What would the world and his dog think if P2 used an x86 SOC? (but not mini SD) ;)


I don't really see what you would need huge amounts of RAM for in a handheld - the only really RAM-hungry tasks I can think of are video editing, large image editing, multitasking a lot, and big compile jobs. Most of those you're not going to do on a handheld anyway, mostly because of the limited screen size.
Interesting you say that, given that you compile on the Pandora every day ;p Well, it's good that someone speaks from the perceived needs of the community rather than just their own needs.

Still, I think given the Pandora's userbase it's not unreasonable to look at the possibilities of it as a docked machine and SSH box. I personally think 2GB would be sufficient, but I wouldn't completely strike 4GB off the table from the start (I would 8GB though :p ), I think a thorough analysis of the options and tradeoffs should be done.
 
If more RAM really means less battery time, I prefer less RAM and more battery time. It's still a mobile device though and I really honor the long hours you can use the Pandora without recharging. No compromises there if ever possible. I see the "10 hrs gaming" as one of the definitions of the "Pandora2". 
 
Depends on the actual scale. If adding an extra 2GB of RAM reduces battery life from 10 hours to 9 hours 55 minutes, I'm ok with that, I can think of more uses for 4GB than I can for 5 minutes. Maybe even as much as 10%, an hour difference, but that's me, other people might be willing to sacrifice 50%, or nothing at all. Point is that some balance can exist, and more RAM shouldn't be automatically discounted because it will cut into battery life until we know exactly how much of an affect it will actually have.
 
If more RAM really means less battery time, I prefer less RAM and more battery time. It's still a mobile device though and I really honor the long hours you can use the Pandora without recharging. No compromises there if ever possible. I see the "10 hrs gaming" as one of the definitions of the "Pandora2".
Worst case scenario would be the ability to deconfigure the RAM on startup, it'll never get powered up that way. There'd probably be much more elegant solutions that still prevent the power consumption when you don't really need the RAM.
 
Never hurts to have more RAM, but I would prefer 2GB of expensive, faster, lower-power RAM over 4GB of cheaper RAM.

Also, if the SD card speeds are significantly faster than they are now (which they will be), then you can have a big swap partition with reasonable performance.

I don't really see what you would need huge amounts of RAM for in a handheld - the only really RAM-hungry tasks I can think of are video editing, large image editing, multitasking a lot, and big compile jobs. Most of those you're not going to do on a handheld anyway, mostly because of the limited screen size.

If we stick to a slim GNU/Linux distro, most of the RAM will be available for applications anyway. In desktop machines running Windows, you need lots of RAM because Windows itself is eating a lot of it itself.
That's where the interesting trade off happens - at least on the Z3770.  There IS an option to run a single channel more expensive faster RAM type on this SoC - but I suspect that it would be better to use two sticks of slower lower power RAM.

Page 19 & 20 of:  http://www.intel.com/content/www/us/en/processors/atom/atom-z36xxx-z37xxx-datasheet-vol-1.html

LPDDR3 is supported in one or two channels.  Max 4GB

  One channel of LPDDR3 yields 8.5GB/sec

  Two channels of LPDDR3 yields 17GB/sec

DDR3L-RS is supported in one channel only.  Max 2GB

  One channel of DDR3L yields 10.6GB/sec

Does anyone know the comparative power draw between these two memory technologies?  How would it translate between these 3 configurations?
 
If more RAM really means less battery time, I prefer less RAM and more battery time. It's still a mobile device though and I really honor the long hours you can use the Pandora without recharging. No compromises there if ever possible. I see the "10 hrs gaming" as one of the definitions of the "Pandora2".
Worst case scenario would be the ability to deconfigure the RAM on startup, it'll never get powered up that way. There'd probably be much more elegant solutions that still prevent the power consumption when you don't really need the RAM.
THAT WOULD BE SO COOL! So you turn it on and you get and EFI type boot screen where-in you can pick the amount of RAM you want to use so when you boot into the OS the non-used RAM doesnt consume power.

I love it
 
Never hurts to have more RAM, but I would prefer 2GB of expensive, faster, lower-power RAM over 4GB of cheaper RAM.

Also, if the SD card speeds are significantly faster than they are now (which they will be), then you can have a big swap partition with reasonable performance.

I don't really see what you would need huge amounts of RAM for in a handheld - the only really RAM-hungry tasks I can think of are video editing, large image editing, multitasking a lot, and big compile jobs. Most of those you're not going to do on a handheld anyway, mostly because of the limited screen size.

If we stick to a slim GNU/Linux distro, most of the RAM will be available for applications anyway. In desktop machines running Windows, you need lots of RAM because Windows itself is eating a lot of it itself.
That's where the interesting trade off happens - at least on the Z3770.  There IS an option to run a single channel more expensive faster RAM type on this SoC - but I suspect that it would be better to use two sticks of slower lower power RAM.

Page 19 & 20 of:  http://www.intel.com/content/www/us/en/processors/atom/atom-z36xxx-z37xxx-datasheet-vol-1.html

LPDDR3 is supported in one or two channels.  Max 4GB

  One channel of LPDDR3 yields 8.5GB/sec

  Two channels of LPDDR3 yields 17GB/sec

DDR3L-RS is supported in one channel only.  Max 2GB

  One channel of DDR3L yields 10.6GB/sec

Does anyone know the comparative power draw between these two memory technologies?  How would it translate between these 3 configurations?
LPDDR3 vs DDR3L

http://blog.virtium.com/ddr3l-vs-lpddr3/

http://www.brightsideofnews.com/news/2012/6/27/samsung-lpddr3-high-performance-memory-enables-amazing-mobile-devices-in-20132c-2014.aspx

http://www.powersystemsdesign.com/performance-vs-power-in-off-chip-ddr-sdram

It sounds like would want the LPDDR3 version - lower power usage, designed for mobile?

That brings us to 'one stick or two'?

Two gets high performance - obviously higher power draw too.  But how much additional power draw compared to the SoC, compared to the screen, etc?
 
What if we had 2 sticks, but a setting in the BiOS that allowed you to enable/disable one of the sticks?
No offense but you have some really outrageous ideas.

Board area is a scarce resource, look at how heated the arguments over using microSD instead of SD got.. the only valid reason anyone ever entertained microSD is because of board area. This is also a big hit to cost and design complexity, just putting a big switch on all these high speed memory I/Os could compromise the design.

All for what, so you get a little bit of extra control in power vs performance? Probably little software would even see a big benefit from slightly more memory bandwidth.
 
I don't really see what you would need huge amounts of RAM for in a handheld - the only really RAM-hungry tasks I can think of are video editing, large image editing, multitasking a lot, and big compile jobs. Most of those you're not going to do on a handheld anyway, mostly because of the limited screen size.
Interesting you say that, given that you compile on the Pandora every day ;p Well, it's good that someone speaks from the perceived needs of the community rather than just their own needs.


Still, I think given the Pandora's userbase it's not unreasonable to look at the possibilities of it as a docked machine and SSH box. I personally think 2GB would be sufficient, but I wouldn't completely strike 4GB off the table from the start (I would 8GB though :p ), I think a thorough analysis of the options and tradeoffs should be done.
For most compiles I've done on the Pandora, 512MB was enough. There were a few occasions when I needed swap space, but even then it stayed below 2GB. But of course if you want to compile in parallel using 4 cores, you'll need more RAM.

Most users most of the time are not going to be compiling stuff though :)
 
This is very true. Aside from emulators, no one thinks of optimization when they are writing code for x86, so none of the software that people want to run would be optimized.
I don't think this is always true. Look at x264. There are some things where ARM may have gotten more optimization in recent years out of necessity.. but let's be serious here, even in this community is writing ARM assembly still? Who is writing heavily tuned NEON code? It's just a few people.. and I don't know if there are a lot of people still doing SSE either but at least some will remember having done it years ago.

The compilers are also probably a little better at auto-vectorization for x86. Especially if you use ICC (not encouraging it but the option is at least there)
 I'll admit, I was not taking video codecs into consideration when I made that comment. However, this does not detract from my point. Most of the software people are interested in would not be optimized, because a lot of the developers are used to having a lot of CPU and RAM to throw around.

I agree, but given the difficulty in sourcing a current Gen ARM SoC it could be time to think the unthinkable. Either way, I'd love to see the P2 designed so a simple change of 'motherboard' would allow an ARM or x86 variant to be manufactured. Not 2 variants at launch, but maybe a year or two down the line.
The P2 should not vary in functionality. It splits the userbase.

-God Ginrai
 
The P2 should not vary in functionality. It splits the userbase.
It does split the community, but not necessarily in a bad way. Obviously lots of people want ARM because they've already invested so much time in it. But obviously others are interested in the first ever non-shit x86 UMPC (such as myself)

Is there anything bad about some people writing x86-only software and some people writing ARM-only software? I personally dont think so.
 
4GB would be nicer if a true side-by-side solution for Linux + Android emerged. I'm trying to think of other big scenarios.. I guess on my desktop having a bazillion Firefox tabs open eats all my memory, and occasionally editing some gigantic text files for debugging stuff.. but these really aren't normal Pandora use-cases..

I guess it's hard to justify anything more than a small cost increase. Or if it hurts other options like RAM technology availability, board space, etc. I think we do agree that 2GB is essential, at least.

Just think, several months ago Craig was saying how it'd have 8GB of RAM and some were behind that :p
 
Someone help me decipher this?

http://www.samsung.com/global/business/semiconductor/news-events/press-releases/detail?newsId=12979

"Samsung's 3GB LPDDR3 DRAM connects with a mobile application processor using two symmetrical data transfer channels, each connected to a 1.5GB storage part. Though asymmetric data flow can cause sharp performance dips at certain settings, the symmetrical structure avoids such issues, while maximizing system level performance."

Does that mean that this 3GB LPDDR3 piece is actually two 1.5GB LPDDR3 in one - meaning that it connects to both sides of a dual channel controller?  I.e. one feature, two channels, full bandwidth in a single chip?  Or am I dreaming that part into the description?

Probably an expensive little piece of reconstituted sand - but I'm curious if I'm interpreting it right.
 
Does that mean that this 3GB LPDDR3 piece is actually two 1.5GB LPDDR3 in one - meaning that it connects to both sides of a dual channel controller?  I.e. one feature, two channels, full bandwidth in a single chip?  Or am I dreaming that part into the description?

Probably an expensive little piece of reconstituted sand - but I'm curious if I'm interpreting it right.
Yes that's what it means. Most PoPs used in high end devices now have two channels internally because high end SoCs out now are dual channel. It's not like you're supposed to stack two PoP modules on top of the SoC :p This isn't something new or exotic. Internally the PoP can actually consist of several DRAM dies in some stacked configuration. Think about how much space on a DIMM you need to get even 2GB, now think about the tiny surface area the PoP module takes up. Even Pandora's PoP module consists of multiple DRAM stacks as well as NAND. This particular one looks like it has 4 DRAM chips, although the stack is only two high laid side by side.

Sorry, I was thinking I should have told you this earlier... the stuff you were thinking about using two of these modules connected to a BayTrail-T made no sense (not least of all because you can't have two PoPs in a system)
 
Last edited by a moderator:
Probably too expensive, but it's the P2 spam area so I feel a license to dream...

http://www.samsung.com/global/business/semiconductor/news-events/press-releases/detail?cateSearchParam=&searchTextParam=&startYyyyParam=&startMmParam=&endYyyyParam=&endMmParam=&newsId=12984&page=&searchType=&rdoPeriod=

64GB eMMC on-board storage capable of 250MB/sec reads and 90MB/sec writes...

I'm not sure I even want to know what this piece would cost, power usage - or if it can be obtained, etc...  But - wow.
 
The P2 should not vary in functionality. It splits the userbase.
 It does split the community, but not necessarily in a bad way. Obviously lots of people want ARM because they've already invested so much time in it. But obviously others are interested in the first ever non-shit x86 UMPC (such as myself)

Is there anything bad about some people writing x86-only software and some people writing ARM-only software? I personally dont think so.
Yes, there is plenty wrong with it. There will be very few users who will buy both, meaning there will be a tiny amount of users for each device. Developers will be encouraged to "choose a side", and either both sides will miss out on killer applications, or one side ends up with a waste of money because no one is developing for their device. You can have variations for small features, (although you normally shouldn't) because the software will remain the same and thus the community remains unified. (i.e. PS3 60GB vs. all other PS3s) However, a difference in architecture is detrimental to the survival of a product, especially a niche product like this.

EDIT:

I guess on my desktop having a bazillion Firefox tabs open eats all my memory
Actually, Firefox uses most of its memory up front. 2.5k tabs only takes up ~2GB of memory when combined w/ the Bartab extension. W/o Bartab, I've got FF on another computer using 1.2GB of memory w/ 473 tabs.

-God Ginrai
 
Last edited by a moderator:
Does that mean that this 3GB LPDDR3 piece is actually two 1.5GB LPDDR3 in one - meaning that it connects to both sides of a dual channel controller?  I.e. one feature, two channels, full bandwidth in a single chip?  Or am I dreaming that part into the description?


Probably an expensive little piece of reconstituted sand - but I'm curious if I'm interpreting it right.
Yes that's what it means. Most PoPs used in high end devices now have two channels internally because high end SoCs out now are dual channel. It's not like you're supposed to stack two PoP modules on top of the SoC :p This isn't something new or exotic. Internally the PoP can actually consist of several DRAM dies in some stacked configuration. Think about how much space on a DIMM you need to get even 2GB, now think about the tiny surface area the PoP module takes up. Even Pandora's PoP module consists of multiple DRAM stacks as well as NAND. This particular one looks like it has 4 DRAM chips, although the stack is only two high laid side by side.


Sorry, I was thinking I should have told you this earlier... the stuff you were thinking about using two of these modules connected to a BayTrail-T made no sense (not least of all because you can't have two PoPs in a system)
Thank you - I'm an admitted novice trying to understand this stuff.  If the part were somehow available, would it be a good 'middle ground' between the 2GB and 4GB options?  Since it's on a ~20nm process, could it be inherently more efficient than the 2GB parts?  Likely very expensive -again.
 
Last edited by a moderator:
Actually, Firefox uses most of its memory up front. 2.5k tabs only takes up ~2GB of memory when combined w/ the Bartab extension. W/o Bartab, I've got FF on another computer using 1.2GB of memory w/ 473 tabs.
You're talking as if this isn't extremely dependent on what's in the tabs >_< Firefox is definitely a big memory sucker for me. Not that the numbers you gave aren't big.
 
Last edited by a moderator:
Back
Top