Ram & eMMC costs / power drain

[Djoga'Ro]

It would be good to hear things like the actual benefit of using 4GB of ram compared to using the Nand for Swap when 2GB ram use is exceeded,

None, if you're an eternal being, that hardly recognizes time passing.

 

[lukey]

But it looks like the 2Gb Version is only Single-Channel when using 1Gb Chips.

 

[Mr_Loon]

From the pre orders thread :

IMHO if 4GB is easy now with no redesign needed to the system to do it, go for it.

It looks like so far that the 2GB version has so few buyers that it may make sense to single stream populate/make 4GB models only then bin RAM population failures as 2GB versions.

Yes I expect ED to make the purchased prototypes as "top of the line" and have the 4GB. We'll have fun figuring out stuff to make use of the additional RAM.

The additional RAM was not needed for a functional device, but won't be a detrimental factor either.

Mind = Blown

Good to hear you've backed off from the utter nonsense you were spouting earlier in this thread, just imagine if the question regarding how much ram people would like given that the cost would be an additional €X and performance impacts Y & Z had been asked before pre orders.....

As strongly suspected people are willing to pay a premium price for additional Ram thus increasing profits per unit thus helping the project.


@EvilDragon : A couple of follow up questions now pre orders are live :

When do you plan to find out if the 4GB Version works correctly + what impact it has on battery life & performance (compared to swap file)?

Based on the fact that over 97% of pre orders are for the 4GB version is it worth considering making 4GB the default option and scrapping the 2GB option?

 

[b_o_b]

As strongly suspected people are willing to pay a premium price for additional Ram

I care more about battery life than ram. If I can have both fine, but if having more ram means 5-10% less battery life I don't need/want it. You can't test the difference if you don't purchase both.

 

[Grench]

From the pre orders thread :


Mind = Blown

Good to hear you've backed off from the utter nonsense you were spouting earlier in this thread, just imagine if the question regarding how much ram people would like given that the cost would be an additional €X and performance impacts Y & Z had been asked before pre orders.....

As strongly suspected people are willing to pay a premium price for additional Ram thus increasing profits per unit thus helping the project.

Nope. I haven't backed off a thing, and none of it was nonsense. ALL of those arguments for the device not needing more than 2GB RAM and 16GB eMMC still stand.

Yes, your's and other's constant demands for 'testing 4 GB' and 'can I have a special unit with 4GB because 2GB isn't enough for me and I'm oh so special,' were potentially harmful should reviewers latch onto them. "Community is upset because of lack of ram - news at 11!" This is a public forum, and as you showed with your quoting me from another thread above, pulling messages from here out of context is pretty darn easy.

It looks like we're going to get devices with 2GB or 4GB RAM and 32GB eMMC. More than half of each will probably be left unused in a very large proportion of sold units. They were not -needed- for the device to function. Getting 4GB doesn't hurt. Complaining and claiming the -need- for evaluating and offering 4GB on a device that was maxed out at 2GB based on available parts -does- hurt.

If you'll read ED's comments on the preorder thread more carefully, 4GB was not even possible until they started listing the 1GB chips. Previously there simply was not room on the SoC daughter board for more than 2GB (in 4 512MB modules). What you were DEMANDING to be evaluated and that you be given the option to purchase, DID NOT EXIST when you were making your demands.

Now it exists, now it is possible, ED has adjusted. As I stated in your above linked messages, trust ED. He will get us the best options possible for what is within reason for the project.

2GB would have been and is enough RAM to make the Pyra very usable. 4GB recently became possible and now that is an option. OK, that's cool.

8GB eMMC would have been functional. 16GB eMMC would have been nice. 32GB eMMC came inside the price points ED needed, so it looks to be happening. OK, that's cool too.

 

[Swordfish II]

Yep grench hit it.

I am in that camp but still ordered the 4gb. The channel thing concerns me (4gb has the advantage there) as does the power usage. Will I ever max out even 2gb, probably not. But since 4gb fits in EDs plan, and I plan to use the Pyra for 10+ years, I'll get the top of the line.

I was irritated at people trying to use their one-off usage case to change the entire production for everyone and the general steam rolling of ED (you were definately guilty of that @Mr_Loon)

 

[rygD]

ED has shown repeatedly that he listens to and thinks about what those in the community have to say. If no one says anything he can't take what others want into consideration.

Although we never got answers based on real testing, both of the things this thread was asking about look to have worked out in cost. I can think of several other times when this has happened, and sometimes people react aggressively toward those speaking up (keyboard, a little with the gaming buttons, both of which are pretty important), while other times everyone seems fine with all sides of the discussion (5GHz wifi). I can kinda see what Grench was saying, but I highly doubt the widespread damage to the reputation of the Pyra because people had differences of opinion on specs.

 

[Mr_Loon]

If you'll read ED's comments on the preorder thread more carefully, 4GB was not even possible until they started listing the 1GB chips. Previously there simply was not room on the SoC daughter board for more than 2GB (in 4 512MB modules). What you were DEMANDING to be evaluated and that you be given the option to purchase, DID NOT EXIST when you were making your demands.

Over 97.5% of pre orders have been for the 4GB model, my REQUESTS for more information regarding the POSSIBILITY & cost implications of 4GB of Ram seem to bear out my argument that an open discussion of these sort of things could be of benefit to the project.

It still strikes me as a shame that ED did not communicate directly with people to find out what they would be willing to pay for a premium Pyra (4GB Ram & 64GB eMMC) however it's fully appreciated that ED seems to work 25 hours a day as it is and the addition of extra models has a considerable time cost attached to it. As mentioned I think the extra profits made from sales of a premium unit could have been used to subsidise the cost of a base unit thus increasing overall sales.

The good news is that once things settle down the option for a premium unit at a premium price could still be explored.

Nope. I haven't backed off a thing, and none of it was nonsense.

Your claim that 'By the time all factors get added in, the -retail- price winds up being closer to 250% of those part cost differences' was utter nonsense as can be seen from the actual retail price difference between the 2GB & 4GB models, 15€ part cost difference leading to a retail price difference of €30.

Your imagined scenario of news posts from 3rd party sites highlighting community discontent due to requests for information / testing being more damaging than a slavish acceptance of dogma also strikes me as utter nonsense.

However it's fully appreciated that even though you are a self proclaimed idiot, in this thread (and all others) you were acting in what you saw as the benefit of ED and the community so I don't take any offence at the nonsense you posted about the requests for discussion being driven by ego etc.

Please accept that these requests were driven by a genuine belief that additional profits could be made by selling a 4GB version should that possibility exist and that additional profit is good especially when it's combined with giving users what they actually want compared to what they are told they need.

 

[_jr_]

The most important reason to order the 4 GB model is that according to currently available information it is impossible to construct 2 GB/4 GB versions from the same pcb by populating less chips without losing half the bandwidth on the 2 GB version. 2 GB slow memory obviously is significantly worse than 2 GB fast memory. While 4 GB fast memory won't have much of a benefit over 2 GB fast memory, because the OMAP5 is a 32bit system and (assuming a 3G/1G split) most processes will have difficulties using significantly more than about 2.5 GB of memory. E.g. Java VMs will still be limited to less than 2 GB heap size even with 4 GB RAM. 4 GB RAM will allow better multitasking between apps with less paging/swapping, but will probably not help much with any single application (apps with mostly random nonlocal accesses - e.g. because of automatic memory management - will benefit most, I'd guess). Depending on the use case this benefit will mostly range from non-existant to marginal and there will be very few use cases where the increase in usable resident memory size will make the differnece between usable and not usable. Therefore 4 GB could be worse than 2 GB at the same bandwidth for some (AFAICT most) users due to higher cost and power demands (but since these downsides are also pretty small, in the end nobody will fight much against 4GB).

 

[b_o_b]

Therefore 4 GB could be worse than 2 GB at the same bandwidth for some (AFAICT most) users due to higher cost and power demands (but since these downsides are also pretty small, in the end nobody will fight much against 4GB).

Assumption is the mother of all ... I like to see a test with two Pyra's 2GB - 4GB and just compare them before deciding what is the best. I don't think you should just forget about 2GB - I am not so sure that is a good idea. The downsides might be underestimated - you can only know for sure if you test in practice.

 

[lukey]

The most important reason to order the 4 GB model is that according to currently available information it is impossible to construct 2 GB/4 GB versions from the same pcb by populating less chips without losing half the bandwidth on the 2 GB version. 2 GB slow memory obviously is significantly worse than 2 GB fast memory. While 4 GB fast memory won't have much of a benefit over 2 GB fast memory, because the OMAP5 is a 32bit system and (assuming a 3G/1G split) most processes will have difficulties using significantly more than about 2.5 GB of memory. E.g. Java VMs will still be limited to less than 2 GB heap size even with 4 GB RAM. 4 GB RAM will allow better multitasking between apps with less paging/swapping, but will probably not help much with any single application (apps with mostly random nonlocal accesses - e.g. because of automatic memory management - will benefit most, I'd guess). Depending on the use case this benefit will mostly range from non-existant to marginal and there will be very few use cases where the increase in usable resident memory size will make the differnece between usable and not usable. Therefore 4 GB could be worse than 2 GB at the same bandwidth for some (AFAICT most) users due to higher cost and power demands (but since these downsides are also pretty small, in the end nobody will fight much against 4GB).

Well i will use the additional 2Gb as Ramdisk for the Firefox Cache most of the time. Except if its possible to Powerdown half of the ram (which should be possible because the OMAP5 has 2 Ram Controllers).

 

[Klumpen]

The most important reason to order the 4 GB model is that according to currently available information it is impossible to construct 2 GB/4 GB versions from the same pcb by populating less chips without losing half the bandwidth on the 2 GB version. 2 GB slow memory obviously is significantly worse than 2 GB fast memory. While 4 GB fast memory won't have much of a benefit over 2 GB fast memory, because the OMAP5 is a 32bit system and (assuming a 3G/1G split) most processes will have difficulties using significantly more than about 2.5 GB of memory. E.g. Java VMs will still be limited to less than 2 GB heap size even with 4 GB RAM. 4 GB RAM will allow better multitasking between apps with less paging/swapping, but will probably not help much with any single application (apps with mostly random nonlocal accesses - e.g. because of automatic memory management - will benefit most, I'd guess). Depending on the use case this benefit will mostly range from non-existant to marginal and there will be very few use cases where the increase in usable resident memory size will make the differnece between usable and not usable. Therefore 4 GB could be worse than 2 GB at the same bandwidth for some (AFAICT most) users due to higher cost and power demands (but since these downsides are also pretty small, in the end nobody will fight much against 4GB).

In an ideal world, ED would already have a huge company and enough spare change to simply order a few thousand 1GB and a few thousand 512MB RAM units and populate those accordingly in sets of four.
Of course in such a scenario the Pyra would probably be made in batches of 10.000 and cost way less, but unfortunately, not enough people care for a device which is made to last and to offer maximum versatility, yet.
Pandora has already shown many people, that such a thing has an immense value outside of simple "maxing out the specs", maybe more will follow and some day, we will get such a scenario in a few years.
Unlikely, I know. We should just be glad, we get this panacea at all.
BTW: Wouldn't Panacea be a fitting name for the Pyra's successor someday? Or for the docking station?

 

[levi]

Pyrrah is a daughter of Pandora. Unfortuately, Greek mythology being what it is, I can't readily identify a relationship between either of them and Panacea. There probably isn't really one except in the broadest sense, since Pyrrah and Pandora were humans, and Panacea was a god. But I don't see why the docking station couldn't be a god Nice idea!

So am I the only one still confused now the question of what to order has come up? Did we get a number for the power consumption of 2GB versus double the number of chips, I forget? We have a good idea of the end user cost difference now; about 25-35EUR extra for the extra 2GB in a 4GB version. And is it double channel only if we get 4GB? Does that mean it'll be twice as fast at getting out of cache data? I guess that might be useful if we can get the 3D chip working, since that needs to pull in the screen RAM and push it back out after rendering and transforming it - if it can do two of those things twice as fast it might give a decent speed boost.

I'm not sure I can be sure of preordering the right unit without at least some of this information, but perhaps it's already out there?

 

[Grench]

The most important reason to order the 4 GB model is that according to currently available information it is impossible to construct 2 GB/4 GB versions from the same pcb by populating less chips without losing half the bandwidth on the 2 GB version. 2 GB slow memory obviously is significantly worse than 2 GB fast memory. While 4 GB fast memory won't have much of a benefit over 2 GB fast memory, because the OMAP5 is a 32bit system and (assuming a 3G/1G split) most processes will have difficulties using significantly more than about 2.5 GB of memory. E.g. Java VMs will still be limited to less than 2 GB heap size even with 4 GB RAM. 4 GB RAM will allow better multitasking between apps with less paging/swapping, but will probably not help much with any single application (apps with mostly random nonlocal accesses - e.g. because of automatic memory management - will benefit most, I'd guess). Depending on the use case this benefit will mostly range from non-existant to marginal and there will be very few use cases where the increase in usable resident memory size will make the differnece between usable and not usable. Therefore 4 GB could be worse than 2 GB at the same bandwidth for some (AFAICT most) users due to higher cost and power demands (but since these downsides are also pretty small, in the end nobody will fight much against 4GB).

I suspect you're starting from an incorrect assumption.

To my understanding:
Yes, the SoC has two memory controllers.
Each controller can address from 1 to 4 chips.
So 8 chips total * 2 per controller @512MB chips = 8GB theoretical maximum OMAP 5 addressable memory.
The Pyra OMAP SoC daughter board only has room for four memory chips.
The original design was for 2 512MB chips on each of 2 controllers for 2 GB RAM.
The new design, now that 1GB chips are available, is for 2 1GB chips on each of 2 controllers for 4 GB RAM.

None of that precludes having one 1GB chip on each of 2 controllers for 2 GB RAM operating at the same speed.

You don't have to eliminate a controller or decrease the net RAM speed by halving the RAM in this scenario.

Now, for the units that HAVE 4GB on them, it MAY be theoretically possible for users to switch off one of the two controllers resulting in a more power efficient unit with a slower RAM speed. The units with 2GB should be able to do the same trick going to one controller with 1GB of RAM attached.

Where did the idea that a 2GB unit wouldn't use both memory controllers come from?

 

[Yoyobuae]

Where did the idea that a 2GB unit wouldn't use both memory controllers come from?

I asked the same question:
Finalizing everything!

Here's the a photo of the daughter board:
https://pyra-handheld.com/boards/attachments/twopopulated-jpg.27203/

Showing the part number for memory chip: IM4G16D3FABG

Specifically this part:
http://www.intelligentmemory.com/parametric-search/detail.php?we_objectID=96

Which is a 16 bit part. But OMAP5 memory controller has 32 bit bus. Thus two chips are needed to fully populate even a single controller.

 

[Faust]

I care more about battery life than ram. If I can have both fine, but if having more ram means 5-10% less battery life I don't need/want it. You can't test the difference if you don't purchase both.

I would also like to know about the battery drain before I choose my Pre-order, I know that the extra would help for virtual machine for Windows and Android softwares, but I rather not spend 30€ more and 10% battery for small gain.

 

[comradekingu]

4GB ram isn't going to break the bank in terms of power, and luckily neither in monetary measures.

Meaning everyone can get the same unit, which needs 4 chips, and they can all be the same 4.

The cost to source two different types of chips, and re-jig the assembly with it, doesn't defend the numbers.

Not doing 4GB seems out of the question, so that is what it boils down to.

4chip 4GB vs a 4chip 2GB I figure has less of a power difference than a theoretical 2chip 2GB vs a 4chip 4GB.

Easy life for production, development and maintenance.

 

[b_o_b]

4GB ram isn't going to break the bank in terms of power

Care to share some tests to back up this statement?

 

[comradekingu]

Its the same maker, operating at the same voltage, with higher density. If regular rules didn't apply, they wouldn't put it on the market. Nor would it stay on the market.
Which is to say its feasible, and that was my intention of saying it doesn't break the bank.

Its rationale by deduction. Exact figures wont sway estimates enough to give merit to other options, which brings the argument out of the theoretical realm and into possible feasibility.

Short of rigging up a benchmark, one could find a comparable product, but a few would be needed to gauge an estimate, since other variables are more taxing. Like with the speed estimate.

Or you could meticulously read the spec sheets for the chips and see if there are any conditions where it varies.

Lest we forget actual use isn't indifferent between the two. A lot of things allocate memory dynamically depending on the free amount. Also as you run out, its not just an adjustable matter, its in favour of the 4GB.

Which is a adjusted use, perceived speed, actual efficiency over time -estimate. And the human factor comes into play.
More possibility for not a lot of money, on a device that is perceived to be more valuable as a result, and has a comparably high threshold of entry, means people go for 4.

Seeing others go for 4 probably also has an effect. And now that it happened, that is good. And people will want (i assume) to be on the same device, and for it to have 4GB.

 

[levi]

But the decision we're faced with isn't 4-chip 2GB versus 4-chip 4GB - it's 2 chip 2GB versus 4 chip 4GB. How much power will those two extra chips burn through potentially?

And secondarily, if we assume the 2-chip solution doesn't allow dual-channel memory, how much of a difference would that make to different tasks? A lot of tasks will be limited by the speed of the SD card, or the performance of the CPU or the GPU in my experience, so faster ram isn't such a big win. Agree?