[poll; updated] eMMC vs. uSD; modular eMMC

[comradekingu]

Im sure that sequential read speed is nice, meanwhile the pandora nand is too small for practical use, also 64MB/s under whatever conditions it applies doesnt matter as much as the real numbers. Its an OS, if you are doing heavy reads, its so that you dont have to do any more reading till the next time you boot.

Show me power figures on eMMC 4.5 for comparable performance to the uSD solution, or for something that makes more sense.

Tablets and cellphones without keyboards are much cheaper to make, typically run android, and by companies that do want your device to break.

 

[vcoleiro1]

^ No one is talking about using the Pandora nand, I only mentioned it as reference for it's read speed.   Remember people that want uSD are wanting to run apps and games of it also , not just the OS.

I know its a little surprising that SD uses so much power , but when you think about it, it's designed for cameras, single shot, not for primary low power mobile storage as eMMC is.   If you ever wondered why the battery drains quickly when you are using high speed SD cards while taking video on your camera, now you know.

I'll post the eMMC link shortly.  Actually here's one for reference - that gives the same speeds as we are talking about (up to 80MB/s). Note for reads or writes it draws 200mA it has a low volt operating range of 1.7V to a high of 3.6V ie  in watts that equates to 340mw to 720mw  - see page 8

http://www.psism.com/MR-%20M1560%20eMMC.pdf   

 

[comradekingu]

Its nand flash memory, why is nand emmc so much more efficient? I can understand why its slower, not why its less efficient.

Short of joining JEDEC, http://www.adata-group.com/index.php?action=bs_main&flag=4&lan=us was the closest i could find, i hope thats not 4 W

Edit: Were your figures for v-nand?

 

[vcoleiro1]

^ No., that's not 4W, it is actually less than the one I just quoted, good find. Its 120mW to 250mW

 

[comradekingu]

Its a worthless find, its by A-data for A-data products. Meanwhile, power, speed, and everything else aside, could you be lured into accepting modularity as the most important feature, or is it irrelevant?

If eMMC i like the hardkernel offering, what is your whole system best case scenario?

I think a system with wearable parts that aren't replaceable is naturally faulty. I end up using SD instead to avoid putting wear on something that will fail the whole package.

 

[thatgui]

I think a system with wearable parts that aren't replaceable is naturally faulty. I end up using SD instead to avoid putting wear on something that will fail the whole package.

Which is how likely ? Almost all parts have some kind of lifespan (that is of interest), and I doubt that the eMMC will fall short in terms of that, if you believe otherwise please explain further.

 

[vcoleiro1]

I think I summed up my position in posts 294 and 334.  For the best all round solution IMO, eMMC + 2 SD cards are what gives the best of all worlds.  Future proof storage on the SD cards that will have better storage options than uSD now and in the future .   Fast , reliable, low power eMMC for the OS and whatever.

In the uSD scenario, uSD adds the ability to add a little more storage than what you will be able to put on SD,  if you never upgrade the stock uSD storage it comes with and stick to the SD slots, it's never a benefit to you - that's why I see it as an option that benefits  some while handicapping everyone with lower speed and poor power consumption for storage..

 

I know you seem to be worried about failure comradekingu, it's not a fear I share with you.   As I said before, there are lots of failure points in a device, capacitors , LCD etc etc - its a race to what fails first. That aside, have you ever considered this.  In the very unlikely event that eMMC failed on you, you still have  2 SD card slots to use in a worst case scenario, they don't dissappear from the equation with eMMC. 

 

[_wb_]

I'll post the eMMC link shortly.  Actually here's one for reference - that gives the same speeds as we are talking about (up to 80MB/s). Note for reads or writes it draws 200mA it has a low volt operating range of 1.7V to a high of 3.6V ie  in watts that equates to 340mw to 720mw  - see page 8

http://www.psism.com/MR-%20M1560%20eMMC.pdf   

OK, so according to that link, that eMMC draws typically 200mA when active and 0.25mA when idle.

The Wikipedia article you linked said about SD power consumption:

During transfer it may be in the range of 66–330 mW (20–100 mA at a supply voltage of 3.3 V). Specifications from TwinMos technologies list a maximum of 149 mW (45 mA) during transfer. Toshiba lists 264–330 mW (80–100 mA).[77] Standby current is much lower, less than 0.2 mA for one 2006 microSD card.[78]

So it seems to me that power consumption is about the same, of course depending on the speed. A slow eMMC will in general consume less power during transfer, and so will a slow SD card.

Maybe you got confused because the Wikipedia article mentions 2.88W? That is just the upper bound on the power consumption that the SDXC spec imposes, it does not mean that actual cards actually consume that much, and the spec allows the host device to restrict max power draw way below that level.

In any case, idle power consumption is way more important than power consumption during transfer. Looks to me like SD and eMMC have roughly the same idle power consumption, and if any of the two is worse, then it's eMMC.

 

[vcoleiro1]

There is no confusion and the power consumption is not the same.

2.88W max is what a SD will consume for reads of 64MB/s (not the top of the table) - which is the equivalent speed to the slow Pandora Nand .  Look at the table. Even at the slower range Wizardstan quoted for the Pandora nand of 50MB/s the table shows a draw of 1.44W.

We are comparing the max power draw for the speeds involved for both eMMC and SD , the comparison is a level playing field 

A eMMC will use 250mW max using the example comeradekingu showed for the same speed (actually 80MB/s).  An order of magnitude less than SD for the same speed. And that's not a high speed, that's just matching the Pandora speed (actually bettering it)

The standby power you quoted is for the much slower SD cards he was talking about in that paragraph, not the ones we are talking about. ie for the ones that run at around 12.5MBs .  You cant compare it to the idle power for the eMMC example which runs at much higher speeds, apples vs oranges.

No matter which way you cut it, SD sucks hard for power consumption .  It's not designed to be used for low power primary mobile storage , eMMC is.   uSD is designed for single shot camera's.   There is no escaping the fact.that eMMC  is vastly better for power consumption in low power mobile devices . Not surprising considering that  is what eMMC is designed for .

\

 

[comradekingu]

I know the SD cards dont dissapear from the equation, nor does warrantability, which is the risk. You are pulling "highly unlikely" out of thin air.

Its modularity versus non-modularity, in any event, all other factors being the same it means you gain size or performance to price ratioover time, you lower risk considably, thats the point. Instead of tunnel-visioning one solution, accept that they all are less than satisfactory on what an OS drive spends any time doing, and its finite level storage. Coparing it to other components is drawing failure rate into the mix, not wearability. It would be a failure, not a foreseen event, that a diode, capacitor, or transistor breaks.

I think a system with wearable parts that aren't replaceable is naturally faulty. I end up using SD instead to avoid putting wear on something that will fail the whole package.

Which is how likely ? Almost all parts have some kind of lifespan (that is of interest), and I doubt that the eMMC will fall short in terms of that, if you believe otherwise please explain further.

I have, but nobody listens. Its like the sequential read speed. People happily judge their experience based on a selection of less than the 10 devices they have employing each technology or the understanding they already have.

Edit: i havent shown any emmc examples.

340mw to 720mw is just read, we double that and you get to an order of magnitude better?

 

[vcoleiro1]

340mw to 720mW is read or write in my example link . - you quadrupple that to get 4 times better as I said.

You did show an example eMMC which showed 250mw max for 80MB/s, the one below . It doesn't matter that it's for a-data products, it's an example of eMMC power draw. 250mW max for 80MB/s is an order of magnitude less than 2.88W max for SD at the same speed.

http://www.adata-group.com/index.php?action=bs_main&flag=4&lan=us

 

[wej]

^ 2.88W is what a SD will consume for reads at 64MB/s - which is the equivelant speed to the slow Pandora Nand.

I don't know were you got those numbers, but neither the 2.88 W for the SD card nor the 64 MB/s for the Pandora NAND are correct.

The power consumption of SD cards can differ considerably between different models, so you can't say SD cards generally consume that much power. Some cards might do that, others might not. Also, idle-power consumption of faster cards is not necessarily higher than that of slower ones.

As for the Pandora's NAND read speed, it is far from those 64 MB/s. When I read from the NAND device I get only 10.5 MB/s (reading the entire NAND from beginning to end). So almost any modern SD card will be faster than that.

 

[vcoleiro1]

That speed is what Wizardstan claimed in his Pandora test in this very thread (50MB/s to be exact looking back at it).  It was only an example number in any event for comparisons.

As for the power draws, I have clearly given the references to them.

Is it so surprising that SD cards designed for cameras suck at power consumption compared to eMMC designed for low power primary storage in low power mobile devices.

 

[_wb_]

In the uSD scenario, uSD adds the ability to add a little more storage than what you will be able to put on SD,  if you never upgrade the stock uSD storage it comes with and stick to the SD slots, it's never a benefit to you - that's why I see it as an option that benefits  some while handicapping everyone with lower speed and poor power consumption for storage..

Not true -- even if you never upgrade from what was shipped, the benefit could be that when you buy the device, you can choose which microSD card gets shipped with it (e.g. maybe there are 4GB, 8GB, 16GB, 32GB and 64GB "models" of the Pyra), pick the one that fits your intended use best, and you don't have to pay for capacity you don't need. And if some day you do need more capacity, you can easily upgrade, and the old card isn't even wasted, you can still put it in an adapter card and put it in one of the big SD slots.

I know you seem to be worried about failure comradekingu, it's not a fear I share with you.   As I said before, there are lots of failure points in a device, capacitors , LCD etc etc - its a race to what fails first. That aside, have you ever considered this.  In the very unlikely event that eMMC failed on you, you still have  2 SD card slots to use in a worst case scenario, they don't dissappear from the equation with eMMC. 

I agree that failure is not a big problem. It is more likely that you'll run out of space.

On my main Pandora I'm booting from a 32GB microSD card (in a fullSD adapter card of course) for quite a while now. It is fast enough for me, and that's with a relatively old card and the Pandora's relatively slow SD interface. On both of my Pandoras I don't use the internal NAND at all, because it is too small for me. So for me, the Pandora NAND is a waste of money. I wouldn't want to have the same experience on the Pyra. One-size-fits-all means that we have to make a compromise in terms of size/speed/price, so we will certainly end up with something that is too small for some and unnecessarily big for others, too slow for some and too power consuming for others, too cheap for some and too expensive for others. I would like to avoid that.

There is no confusion and the power consumption is not the same.

2.88W is what a SD will consume for reads at 64MB/s  - which is the equivelant speed to the slow Pandora Nand.  Look at the table. it's 1.44W just at 50MB/s. We are comparing the max power draw for the speeds involved for both eMMC and SD , the comparison is a level playing field 

A eMMC will use 250mW max using the example comeradekingu showed for the same speed (actually 80MB/s).  An order of magnitude less than SD for the same speed. And that's not a high speed, that's just matching the Pandora speed (actually bettering it)

No, you are comparing the maximum current draw that the SD standard allows with the maximum current draw for some specific eMMC. You are comparing apples and oranges here. The max current draw that the eMMC standard allows is also much higher (http://rere.qmqm.pl/~mirq/JESD84-A44.pdf, page 135: up to 550mA at 3.6V, so almost 2W).  It's not that relevant how much the maximum is allowed by the standard; what matters is the actual power draw of actual SD and eMMC cards, which can be anything between zero and the maximum the spec allows (it could even be more than that, they don't necessarily follow the spec).

 

[comradekingu]

You need to learn to not trust a-data as a source for a-data information. "With interface speeds of up to 200 (IOPS)" is probably not more i reckon, so that just below mediocre at best.

Besides https://docs.google.com/viewer?url=https://www.sdcard.org/downloads/pls/simplified_specs/part1_410.pdf page 39 you will find 2.88W is a maximum for the whole ordeal, not write alone.

Page 17:

Host can choose suitable output driver strength by CMD6 Function Group 3.

Host can choose one of UHS-I modes by CMD6 Function Group 1. Each UHS-I mode is specified by

the maximum frequency, sampling edges (rising-only or both) and maximum Power consumption for
compatibility with existing cards. Host can choose one of UHS-I mode depending on capability of

generating SDCLK frequency and capacity of power supply host supported.

Unless you want to deal with CMD19 too, thats where we will leave this, at UHS50.

On page 13 you can see pin 14 being extra supply voltage for uhs-2 cards, as expected.

Yes emmc is faster, no it isnt faster in the long run, and the pyra is in it for the long run. Modularity wins in my book, be it emmc or uSD or both.

Show me the emmc that was so much better speed wise, because im quite content with what uSD has to offer.

http://wiki.cyanogenmod.org/w/EMMC_Bugs does this seem like something you want to happen to anything thats soldered down and cant be easily soldered off?

 

[vcoleiro1]

@_wb_

eMMC gives the best price/speed options, so I find it hard to see how that's a compromise - you cant even match eMMC for speed with uSD because of its OMAP5 100MB/s bus limit.  

As for size, if you need more than the enormous amount the 2 SD can provide now/future, sure, not having a uSD is a size compromise

I'm not comparing any max currents for SD, there are no figures of such that I linked in that table, its just the  power draw for SD at different speeds - presumably its max (doesn't say that though) .   I have given the max power for eMMC also - that is based on the current draw for the specific eMMC product that gives 60MB/s of speed. That's its max power draw . 

The standard you just linked is for 4.4 by the way not 4.5 which we are talking about. Not that its relevant, as we are talkiing about max's at different speeds. We don't know what speed you would get with that doc max rating.

There's one way to settle this, show me the power draw of the 90MB/s uSD card you linked last time.

 

[wej]

As for the power draws, I have clearly given the references to them.

Oh, now I see it, you are refering to that table in the Wikipedia SD card article. Those numbers are maximum power draw, which is interesting for designing the power supply (it has to be able to supply that much current), but which is pretty much irrelevant for the battery runtime of the device (the card does not necessarily consume that much power while transfering data on average, most likely quite the opposite, it probably will do that for very short periods of time only).

Also, people rarely transfer data from SD card at high speed all the time.

 

[Natsu]

Has anyone find a suitable µSSD?

If it becomes µSD only then just give them a capacity option from the same speed cards from the same

company from the start.

SanDisk Extreme Pro Class 10 8GB, 16GB, 32GB.

 

[WizardStan]

I think a system with wearable parts that aren't replaceable is naturally faulty.

You mean like the wifi chip, the nubs, the SoC, the RAM, all the little connectors, pretty much everything on the board, including the board itself?They all have finite use, some much (much much) longer than others with varying difficulties in replacement, but given enough time everything will eventually wear out and need to be replaced. The fact that something can wear out should not be an argument that it must be replaceable, else you'd have to apply the same argument to every other component.

What is the actual life expectancy of eMMC? What is the probability that it will, in all honesty, need to be replaced within a lifetime? Can we assign some arbitrary value after which it's reasonably acceptable for things to go bad and need to replace the entire device? We can then do some statistical analysis to figure out whether eMMC (and any other component) would fall outside this timeframe or not.

Anecdotally speaking, my phone is 4 years old, I install and uninstall all manner of crap to it, copying between the flash and the SD card when internal storage gets full, and every 6 months or so I completely reflash a new ROM to it to see if there's been any improvements before realizing there hasn't and then flashing back to CM7; so you're looking at regular frequent use, complete formats and rewrites 2 to 4 times a year, and no sign of damage.

This is the part where someone else comes along and mentions how their phone's flash died after only a couple months proving that anecdotes are bad science.

 

[comradekingu]

That is of academical interest if any. Battery failure and nand/flash is real, because they have limited cycles which you can spend in less than a lifetime, and they can often be faulty, making even that premise a stretch.

Your phone use isnt close to what you theoretically could put it to. Newer android has trim support though, you shouldnt skip on that. Ive had MLC nand fail on me, my anecdote is more important in that respect, because its when things fail you realize the problem, at which point you kind of want your data to come back, and not have the problem be systemic to the device it runs on.

In the real world these devices fail, controllers, nand, its all a pretty messy industry, and we don't get top picks.

As process node goes up, down goes lifespan of cells. You get more storage per area, but it also means complexity of controllers go up, which is a new cause for concern.

A (modular) uSD isnt horrible today, and its a prospect that gets better. You can test them beforehand and everyone is covered, maybe not 100%, but still. Losing some speed is not the same as some users being downright unhappy with the solution. Similarly, if you tell people about it and be honest, im confident people will be more happy to buy it based on the longevity of that proposition.

Besides, modularity means more people can afford it.

An imposed hard limit on storage means going down the pnd route again, which is making a limited nand space problem systemic to the whole OS.

This could just be solved by having expandable internal memory of a base level ample size. There isn't even a need to extend that via better means.

However much work you want to pile into a little better speeds, soldered eMMC is a 100% total fail right out of the gate if you are unlucky.