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

[vcoleiro1]

I think Exophase already pointed out that fast eMMC is relatively cheap.

Why should I and everyone give up speed for nothing that I would use in return.  That is the compromise here.   For those that want the flexibility of card slots , they would still exist in an eMMC + 2 card slot config.  Plus not to mention you could add a flash drive etc via the USB if you wanted even more storage options. Seriously, with so many storage options already available - do you really want to sacrifice speed for everyone , for yet another storage option.

 

[comradekingu]

You don't explain why you want to take away proposed modularity, only that there is modularity already planned. A non-volatile chip failing because its dead doesnt get alleviated by other options taking over so much as the problem being modular-option solves it.

Just because you personally wont use it, or don't think so, doesn't mean a percentage of people will. Cutting the percentage of people that will refund their whole item because the soldered storage died makes sense even if small compared to the rest. It is also, potentially life-saving for dragonbox. Also it seems people like the upgrade prospect, possibly dont want to pay more than barebone model upfront, and as me, like the peace of mind that wearing out storage is a modular problem.

If the dual sim/sd card connector/bay is used then it isnt even an extra part, the uSD vs the emmc, fits into a 1 pc connector already present.

Show me the eMMC options that are performance wise better than 1k iops across the board.  Its not just uSD vs emmc, you can also spend money on ram, which sweetens the deal for everyone, gets rid of that hard limit, smokes all options in regards to performance, has extra abilities, and alleviates problems with storage-memory.

 

[_wb_]

The way I see it, most current mobile devices that aren't dead cheap have eMMC chips instead of soldered down uSD cards. For these manufacturers there has to be a compelling cost vs quality reason to make this decision. Maybe it changes with economy of scale but I doubt it, if you get low quantities of SD cards you'll be paying a distributor their cut same as with eMMC; while SD cards may be more marketed for consumer purchase from the get go they're also commonly sold in brick and mortar stores that have a supposed markup.

Maybe I'm too cynical, but I always assumed that having non-upgradeable internal storage is a nice feature if you want to make sure that the device feels obsolete by the time you want your customers to get the new model. Of course if the choice is between eMMC and a soldered down uSD card, then eMMC wins because it is a bit cheaper at any given (size,speed) point.

Can someone tell me the price of an eMMC of 32, 64 or 128GB (less than that will look cheap and silly in a couple of years) that can reliably reach more than 100MB/s on all operations (sequential and random reads and writes)? Because that is the minimum that is needed in order for eMMC to be a better option than microSD. If it does not reach that performance level, it is almost sure that at some point in the next few years, there will be (micro)SD cards available that are both faster and larger.

 

[TrashyMG]

Faster SD cards are irrelevant if we're limited by the bus speed.

 

[Exophase]

The point of the external SD slots is that you are free to eject the cards in them at any time. The point of internal storage is that it contains all the software and data that you want to have available all the time. External storage is not a real replacement for internal storage, unless you sacrifice one of the external slots to have a non-ejectable card. Having two free-to-eject external SD slots is a really nice feature, it allows you to copy things from one SD card to another SD card directly. Non-upgradeable internal storage effectively forces people to give up one ejectable external SD slot at some point in time.

Only if you actually need more internal storage in the first place. Most will agree that the amount of stuff wanted for this purpose is limited. And for it to get in the way of copying from one SD card to another you'd be limited to content that cannot be removed while the copying is done, not simply content that you'd like to always have present.

Also, a minor adjustment to what you're saying - if you need something available all the time, you could put it on all of your removable cards, instead of restricting one to always being in the slot.

IOPS of the microSD and eMMC itself should be roughly similar, since the underlying technology is the same. On decent SD cards, for random small reads ~1000 IOPS is typical, for random small writes ~100 IOPS is typical (writing is always slower because of the underlying technology, i.e. block-based NAND flash). For comparison, a spinning HDD typically does ~100 IOPS (reading or writing makes no difference), while a good SSD can usually do 10,000-100,000 IOPS.

IOPS is not an issue, and the difference between eMMC and µSD is not significant in that respect.

But IOPS are not merely determined by the NAND technology, they have a lot to do with how sophisticated the controller is and the associated caches. I also thought comradekingu's remarks of SD cards with single digit IOPS sounded too bad to be true but he's right, there are benchmarks of several cards a couple years old that got only a few KB/s random read performance under 1KB/s random write performance with 4KB blocks (basically equivalent to IOPS measurement). Many fell under this category, while the best ones were orders of magnitude better. I don't know what it's like for eMMC, they're not exactly the same market segment as SSDs, but they could still have a higher baseline for random access performance.

I agree that the differences between thousands and tens of thousands (much less hundreds of thousands or millions) are mostly irrelevant for our purposes, though.

Maybe I'm too cynical, but I always assumed that having non-upgradeable internal storage is a nice feature if you want to make sure that the device feels obsolete by the time you want your customers to get the new model.

But what does this sort of planned obsolescence have to do with eMMC? As I said, they could just use soldered down uSD instead. They almost certainly would instead of having removable uSD, and although it would technically be easier for a user to upgrade that almost no one would bother.

I will go through the trouble of upgrading the internal micro SD as soon as I can upgrade for a reasonable price.

Don't see any reason why I wouldn't.

Do you always upgrade things just because you can? The only reason you need is "why not"?

 

[comradekingu]

WB: Your cynicism backfires though, its not a hardware support operation. Turn the argument and say that more people will get it because its not a planned obsolescence device. Means more customers at the same base platform, easier to support just one, and easier to support with modularity, which wins in the long run, which is what its about.

If what you describe existed at a price point that was feasible, i wouldn't be spending hundreds of dollars on SSD drives.

More memory still smokes the perfect emmc at perfect price. We should be looking at ram prices.

TrashyMG: We arent limited by the bus speed, because random writes will never saturate 100MB/s random 4k 32QD, it _may_ saturate the 100MB/s top end on sequential reads, which is what we can get in the real world, and thats how to load the OS, at which point ram takes over and that has a higher transfer threshold too, whats not to love.

Exophase: You are explaining workarounds, but not why emmc is the better option. Copying takes time, and redundancy for availablity uses more space, union type systems are added technicality for dubious reward, namely why _isnt_ uSD good when it has the better options checked.

Why is "most" a good argument when the prospect of having modular internal storage isnt elitist? The opposite to what most get along with, instead factors in everyone, including the supplier.

uSD cards continue to grow at a rate peoples / and /home does. Everyone can agree that this is an expanding size, also need over time.

Why do we even assume people have/want to buy a fullsize SD card anyhow, why not just the one uSD. Even if your slots go empty, when you use uSD for everything you are at any given time free to transfer from any SD to any other SD, thats a good way of having modularity for the sake of it.  Home on a separate card is faster overall, but again i feel the need to point to ram as the speed option that for the people not interested in speed can be defended in other ways.

While /home can be ejected when its not on the internal, it for the most part ties up the first SD card slot.

TL;DR uSD and more ram is for everyone, and they share their strengths.

 

[zonova]

It depends what you plan on using the eMMC/uSB for. In my mind, the only internal storage we really need is for the OS. That way, it's clean and fast. Perhaps it could have some extra storage space, but I'm not thinking of either of these options as a main form of "internal storage" or something similar that you might see on a desktop or laptop. I think we will have plenty of slots for replaceable storage, I think we should have something fast, cheap, and secure for the OS. If the eMMC does have problems, how difficult exactly is it to be removed/replaced?

 

[Exophase]

Exophase: Why is "most" a good argument when the prospect of having modular internal storage isnt elitist? It factors in everyone, including the supplier. uSD cards continue to grow at a rate peoples / and /home does

why do we assume people want to buy a fullsize SD card anyhow, why not just the one uSD

While /home can be ejected when its not on the internal, it for the most part ties up the first SD card slot.

comradekingu, I'm going to be brutally honest with you. I don't have the faintest idea what you're trying to say most of the time and this post is no exception :/ It's vexing because your spelling and grammar all seem fine. This could be a total failing of my reading capability, in which case maybe someone could translate it for me.

 

[comradekingu]

I fixed my above post, in a post-edit because it was unclear. I often do that, so if it doesnt make sense, give it a while.  However I'll try to do better.

Edit1: Option A has a problem, while option A is not option B, its shortcomings could be lessened by options C or D

C and D, when used this way, have their own set of shortcomings,

While all of this is fine for most people, most of the time

It doesnt say why A is better than B, especially since problems relating to A is how not only A is problematic, but also how B is superior.

We then scrap everything about saying A is almost comparable in a way that B is just better, and ask instead:

What does A have which is better than B, knowing that B not only fixes the problem arising from the direct comparison, but also has distinct and exclusive features to B

(emmc is A,  B is uSD)

Edit2: What I am saying is that your defence of B is like using option C and D, it only takes away something, most like "most" does, pun intended

if you dont have to, dont. It falls to the nature of the argument.

edit3 - Plain english:

There are two options, one of which has a problem.

The problem can be worked around, but thats just moving the problem elsewhere.

The problems of the first option can be lessoned by two additional solutions, but they are detrimental.

The unproblematic solution is unproblematic.

How the workaround applies to the unproblematic in much way is how the problem is inherrent to its solution, its like trying to explain a benefit away with what is always not as beneficial.

Just like most sounds good, you could say it is always a good ad populace argument, but "everyone" is always better.

Edit4: The problem with the answer to the problem of the problem, goes away.  Confuzingu strikes again.

 

[_wb_]

IOPS of the microSD and eMMC itself should be roughly similar, since the underlying technology is the same. On decent SD cards, for random small reads ~1000 IOPS is typical, for random small writes ~100 IOPS is typical (writing is always slower because of the underlying technology, i.e. block-based NAND flash). For comparison, a spinning HDD typically does ~100 IOPS (reading or writing makes no difference), while a good SSD can usually do 10,000-100,000 IOPS.


IOPS is not an issue, and the difference between eMMC and µSD is not significant in that respect.

But IOPS are not merely determined by the NAND technology, they have a lot to do with how sophisticated the controller is and the associated caches. I also thought comradekingu's remarks of SD cards with single digit IOPS sounded too bad to be true but he's right, there are benchmarks of several cards a couple years old that got only a few KB/s random read performance under 1KB/s random write performance with 4KB blocks (basically equivalent to IOPS measurement). Many fell under this category, while the best ones were orders of magnitude better. I don't know what it's like for eMMC, they're not exactly the same market segment as SSDs, but they could still have a higher baseline for random access performance.


I agree that the differences between thousands and tens of thousands (much less hundreds of thousands or millions) are mostly irrelevant for our purposes, though.

Of course there exist cheap, old, very slow µSD cards. Just don't use those We don't have to make internal storage non-replaceable just to avoid that someone shoots himself in the foot by replacing it with something that sucks.

The relevant comparison is: what size/speed/price can you get with eMMC today, and what can you get with µSD today. And what can we expect in a couple of years (comparing todays eMMC with future µSD, taking bus limitations into account).

I don't see any reason why eMMC would inherently have better controllers etc. I don't think there's any technical argument for that. There may be somewhat different target markets, but that's mostly something of the past: for cameras (the original target for SD cards) you don't need good random access, but µSD is used in so many phones and tablets that there is at least a large segment that will create demand for fast general-purpose cards.

Anyway, like I already said, random small write speed is not that important because we'll have plenty of RAM and a decent kernel that will do a lot of caching and buffering anyway. Since it's internal storage, we don't have to do immediate writebacks like on the external slots (to deal with sudden ejects by users who don't unmount before swapping a card).

Is there a way for us to get UHS-II speeds instead of just UHS-I ? Also for the external SD slots that would help to make things even more future-proof.

 

[Exophase]

Of course there exist cheap, old, very slow µSD cards. Just don't use those We don't have to make internal storage non-replaceable just to avoid that someone shoots himself in the foot by replacing it with something that sucks.

The relevant comparison is: what size/speed/price can you get with eMMC today, and what can you get with µSD today. And what can we expect in a couple of years (comparing todays eMMC with future µSD, taking bus limitations into account).

I don't see any reason why eMMC would inherently have better controllers etc. I don't think there's any technical argument for that. There may be somewhat different target markets, but that's mostly something of the past: for cameras (the original target for SD cards) you don't need good random access, but µSD is used in so many phones and tablets that there is at least a large segment that will create demand for fast general-purpose cards.

Anyway, like I already said, random small write speed is not that important because we'll have plenty of RAM and a decent kernel that will do a lot of caching and buffering anyway. Since it's internal storage, we don't have to do immediate writebacks like on the external slots (to deal with sudden ejects by users who don't unmount before swapping a card).

Is there a way for us to get UHS-II speeds instead of just UHS-I ? Also for the external SD slots that would help to make things even more future-proof.

Here's read IOPS for SD cards today:

http://www.tomshardware.com/charts/sd-cards-2014/-7-Random-Read-4-KB-QD-32-MB-s,3478.html

The baseline is much higher than it has been and there's much less difference between the top and the bottom, but it still tops out at around 780 and bottoms out around 450.

And write IOPS:

http://www.tomshardware.com/charts/sd-cards-2014/-8-Random-Write-4-KB-QD-32-MB-s,3479.html

There's still a tier that has very, very bad performance.

Here's some numbers an Odroid eMMC module:

http://forum.odroid.com/viewtopic.php?f=54&t=1152

Which shows better performance than the best SD cards. Similarly with Odroid-XU:

https://www.linux.com/learn/tutorials/752860-benchmarking-the-odroid-xu-a-fast-clocked-quad-a15-arm-machine

(look for random read/random write scores)

Some older numbers are much worse, but still have better read IOPS and poor write IOPS but not nearly like the very bad class we still see today with some SD:

http://blog.datalight.com/emmc-problems

 

[_wb_]

Yes, todays eMMC is faster than todays µSD cards. Both will keep getting better though. Is there any eMMC available today that can do better than the max speed the bus allows for µSD, i.e. faster than 100MB/s? If so, then that's an argument in favor of eMMC. If not, then we'll be stuck with something that will inevitably at some point become both slower and smaller than µSD.

 

[Askarus]

Yes, todays eMMC is faster than todays µSD cards. Both will keep getting better though. Is there any eMMC available today that can do better than the max speed the bus allows for µSD, i.e. faster than 100MB/s? If so, then that's an argument in favor of eMMC. If not, then we'll be stuck with something that will inevitably at some point become both slower and smaller than µSD.

That's exactly the point

 

[slaeshjag]

Yes, todays eMMC is faster than todays µSD cards. Both will keep getting better though. Is there any eMMC available today that can do better than the max speed the bus allows for µSD, i.e. faster than 100MB/s? If so, then that's an argument in favor of eMMC. If not, then we'll be stuck with something that will inevitably at some point become both slower and smaller than µSD.

Isn't that exactly what was linked a few pages back? A real-world existing eMMC with mesured read speeds of 128 MB/s?

 

[vcoleiro1]

eMMC 4.5 which was announced June 2011, has top speeds of 200MB/s. There are obviously lots of chips to choose from with speeds from low 100'sMB/s to 200MB/s 

eMMC 5.0 which was announced last year and will eventually supercede 4.5, has a top speed of 400MB/s.  The first eMMC 5 module I saw a while back had a speed of 250MB/s, there may be even faster ones now.

Obviously the latest eMMC 5 is not really want we want as 250MB/s is greater than the OMAP 5 bus can handle.  We should be able to get an older eMMC 4.5 class module for close to 200MB/s relatively cheaply IMO. ED will need to investigate the size,speed,price options.

 

[_wb_]

Yes, todays eMMC is faster than todays µSD cards. Both will keep getting better though. Is there any eMMC available today that can do better than the max speed the bus allows for µSD, i.e. faster than 100MB/s? If so, then that's an argument in favor of eMMC. If not, then we'll be stuck with something that will inevitably at some point become both slower and smaller than µSD.

Isn't that exactly what was linked a few pages back? A real-world existing eMMC with mesured read speeds of 128 MB/s?

Sure, for sequential reads. For sequential reads you can find µSD that reaches 95MB/s, so that's not really a significant difference. What about writes? What about random access?

 

[Exophase]

Yes, todays eMMC is faster than todays µSD cards. Both will keep getting better though. Is there any eMMC available today that can do better than the max speed the bus allows for µSD, i.e. faster than 100MB/s? If so, then that's an argument in favor of eMMC. If not, then we'll be stuck with something that will inevitably at some point become both slower and smaller than µSD.

The point here is that today's eMMCs are exhibiting lower access times than today's SD cards. Meaning eMMC isn't just SD with double the bus width, just like SSDs aren't just SD with an SATA interface.

Maybe latency will get better with the best SD cards to the point where it eclipses the best eMMC cards today, but you don't know if that'll actually happen any time in the reasonably near future. The argument that there's a big push for high performance SD because it's available on lots of mobile devices doesn't really hold up when those mobile devices are running their OS and apps off of eMMC, and some of the most popular ones still lack SD expansion entirely.

 

[_wb_]

Yes, todays eMMC is faster than todays µSD cards. Both will keep getting better though. Is there any eMMC available today that can do better than the max speed the bus allows for µSD, i.e. faster than 100MB/s? If so, then that's an argument in favor of eMMC. If not, then we'll be stuck with something that will inevitably at some point become both slower and smaller than µSD.

The point here is that today's eMMCs are exhibiting lower access times than today's SD cards. Meaning eMMC isn't just SD with double the bus width, just like SSDs aren't just SD with an SATA interface.


Maybe latency will get better with the best SD cards to the point where it eclipses the best eMMC cards today, but you don't know if that'll actually happen any time in the reasonably near future. The argument that there's a big push for high performance SD because it's available on lots of mobile devices doesn't really hold up when those mobile devices are running their OS and apps off of eMMC, and some of the most popular ones still lack SD expansion entirely.

OK. But in my opinion, todays SD access times and speeds are already good enough. Of course more speed is always nice, but not if it comes at the price of not being upgradeable and having to settle for a one-size-fits-all compromise that will probably not be that fast and big because of price considerations.

 

[vcoleiro1]

IMO


With eMMC -> Better latency figures and twice as fast as uSD + 3 replaceable and upgradeble storage options -> 2 x SD and 1 USB flash drive option


is better than


With uSD which gives worse Latency , twice as slow speed (always because of the bus limitation) + 4 replaceable storage options -> 2 SD , 1 uSD and 1 USB flash option.


I wouldn't want to sacrifice twice the speed and better latency for the OS , just so you can have 4 removeable storage options as opposed to 3. You can always use one of the 3 options to upgrade the removable storage. You do not lose that ability in the eMMC option.

 

[Askarus]

IMO


With eMMC -> Better latency figures and twice as fast as uSD + 3 replaceable and upgradeble storage options -> 2 x SD and 1 USB flash drive option


is better than


With uSD which gives worse Latency , twice as slow speed (always because of the bus limitation) + 4 replaceable storage options -> 2 SD , 1 uSD and 1 USB flash option.


I wouldn't want to sacrifice twice the speed and better latency for the OS , just so you can have 4 removeable storage options as opposed to 3. You can always use one of the 3 options to upgrade the removable storage. You do not lose that ability in the eMMC option.

I agree with _wb_.

SD-cards will get faster.

It's possible to upgrade.

As long as they are faster than Pandora NAND it's good anyways.