, like they probably will again with the Snapdragon 820 even though they now have a custom CPU in Exynos.

The other relevant players like MediaTek just weren't high end enough to have A15 or A57 SoCs suitable for phones. That'll probably start changing with A72.

You can see there are way more A57 devices out there, is that because A57 had much better implementations than A15? No, it's because Qualcomm had to use it and their implementation was notoriously terrible, I would say it had significantly worse efficiency than any A15 SoC we had, and that efficiency was also very bad for the A53 part so it was a huge mess. But it was used a fair amount anyway because Qualcomm. Phone makers aren't using Qualcomm because of their amazing CPU prowess, it's because of their modem integration and a bunch of red tape surrounding it.

For what it's worth I think Atom would probably be better for Pyra at this point too but that ship has sailed. The 14nm SoCs weren't even out when these decisions were being made and the BayTrail ones has other issues. ED did actually talk to Intel about this. The options were just not good.
[doublepost=1456603293,1456603084][/doublepost]

I'm in contact with HiSilicone, but I also closely follow Rockchip and Intel ones.
So who knows... but let's first finish the main system.

Kirin 950 would be amazing! I would have never even considered that to be possible. Good luck with those talks.

 

[EvilDragon]

Kirin 950 would be amazing! I would have never even considered that to be possible. Good luck with those talks.

Do you happen to know how Linux support (also 3D drivers, etc.) Is for that?

 

[Exophase]

Do you happen to know how Linux support (also 3D drivers, etc.) Is for that?

Sorry, I don't really have any idea..

All I know about HiSilicon is that they're owned by Huawei, and I expect that the Kirin SoCs are only meant for Huawei phones. But they make a bunch of other SoCs for STBs and network equipment and CCTVs and stuff, and they have some documents for them on their site so they must not be intended only for Huawei products. They have some of their Linux driver stuff up here: https://github.com/hisilicon

This Linaro promoted board uses a fancy server-oriented HiSilicon SoC and runs Linux:

http://open-estuary.org/d02-2/

But I think it has no GPU anyway. Still at least something.

They use Mali and AFAIK ARM has been providing releases for Mali userspace blobs on Linux so there might be realistic avenues here.

 

[levi]

So will i still be happy a long time whit Omap 5, or do i have to upgrade at some times??
And how will this work whit the Pyra Ports, if there are lot diverent CPU Boards in Future??

Unless the A15 turns out to overheat so you can't even run PS1 for any length of time without it cooking and throttling back, or has some weird bug a little like the way some software doesn't work on CC units but worse, I'd expect many people to stick with the A15 like people have stuck with CC units even while there's another new CPU or two out. It should do everything you listed, but maybe you'll see some software released that you'd like to try that won't run so well on your A15 cores.

But yeah, I'm also concerned about the extra work multiple different revisions will have on devs. Perhaps it's less of an issue that it would have been with devices like the GP2X those years ago, but with the Pandora, most stuff is compiled down from something portable, so it's just a matter of adding an extra cross-compile target and you've got a new DBP. It will mean more testing, but thankfully the community here is quite useful for that I guess. I'd expect less optimisation, like Notaz's PS1 DSP work, if you can just buy a new CPU and run everything faster, but we'll see.

 

[EvilDragon]

But yeah, I'm also concerned about the extra work multiple different revisions will have on devs. Perhaps it's less of an issue that it would have been with devices like the GP2X those years ago, but with the Pandora, most stuff is compiled down from something portable, so it's just a matter of adding an extra cross-compile target and you've got a new DBP. It will mean more testing, but thankfully the community here is quite useful for that I guess. I'd expect less optimisation, like Notaz's PS1 DSP work, if you can just buy a new CPU and run everything faster, but we'll see.

Most of the work should be kernel and low-level drivers.
The rootfs should be the same for all of them.

 

[comradekingu]

Where does Cherry-Trail have official Linux support? Not to people who actually make products based on them. Some do it unofficially, some don't.

Going beyond Cherry-Trail, Broxton with Skylake seems to require loading firmware blobs https://lists.freedesktop.org/archives/intel-gfx/2015-June/068167.html

Not a good upgrade path then, and I dont think AMD, (who now has ARM on low end), is about to get more agressive on the low end X86, considering intel is throwing huge amounts of subsidies at that particular field. It might stop doing so.

https://www.linux.com/news/embedded-mobile/mobile-linux/837216-linux-41-speeds-up-intel-atom-soc
https://en.wikipedia.org/wiki/List_of_Intel_Atom_microprocessors#.22Cherry_Trail-T.22_.2814_nm.29
Cant find intel sources.
The point remains that while good on paper, intel atom systems require a lot of cancer that its convenient to forget about.


HiSilcon has their own phones too, they are the most promising vendor.
Other than freescale finally getting on with it and releasing something new in the powerful department.
(Newer than the excellent i.MX6 Quad)

Nvidia Tegra X1 is maxwell, so firmware blob required https://en.wikipedia.org/wiki/Tegra#Tegra_X1

What most often gets forgotten is that just because the GPU blazes syntethic bench du jour, doesnt mean that the SoC is free of modems that cant be turned off.

Samsung Exynos 8 Octa 8890 with Mali T-880 and Kirin 950 lists on Wikipedia as integrated LTE, dont know if it comes without. Same with Qualcomm 820 and MediaTek Helio X20.

So if we look at No powerVR, No integrated modem:

------------------------------------------------------------

Rockchip RK3288 https://en.wikipedia.org/wiki/Rockchip#Tablet_application_processors_.28ARMv7-A_and_later.29
is the only thing i can find with nonPowerVR graphics, looking more performant than OMAP5, but then again not by a huge amount.

Mediatek MT8163 https://en.wikipedia.org/wiki/MediaTek#Standalone_application_and_tablet_processors

Allwinner A64 https://en.wikipedia.org/wiki/Allwinner_Technology#H-Series_.28OTT_box_application.29

This is probably available ↑

Exynos 5 Octa](Exynos 5800) https://en.wikipedia.org/wiki/Exynos

So the very high-end, like apples current non-offering, is off limits.

------------------------------------------------------
Anything intesting at MWC that I overlooked?

 

[oskda]

If your argument is that it's bad because it wasn't used much in phones then how can you favor Atom x5 or x7 instead?

Really you are totally lost in my words.

I said manufacturers mostly didn't choose Cortex-A15 BECAUSE it had lots of problems; in response EvilDragon said there were a few smartphones with Cortex-A15 and I said him it is 15 vs 245 (searching in PhoneArena data).

So, my argument isn't what you say --"it's bad because it wasn't used much in phones"-- but the opposite: it was bad, was delayed, had lot of problems with heat and power, and because of all that it was very little used, while other ARM equivalent in power (Qualcomm with Krait cores) ate near all market.

then how can you favor Atom x5 or x7 instead? Where does that show up on GSMArena? .... hows up in far fewer phones than Cortex-A15.

I thin you continue lost here: Atom can't go well on smartphones market because they use x86 ISA, while most smartphone market use ARM ISA. Even Intel had have to build an ARM emulator (really it is a binary translator) so it can run native ARM code (some Android programs runs native ARM code and other, although don't use native code on their own, use native code indirectly as they use libraries with native ARM code).

I don't know if those Intel will be good or bad, but we can't say it is bad counting smartphones using it, as they have near no market for smartphoes: a market dominated by ARM. In fact Intel is desperate in mobile.

I know Intel support is superb, and I know this Intel SOC with Intel GPU (not PowerVR like earlier Atoms) have open drivers.

the real reason why Krait dominates the search on phones, is that at that time Qualcomm cornered the market

And Qualcomm dominated (high-end market) with Krait because the equivalent ARM, the Cortex-A15, was a FLOP with lots of problems. Even Samsung manufactured some Exynos with Cortex-A15 but in so low volumes that it was only for some markets, while the same Samsung phone with Qualcomm was shipped to most of world.

You can see there are way more A57 devices out there, is that because A57 had much better implementations than A15?

Cortex-A15 was a flop and Krait (Qualcoomm) ate most high-end market (near all).

With Cortex-A53/A-57 the history and logical reasoning is completely different: this time the flop was with Qualcomm: Qualcomm didn't have own 64-bit cores on time, so they had to use 64-bit ARM cores Cortex-A57/A53. At the same time this ARM cores were good so there is a lot of manufacturers using them without problem (except in Qualcomm 820 that using x4 Cortex-A57 have some thermal problems).

Now Qualcomm has developed its 64-bit high performance cores: Kyro (used in Snapdragon 820).

I think Atom would probably be better for Pyra at this point too but that ship has sailed

We agree: ship has sailed and there is no option going back now.

Kirin 950 would be amazing

It has lots of power, but the same problem other ARMS: it has closed drivers GPU. If I prefer Intel SOC (with Intel GPU, no rebranded PowerVR GPU) is because it have open drivers for GPU.

 

[levi]

Most of the work should be kernel and low-level drivers.
The rootfs should be the same for all of them.

Even if you opt for an x86 chip in a later revision?

But yeah, I hadn't really considered the driver work needed to pull a new board up from its bootstraps, fair point.

 

[TrashyMG]

Even if you opt for an x86 chip in a later revision?

well a different rootfs would be needed in that case, but we still can use the same Debian base packages, just x86 vs armhf

 

[Exophase]

And Qualcomm dominated (high-end market) with Krait because the equivalent ARM, the Cortex-A15, was a FLOP with lots of problems. Even Samsung manufactured some Exynos with Cortex-A15 but in so low volumes that it was only for some markets, while the same Samsung phone with Qualcomm was shipped to most of world.

Cortex-A15 was a flop and Krait (Qualcoomm) ate most high-end market (near all).

With Cortex-A53/A-57 the history and logical reasoning is completely different: this time the flop was with Qualcomm: Qualcomm didn't have own 64-bit cores on time, so they had to use 64-bit ARM cores Cortex-A57/A53. At the same time this ARM cores were good so there is a lot of manufacturers using them without problem (except in Qualcomm 820 that using x4 Cortex-A57 have some thermal problems).

Now Qualcomm has developed its 64-bit high performance cores: Kyro (used in Snapdragon 820).

Here is the basic problem I have with what you're saying: you're making the assumption that Qualcomm dominated most of the high-end market because of problems with Cortex-A15. Yet Qualcomm continued to dominate much of the market with their A57 SoCs despite having similar but worse problems vs early A15 ones. Cortex-A57 was actually used by fewer SoC manufacturers than Cortex-A15 was, and Qualcomm's implementation was exceptionally poor. Much worse than the implementations of A15 that Samsung or nVidia had, especially the later ones.

Bottom line is Qualcomm dominates for reasons that have to do with a lot more than CPU.

 

[oskda]

Where does Cherry-Trail have official Linux support? Not to people who actually make products based on them. Some do it unofficially, some don't.

Cherry Trai with Linux: http://linuxgizmos.com/latest-intel-compute-sticks-use-skylake-and-cherry-trail-cpus/
http://liliputing.com/2016/01/intel...tick-with-atom-x5-cherry-trail-processor.html


[doublepost=1456651779,1456651605][/doublepost]

You should include big.little as well, as they use the A15 for higher processing as well, only fall back to the A7 when that power is not needed.

They are included in the 15 search result (the search gives results containing "Cortex-A15" even if they are only cores or big.little)
[doublepost=1456652309][/doublepost]

Couldn't you have told us these informations a liiiiitle bit earlier, like 2-3 years ago?

EvilDragon explained why they chose TI and OMAP. I suppose you read those post. And they wanted ARM to reuse Pandora code. I suppose there were little to chose on.

No clue how good the Atom graphics unit is compared to the one ftom the OMAP5, hard to compare I guess (like everything between x86 and ARM).

I suppose it will be much more capable as it is a much recent SOC, but even if it weren't, the most important thin is that in theory Intel HD GPU have open drivers while PowerVR (and all GPUs in ARM ecosystem) are closed.

If I like Intel with this SOC it is only for its GPUs having open drivers (previous Atom SOC use PowerVR GPU, rebranded by Intel, so they have the same closed drivers we see in ARM).

On other side Intel is knwon for giving good Linux support.

 

[WizardStan]

Couldn't you have told us these informations a liiiiitle bit earlier, like 2-3 years ago?

We knew this 2-3 years ago. There was no other option, ED tried really hard. Did you forget the hundred-page-war about SoC options? That every so often someone well meaning but unfortunately ignorant posts into?
The simple fact is there is nothing unknown about the dangers of using the OMAP5, it was discussed to death and many alternatives were suggested, all of which had to be shot down as impossible: the OMAP5 was the only option.
There are options now, probably even a few months back, but restarting design now, or even 6 months ago, would've pushed things out and ED is already taking a huge risk.
Development times being what they are, before you finish one design there will always be something better just come out, or just about to, before you finish: sooner or later you need to stick your design and say that it's final, even if something better comes along.
That is literally why ED chose the CPU board design though, so when it's time for an upgrade the development path is shorter. But it's still not short enough to not have a major impact on the very first product.

 

[oskda]

Here is the basic problem I have with what you're saying: you're making the assumption that Qualcomm dominated most of the high-end market because of problems with Cortex-A15. Yet Qualcomm continued to dominate much of the market with their A57 SoCs despite having similar but worse problems vs early A15 ones. Cortex-A57 was actually used by fewer SoC manufacturers than Cortex-A15 was, and Qualcomm's implementation was exceptionally poor. Much worse than the implementations of A15 that Samsung or nVidia had, especially the later ones.

Bottom line is Qualcomm dominates for reasons that have to do with a lot more than CPU.

1/ In first place I was talking about core/cpu dominance (Cortex-A15 vs Krait etc), no about brand dominance (Qualcomm vs Exynos etc).

2/ « Cortex-A57 was actually used by fewer SoC manufacturers than Cortex-A15 was ».

The technology situation is differente in those periods: in the time of Cortex-A15 and Krait you have to use one of those SOC (most used krait because Cortex-A15 was late in real SOC and a flop) to obtain very good results; today there are a lot of phones using Cortex-A53 multicores with excelent result, so using Cortex-A57 SOC is only for very high-end phones.

But if you want to compare without considering those historical situations go ahead:

Its even worse: there were more Cortex-A15 SOC (6 SOC) but used by FEWER PHONES (15 Cortex-A15 phones) while there were less Cortex-A57 SOC (4 SOC) but used for much more phones (33 Cortex-57 phones).

In the search I made on cellphones using Cortex-A15 I only found 15 phones using Cortex-A15:

HiSilicon made 2 Cortex-A15 SOC Kirin 920 and 925 used in 2 phones each
LG made 1 only Cortex-A15 SOC Nuclun used only in 1 phone
Samsung Exynos made 3 Cortex-A15 SOC used like in 6 phones (most of them only realeased in little countries, while the same phone with Krait was realeased for most contries).

It is very 'strange' that LG only used its Cortex-A15 in one only phone, and HiSilicon only used its 2 Cortex-A15 SOC in only 2 phones each

Where those own Cortex-A15 so bad/problematic that they decided no to use in more phones?
Instead of using their own Cortex-A15 SOC they run to search for Krait SOCs.
[doublepost=1456657571,1456656022][/doublepost]Yesterday I was talking about a Atom x5/x7 for Pyra CPU board, and today I have found a Indiegogo proyect with a palmtop computer very similar to Pyra but with Windows 10 (I don't like if you can't install Linux too) and with Atom CPU:
It seems a totally Pyra plagiarism :O