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

[_wb_]

You don't need two physically different media to dual boot either -- you can have more than one partition

 

[flinx0]

Well doing the emmc and sata to sd would satisfy a majority of people from what I have seen in the thread.

Sent from my SCH-I545 using Tapatalk

 

[Grench]

The SATA solution would add too much complication without much yield. Maybe 3+ years from now with UHS-II it would be a viable option.

From what I can tell, the SoC appears to have 4 media controllers.

1. Bootable eMMC OR SD.

2. Bootable SD

3. SDIO

4. non-bootable SD

#3 The SDIO has been taken for Wifi & Bluetooth. This means neither of the SD slots on the Pyra will be SDIO. That leaves three.

We know we want two full sized SD slots out the front - and that one of them needs to be bootable. That would be #2 and # 4 above.

So - we come down to #1, the bootable eMMC/SD. Whether that should be eMMC soldered to the board or a microSD cage piggy backed on the SIM or by itself on the board (similar footprint to eMMC) seems to be the main question.

eMMC can be faster - but ED & team would need to be VERY selective in which eMMC they get. That Kingston part referenced earlier has me concerned. When I think 'fast NAND' I generally don't think of Kingston. It looks like the eMMC on the dev boards is rated at 20MB/s, which is easily surpassed by uSD speeds. It would be pretty pitiful if the end unit could run faster off of the left SDXC card slot than it could off of the eMMC. If they're not careful - it could. eMMC has a faster bus - but for that to mean anything requires finding an eMMC module that can take advantage of it. In theory a 'future model Pyra' COULD be built in a few years with a larger, faster eMMC up to 128GB and 200MB/s (should such a part ever exist). However, similar to the 1Ghz Pandora, it would require a motherboard swap to upgrade to that level. If this option is chosen, it will likely be a 16GB part. If this path is chosen, SanDisk and Toshiba both make parts that push into the 150MB/s territory.

Relatively inexpensive microSD or uSD cards of today can nearly outpace the SD bus on the OMAP. The fast uSD cards are -faster- than normal or slow eMMC. The fast uSD cards price out about the same as 'normal' eMMC. These top out on this controller at a theoretical 104MB/s. Actual inexpensive and available cards top out at 96MB/s - but have not been tested with the controller on the OMAP. There are some concerns about the uSD cage and whether or not it is robust enough to survive the wear and tear that the users could/might apply to it, but in reality there would be no reason to remove this part unless upgrading to a bigger or faster one - as the Pyra would be able to read/write to it (and the other SD ports) at rates comparable to a USB 3.0 card reader. I.e. the Pyra IS a card reader/writer. This would likely be a 16GB 95MB/s uSD card. Toshiba and SanDisk both make cards in this class. Kingston does not.

Starting capacity, space used on the board and price are pretty much a wash between the two.

*eMMC has an edge on speed IF the right part is chosen and works to spec on the OMAP.

*uSD has an edge on potential capacity increases while maintaining very good speed IF the 95MB/s media works to spec on the OMAP.

I could see doing the eMMC option IF a truly fast 16GB eMMC module were obtained. The only real advantage eMMC has is speed beyond 100MB/s. So, to overtake the advantages of uSD, the eMMC would need to be substantially faster and in the >=125MB/s range.

If the eMMC modules available to the project top out at speeds < 125MB/s sustained reads, then I think that the uSD option becomes a clear winner based on being able to swap it out as technology advances. Higher capacity cards that can fully saturate the SD bus should be coming eventually.

So, we need to know more.

*The eMMC on the dev boards is clearly not what is going to be on the Pyra. There really isn't a good way to evaluate eMMC alternatives. Does anyone have a good suggestion here on how to evaluate these modules? It's not like someone can plug/unplug these from the board at whim.

*ED and team need to acquire a 16GB 95MB/s uSD card (<$20) and test it directly on the SD controller on the OMAP with it set to the 104MB/sec bus spec. I suggest trying one of the Toshiba Exceria 16GB 95MB/s cards or the SanDisk equivalent. Both appear to be readily available and inexpensive.

 

[Natsu]

The SATA solution would add too much complication without much yield.

Except being able to have both uSD and eMMC?

I suggest trying one of the Toshiba Exceria 16GB 95MB/s cards or the SanDisk equivalent.

SanDisk would be better.

 

[WizardStan]

This means neither of the SD slots on the Pyra will be SDIO

This is false. They're all SDIO.The one for the wifi is specifically a WLSDIO. I couldn't tell you what makes it so different from regular SDIO though, maybe different power levels or something.

 

[vcoleiro1]

I think it's best left for ED to do some tests.  It could well be that the eMMC he gets is faster than uSD  - remember the OMAP5 bus has a 100MB/s speed limit for uSD.  eMMC 4.5 which he is looking at can go to 200MB/s (current eMMC 5.0 starts at 250MB/s and will reach a limit of 400MB/s eventually).   There is also the point Exophase raised on latency, it seems from what he said, Latency is far worse on uSD than eMMC.   Then there comes power draw, we have been able to obtain some max power figures for eMMC products which are quite low, but haven't for particular uSD products - this is also something ED needs to look at.  Durability of both then needs to be looked at , including expected warranty return costs within the first 1-2 years(whatever it is)  warranty cover period. 

 

[comradekingu]

bt/wifi has not been taken by sdio, you can just as well use realtek rtl8723 which uses a single usb interface to do both

 

[slaeshjag]

I can find neither a datasheet nor a place to buy it from?...

 

[thatgui]

bt/wifi has not been taken by sdio, you can just as well use realtek rtl8723 which uses a single usb interface to do both

Wouldn't we end up with a USB2 controller that never sleeps as it would constantly poll the wifi-chip?

 

[_wb_]

There is no advantage to eMMC for us. The only potential advantage would be speed, but in order to get to that speed (>> 100MB/s, also good random access), cost would go up by too much. Even then, it will probably not keep the speed advantage forever, since even though it would be faster at sequential access (because of the SD speed is capped by 104MB/s), cutting edge SD cards from the future could very well have faster random access and write speeds. So it has one potential temporary advantage that we cannot actually get because we can't afford it.

There are plenty of advantages to µSD.

It took some time to realize it, but this is a no-brainer

 

[Natsu]

What's the speed cap of the other cards?

 

[_jr_]

Except that the available numbers show that reasonably available microSD are by far the more crappy solution in terms of overall performance and to day it seems only wishful thinking that microSD would surpass even mediocre eMMC one day.


Perhaps the numbers are wrong, but there have been no serious attempts to actually produce numbers favourable for microSD. And you can claim that 4k read/write numbers are irrelevant as long as you want, that doesn't make it true for a lot of usecases involving lots of small files.


Upgradeability is nice, but if the polls show one thing it is that upgradeability of internal storage is not a killer argument that tops all other considerations.


edit: concerning life span and anecdotical evidence: I've never lost an SD card, but the 8G eMMC in an Aspire 110L died shortly after two years on me (which is a shame because this was a great little machine; fortunately boots from USB, but my ex-wife took it with her Anyway, I still think the available data favors eMMC.

 

[vcoleiro1]

The most important thing for the OS as Wizardstan has said is the read speed, uSD will always be capped at 100MB/s for read due to the OMAP5 bus.  eMMC 4.5 can go much higher up to 200MB/s. (eMMC 5 to 400MB/s) Exophase already showed an example of an eMMC that was reasonably cheap that is quoted at 160MB/s.  Cheap High speed eMMC that is much faster than the 100MB/s capped uSD is available.   

eMMC  has plenty of pro's: read speed, much ower latency than uSD, low power consumption etc.   ED still needs to check power consumption for uSD which is still unknown.   Lets not forget EDs' statement that he has had 6 out of 9 uSD cards die on him.  I've had the same experience including reader failure.  ED also mentioned how the storage used in the Pandora has been robust and stated his preference to use similiar eMMC because of that - and because of his concern re: uSD .   SD issues have also been mentioned  in several threads on the GCWZERO forum.   It's a recipe for failure IMO - and ED is wise to be concerned about it.

Do we want the OS to be sitting on something that has proven to be robust in the Pandora or something that has proven to fail in ED's and many others experience.

There's a reason eMMC is used for OS storage in practically every  tablet, phone and mobile device and not uSD.  

It's a no brainer, eMMC is best  

 

[Natsu]

I asked that, because I was wondering, Is there no room for a full SD in the inside?

 

[vcoleiro1]

^ ED was concerned re: space required for uSD , so a full SD card is probably going to be to big.

 

[_wb_]

Except that the available numbers show that reasonably available microSD are by far the more crappy solution in terms of overall performance and to day it seems only wishful thinking that microSD would surpass even mediocre eMMC one day.


Perhaps the numbers are wrong, but there have been no serious attempts to actually produce numbers favourable for microSD. And you can claim that 4k read/write numbers are irrelevant as long as you want, that doesn't make it true for a lot of usecases involving lots of small files.

What numbers are you referring to?

ED mentioned Kingston eMMCs.

I don't find speed claims on the Kingston website.

The Kingston eMMC in the Nexus 7 does 23MB/s sequential reading, 8MB/s sequential writing, 5MB/s random reading, 0.32MB/s random writing (according to some anandtech benchmark).

There are microSD cards readily available right now that do 95MB/s sequential reading and 90MB/s sequential writing, 87MB/s random reading, 13MB/s random writing.

 

[vcoleiro1]

^ I was referring to the numbers Exophase gave a while back, you'll have to ask him.   Even TrashyMGs round of the mill eMMC gave much higher figures than you're quoting.  eMMC 4.5 which goes to 200MB/s is old, its being phased out with 5.0 which goes to 400MB/s , the first 5.0 chips were rolling out after mid last year sometime by Samsung (250 MB/s read) .  I've personnaly seen eMMC 4.5 available at 168MB/s read  - by the time ED gets to buying it, it should be bargain basement cheap, let alone cheap as it is now.  Lets not forget the much worse latency uSD has. 

At the end of the day, as I said a few posts ago, ED needs to do the comparison with what he can find - both for uSD and eMMC.  And let him tell us the costs and pros/cons - there's nothing either side of the fence eMMC or uSD can add to the discussion that hasn't been said a million times in this thread. It's up to him now.

 

[_wb_]

^ I was referring to the numbers Exophase gave a while back, you'll have to ask him.   Even TrashyMG round of the mill eMMC gave much higher figures than you're quoting.  eMMC 4.5 which goes to 200MB/s is old, its being phased out with 5.0 which goes to 400MB/s , the first 5.0 chips were rolling out last year sometime by Samsung .  I've personnaly seen eMMC 4.5 available at 168MB/s read  .  

Well OMAP5 cannot do more than 200MB/s, so eMMC 5.0 is as irrelevant as UHS-II SD. And of course faster eMMC than what I quoted exists, but what makes you think we will pay the price for that?

 

[vcoleiro1]

What makes you think its expensive when I a just said it's not and gave the Exo reference as an example (which is 160MB/s read).  As I said, fast  eMMC  4.5 or earlier (160MB/s) is not expensive, its cheap and will be coming down in price further because of the eMMC 5 rollout - (not because we can use eMMC 5)

 

[_wb_]

What makes you think its expensive when I a just said it's not and gave the Exo reference as an example (which is 160MB/s read).  As I said, fast  eMMC  4.5 or earlier (160MB/s) is not expensive, its cheap and will be coming down in price further because of the eMMC 5 rollout - (not because we can use eMMC 5)

Exophases' example (this one: http://www.hardkernel.com/main/products/prdt_info.php?g_code=G135393548479) claims 160MB/s but only does 128MB/s according to the benchmarks, and it costs $79. You may consider that to be cheap and its speed to be significantly better than what can be obtained with µSD, but I don't think that the extra speed makes a significant difference (95MB/s vs 128MB/s is not a speed difference I want to sacrifice other things for) and I do think that $79 is relatively expensive.  EvilDragon was considering a $19 eMMC, and wouldn't use a $41 eMMC because it would make the final Pyra price too high.