My Thoughts on the Pyra

[Saber]

Now I don't want to insult anyone but...

I very mixed feelings about the current state of the pyra's hardware. On the one hand I love the fact that the pyra is continuing to use almost exactly the same design the pandora had. I love the pandora's design and I can't envision a more perfect design for a handheld device. Also, I actually like the fact that the pyra is being forced to use a 720p screen, as 1080p puts way to much stress on the gpu for the image quality increase it provides on that size of screen.

I understand that ED is trying as hard as he can, and the reason we're getting an OMAP was because absolutely no one else would even talk to him. That said, the pyra's cpu is very weak by today's standards, and will be completely obsolete by the time the pyra comes out. The pyra's website advertises that it "has a state-of-the-art mobile CPU." Seriously? That's a bald-faced lie! Even most one year old smartphones have better cpu's than the OMAP that the pyra is going to be using. If we look at the pyra's closest competitor (not that that's very close), the nvidia shield, it has cpu that is twice as powerful as the pyra's cpu despite being almost a year old! Also, the shield has double the ram. However, my biggest disappointment was learning that there is an exactly 0% chance of there ever being an x86 version of the pyra simply because intel doesn't even offer atoms as socs. An x86 pyra would, in my eyes, be infinitely superior to an ARM version, even if the battery life would be much shorter. Almost all games I would want to play on the pyra are closed source and x86. Even if the pyra could easily change cpus to a better ARM chip later because of the daughter-board, is feasible to actually do so because of the software side of things? And remember the outrage that just the suggestion of a 1 ghz model of the pandora caused? Also, is there any reason why we can't have 4 gb of RAM?

Also, I would like to see a trackball as either a replacement for the right nub or as an external sort of thing that attaches to the side. Face it, any sort of analogue nub or joystick will always be inferior to the precise input of mice and trackballs. There's a reason console games always have ridiculous auto-aim. While the touchscreen has the desktop covered, it can't be used for any first or third person games. I use a thumb trackball instead of mouse normally, and for me it would be awesome to have one on the pyra too. 

I disagree.

The Nvidia Shield is a beast as far as number boasting is concerned, yes, but it's a weighty one at that at over 20oz and bulky. Try sliding one of those sea turtles into a pocket. The Pyra though is plenty powerful for real world use and a highly versatile device too(multiple USB/SDXC ports, superior gaming controls, backlit keyboard) with the added advantage of this long established community behind it. Also, the nubs have been vastly improved(precision, center clicking) and one of the biggest advantages in my eyes is the owner of a Pyra is not forced into being a money conduit to the Android marketplace for most of the software, although you could run Google's OS(I believe) if you so choose to.   

As for ED "trying hard" to bring the Pyra to the market, that's a severe understating of the logistics involved. It's a staggering amount of work to assemble the pieces just right into a final product. There is a reason we don't have a multitude of full, compact handheld computers with built-in gaming controls with backlit keyboards out in the world. 

 

[Silent-Hunter]

I think ED's absolutely right, android only need 2gb of ram is because android is inefficient. Apple's products never need that much ram even they come with the so called retina display.

Yes, all the so called next gen socs have 8 cores, but no people are going to use that in many years. Even big companies struggle with multicore optimization, I would assume Pyra's devs would not even bother with a quad core setup.

Pyra's as powerful as it needs to be if it's coming out next year. (cough, cough, I hope ED can push pyra out in 2015, not 2018)

Pyra has a quad core setup, two A15s, and two M4s.

 

[Grench]

About the RAM:


I ran KDE4, GIMP, FireFox, two PSX Emulators and a GameBoy emulator at the same time - and had about 600MB RAM left on the devboard with 2GB RAM. More RAM needs more power and is more expensive - so I wonder: what would you need 4GB for?

I want to run my Pyra with desktop, web and game applications and -two- Android 'virtual machines' (one for business/work apps and one for games) and an emulated X86 command line desktop - maybe running Desqview to multitask DOS applications there too.

CPU is 32bit, so anything over 3.5 GB is a waste.  Can I have 3GB RAM?  What is the price jump from 2GB to 3GB?

http://www.tomshardware.com/news/Samsung-3GB-LPDDR3-20nm-class-4Gb-LPDDR3-memory-module,23662.html

Yeah - my use case scenario is extreme - but not totally unrealistic.  Keep in mind that using today's cards the device could hold 740GB of data internally (256+256+128) and 1016GB if you add on microSD cards in tiny adapters the USB ports.  Just querying the file structures could be a lot of data to process.  If end users even used 'spare' RAM to cache their storage reads, there could be some advantages.

So - what is the max RAM that the OMAP5 can address?  How much more would it cost to 'max it out'?

 

[Askarus]

So - what is the max RAM that the OMAP5 can address?

8 GB

Some sort of banswith expansion so that 32 bit machines can access 8 GB.

How much more would it cost to 'max it out'?

A lot of Money.

Don't get me wrong, I'm the "more is better" person.

I still would like 4 gb but only of it is that sort of RAM that can be powered down if it's not used.

And remember: The RAM is also on the PCB board.

If we might get the A80 upgrade then we can always use more RAM when it got cheaper.

 

[WizardStan]

CPU is 32bit, so anything over 3.5 GB is a waste

32bit but with 40bit addressing, plus I'm 99.9% certain that since the video RAM is the system RAM, there's no need for the 500MB hole that plagues x86 systems, it just gets allocated as needed like any other RAM, so the full 4GB would be available (minus what the video driver is actually using).

So - what is the max RAM that the OMAP5 can address?

Theoretically I think the physical limit is 1tb (40 bit address), but each application still only has access to 32 bits (4GB) at a time.So more practically you could have 8GB with individual applications able to consume up to 4GB.

 

[Granitehead]

More RAM needs more power and is more expensive

 How much more would 4GB cost and how much more power usage would it mean (in relation to 2GB and to the complete power usage of the Pyra)?

A probably even more difficult problem is that you seem to need a lot of separate chips: Look at the RAM in your computer, it probably has at least 8 RAM chips on it which it combines. Maybe 16.The picture of the SoC board there (http://boards.openpandora.org/topic/16499-some-pcbs-holidays-a-new-case-revision-and-a-theme/) shows places for 8 identical chips. I assume that's the 8 RAM chips you need for 2GB of RAM. I don't think 16 of them could fit. Maybe there would be more expensive higher density chips, maybe not.

(yes, I doubt any smart phone has 8 RAM chips in it. Smart phones tend to use RAM which is actually supposed to be used in mobile devices. It uses less power and probably comes in chips containing as much RAM as the smart phone needs, i.e. 2GB or 4GB. OMAP5 doesn't support such RAM because it's not a processor for mobile products)

Edit:

I think ED's absolutely right, android only need 2gb of ram is because android is inefficient. Apple's products never need that much ram even they come with the so called retina display.

Yes, all the so called next gen socs have 8 cores, but no people are going to use that in many years. Even big companies struggle with multicore optimization, I would assume Pyra's devs would not even bother with a quad core setup.

Pyra's as powerful as it needs to be if it's coming out next year. (cough, cough, I hope ED can push pyra out in 2015, not 2018)

Pyra has a quad core setup, two A15s, and two M4s.

Bullshit.The Pyra has a dual core processor and two useless little cpus which probably won't find more uses than Pandora's DSP: Only a couple of years after release, there is a handful (i.e. less than 5) programs using it (I couldn't name any off the top of my head), which that way gain a bit of performance or use a bit less energy (in exchange for a huge amount of work on the programmers side).

 

[Orz]

Oh boy, I really shouldn't have started this...

I way overstated how underpowered the pyra's cpu was going to be. Sorry about that.

I misread the pyra's website as saying that the soc was state of the art (its been available for over a year I think), but now I see that what it was saying is that the cpu's themselves are state of the art (which is still true).

The shield has A9 CPUs, doesn't it?

Huh. I looked at wikipedia and it said that the nvidia shield uses a quad-core cortex a15, not a9. But now looking at the actual cpu, I see that it only has one A15 and four A9's. That should teach me to trust wikipedia, damned paid editors.

Also, I stated that I understand why the pyra probably wont have an x86 cpu. I'm just disappointed, thats all.

That's surprising considering ED hasn't finalized any decisions on the RAM,

More RAM needs more power and is more expensive - so I wonder: what would you need 4GB for?

I thought that ram was finalized at 2 gb, as it is what the pyra's website states. If more ram is too much more expensive and/or uses up too much more power, then I'm fine with 2gb.

If you like closed source stuff then you better get a vita or something. And enjoy it being obsolete when noone develops closed source software for it.

Really? Is there anything wrong with wanting to play closed-source games? And suggesting that I should get a vita instead is just plain insulting.

The CPU is outdated argument was also made when the pandora was being produced

Not very early on. I've looked at topics from way back when the pandora was first announced, and there were a few comments questioning if we really needed a cpu as powerful as the omap.

can i dislike a post? because original post gets all my downvotes/dislikes

Once again, I manage to piss everyone off.

 

[Grench]

OK - I could live with 2GB, but suspect that 4GB may be 'better' for my intended use.  8GB would just be wasting power for 90+% of use cases.  Of course there is swap space - which should be relatively fast on the Pyra's SD bus.

So - can the current MB design accommodate higher density RAM to reach 4GB?

What is the power draw difference between current planned 2GB RAM and higher density (if it exists) RAM to have 4GB?

Within the system's total power envelope and battery capacity is that difference measured in minutes or tens of minutes?

What would the cost per unit difference be to go to 4GB?  

If the difference is under $30 or 20EUR and looses less than 20 minutes runtime, then it's a good trade-off IMHO.

 

[Granitehead]

What is the power draw difference between current planned 2GB RAM and higher density (if it exists) RAM to have 4GB?

I think it might be possible (likely?) that chips with higher individual capacities are just the same RAM twice on the same die. So twice the power for twice the RAM.If they use a smaller process they should even take less power/GB if I understand that right, but would likely cost more (per GB) to buy.

I think it's highly unlikely double the RAM uses more than double the power.

 

[rohezal]

An important thing about power consumption is the nanometer size of the transistors. If you make a 100nm*100nm transistor it has 4 times the area of a 50nm*50nm transistor. And this means not only area but also power consumption. If you are able to get RAM which was produced with latest TMC technology, it could be 4GB which consumes just as much area and power as the 2GB RAM. But this can become very expensive.

 

[Granitehead]

An important thing about power consumption is the nanometer size of the transistors. If you make a 100nm*100nm transistor it has 4 times the area of a 50nm*50nm transistor. And this means not only area but also power consumption. If you are able to get RAM which was produced with latest TMC technology, it could be 4GB which consumes just as much area and power as the 2GB RAM. But this can become very expensive.

That's what I've written here:

If they use a smaller process they should even take less power/GB if I understand that right, but would likely cost more (per GB) to buy.

But nicely explained.

 

[WizardStan]

Huh. I looked at wikipedia and it said that the nvidia shield uses a quad-core cortex a15, not a9. But now looking at the actual cpu, I see that it only has one A15 and four A9's.

Actually you may be right about this. It looks like the Tegra4i is the quad-core A9. The Shield has the regular Tegra4 which by all my searches actually does have quad A15s in it.Sorry about that.

 

[edgex004]

I wonder if the Pyra will be able to run the shield version of portal. That would be fun.

 

[Yoyobuae]

Hmm, do these closed source games you want to play work on x86 tablets?

I'm not even sure if the CPU those tablets use (ie. Core i5) could be put into something the size of Pyra. Probably would need to be a smaller/weaker x86 CPU, at which point the only x86 games that would work well are the ones which are a good number of years old.

Also whenever there's two similar products (ie. tablets) one ARM and one x86, the x86 counterpart is a whole lot more expensive. Probably part of that is the Windows Tax, but the CPU probably plays a part on the cost increase too.

 

[rygD]

I was going to say that my weak research suggests that Tegra 4 has A15s (and something else for power saving when needed), but it seems Orz and WizardStan found the same thing.  http://www.nvidia.com/object/tegra-4-processor.html

My ideal Pyra usage would be somewhat similar to Grench's.  I would like to be able to run VMs, but I have never tried this on an ARM device, and don't know if it is easily possible.  It seems VMware's products won't work.  Same for VirtualBox, I guess.

As for the RAM, I am willing to put in more myself, or pay someone with the proper tools to do it.  It would be nice to have more when it ships, but I understand that there may only be a few of us who will use it, so it might be unneccessary or useless for most people.

 

[Natsu]

3GB would be a compromise.

 

[Grench]

Hmm, do these closed source games you want to play work on x86 tablets?

I'm not even sure if the CPU those tablets use (ie. Core i5) could be put into something the size of Pyra. Probably would need to be a smaller/weaker x86 CPU, at which point the only x86 games that would work well are the ones which are a good number of years old.

Also whenever there's two similar products (ie. tablets) one ARM and one x86, the x86 counterpart is a whole lot more expensive. Probably part of that is the Windows Tax, but the CPU probably plays a part on the cost increase too.

Sorry - mostly wrong.  Though you're right on only playing games a few years old.

Closed source games on Intel's Z3000 (Bay Trail-T) series has been around for a while.

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

 

The tablet SoC from Intel is the Z3000 series - not a Core I5.  The flagship of that line is currently the Z3795.  It is being used in business tablets from Lenovo and HP - now running 4GB RAM and 64bit Windows 8 (yuck).  The new ones with a 64bit boot loader are so new that I haven't seen any citations of someone trying to loading Linux on one.

x86 is -not- more expensive on an SoC to SoC basis.  The flaship Z3795 is $40 per SoC in tray prices.  

http://ark.intel.com/products/80267/Intel-Atom-Processor-Z3795-2M-Cache-up-to-2_39-GHz

They're more 'open' than the ARM alternatives - though that varies.  Intel HD graphics has been in the Linux chain for a long time now.

However:  There is no dev board available for Bay Trail-T.  Apparently these tablet manufacturers are all rolling their own system designs and BIOS?  The development time would have added a year or more to the Pyra's development cycle.

Even though a Pandora/Pyra class machine with a Bay Trail-T processor running Linux would rock - it isn't going to happen for now.

Maybe some day.  From ED's comments that would require a user-community designed SoC board to be developed after the Pyra's release.  I could see that gain some interest IF Linux gurus can get past the UEFI bootloader and make Linux on these tablets work AND the requisite control chips can be shrunk down to fit on the SoC daughter board.

So - 3 cheers for Pyra on OMAP5.  It's going to be a great device with tons of possibilities that will keep us all busy for years.

 

[Yoyobuae]

Sorry - mostly wrong.  Though you're right on only playing games a few years old.

I knew I was probably wrong. I also knew it wouldn't be long someone would want to correct what I got wrong and give all the information I was too lazy to look for myself. So thanks. ^_^

 

[EvilDragon]

x86 is -not- more expensive on an SoC to SoC basis.  The flaship Z3795 is $40 per SoC in tray prices.  

http://ark.intel.com/products/80267/Intel-Atom-Processor-Z3795-2M-Cache-up-to-2_39-GHz

That's more expensive than the OMAP5, which is one of the most expensive SoCs out there.

MediaTek and Allwinner are usually lower than $10...

 

[Kiga]

That's more expensive than the OMAP5, which is one of the most expensive SoCs out there.

MediaTek and Allwinner are usually lower than $10...

Maybe Allwiner could offer Linux drivers if you pay those SoCs at OMAP5 price or higher. Or maybe you could even buy the drivers.