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

[WizardStan]

Have you not seen the OS topics? All the discussion of the PND format for this future device has leaned towards using it only for games and to use the repo for everything else.

I've seen them. I'm choosing to stay out of them as much as possible.

That's not what I meant. I meant it would be cheaper for EvilDragon, and thus easier to produce.

Is the development cost of uSD significantly lower than eMMC? I'd expect it to be higher, if even just a few pennies for the connector. The actual costs of the chip itself shouldn't make a difference, whether uSD or eMMC that cost will be directly translated to the end user during production, nothing to worry about during development.

Although it would be -possible- to put a 'fast' eMMC module in the device at twice the price of a 'normal' eMMC module that would be somewhat faster than the microSD cards of today, two years from now it will likely be slower than the 'bleeding edge' microSD cards on the market - and seem pathetically small

The limiting factor isn't the card, it's the device. Just as the Pandora has a maximum theoretical transfer speed of 24MB/s (and practical transfer speed of about 16MB/s) the SoC chosen for the Pyra is going to have a maximum theoretical speed as well. The bus speeds for eMMC and SD will be identical, so the difference will be bus width, wherein eMMC is double that of SD (at least). Choose a good eMMC chip that is rated near the maximum bus speed and it will necessarily be twice as fast as any uSD card you might put in, regardless of how bleeding edge that SD card might be.

 

[Askarus]

Using eMMC locks the device into whatever speed and capacity is chosen today. Using microSD gives us the flexibility to use whatever speed and capacity exists now and tomorrow. So - to me, the real question is: How fast can the eMMC/microSD connection on the SoC go?

That's a good question.

I very much like the fact that microSD is upgradeable after some years.

Will the maximum speed the processor can deal with be higher than todays best eMMC?

 

[God Ginrai]

That's not what I meant. I meant it would be cheaper for EvilDragon, and thus easier to produce.

Is the development cost of uSD significantly lower than eMMC? I'd expect it to be higher, if even just a few pennies for the connector. The actual costs of the chip itself shouldn't make a difference, whether uSD or eMMC that cost will be directly translated to the end user during production, nothing to worry about during development.

I explained that right after the part you quoted. The average user can get a cheaper cost due to the cost of production lowering, which is likely to happen, because EvilDragon doesn't have to feel constrained to picking a larger storage size with microSD since those users who do want more could replace it themselves.

Although it would be -possible- to put a 'fast' eMMC module in the device at twice the price of a 'normal' eMMC module that would be somewhat faster than the microSD cards of today, two years from now it will likely be slower than the 'bleeding edge' microSD cards on the market - and seem pathetically small

The limiting factor isn't the card, it's the device. Just as the Pandora has a maximum theoretical transfer speed of 24MB/s (and practical transfer speed of about 16MB/s) the SoC chosen for the Pyra is going to have a maximum theoretical speed as well. The bus speeds for eMMC and SD will be identical, so the difference will be bus width, wherein eMMC is double that of SD (at least). Choose a good eMMC chip that is rated near the maximum bus speed and it will necessarily be twice as fast as any uSD card you might put in, regardless of how bleeding edge that SD card might be.

Grench's argument still stands. If we choose "near" the bus speed w/ eMMC, then we will always be at just that. If we choose microSD, then when microSD speeds surpass the bus speed, we are guaranteed to be able to transfer at bus speed, thus still allowing microSD to surpass eMMC in speed.

-God Ginrai

 

[Exophase]

So - to me, the real question is: How fast can the eMMC/microSD connection on the SoC go?

For any SoC specifications I'm currently aware of, around 200MB/s for eMMC and 100MB/s for SD would be the bus limits.

Keep in mind, performance has to do with more than long peak burst read speed or write speed. Random access performance matters too, seek time is important. That's a big reason why people value SSDs over HDDs.

 

[God Ginrai]

So - to me, the real question is: How fast can the eMMC/microSD connection on the SoC go?

For any SoC specifications I'm currently aware of, around 200MB/s for eMMC and 100MB/s for SD would be the bus limits.

Keep in mind, performance has to do with more than long peak burst read speed or write speed. Random performance matters too, ie seek time is important. That's a big reason why people value SSDs over HDDs.

Ok, so it turns out that eMMC has a higher bus limit. However, if what Grench said about fast eMMCs being expensive is true, then the cost would go up a decent amount just to ensure a fast enough eMMC that couldn't be surpassed in speed by microSD.

-God Ginrai

 

[comradekingu]

Keep in mind the figures given for SD cards arent sustained writes, its burst speed. Also, they typically suck at random IOPS. Which makes them not so hot for OS use.

On omap 5 what are the available bus-es? Maybe that could help to solve the puzzle

 

[Klumpen]

Grench on the left

 

[WizardStan]

Grench's argument still stands. If we choose "near" the bus speed w/ eMMC, then we will always be at just that. If we choose microSD, then when microSD speeds surpass the bus speed, we are guaranteed to be able to transfer at bus speed, thus still allowing microSD to surpass eMMC in speed

I said "near" but what I really meant was "slightly higher than bus speed but as close to bus speed as possible so that we're not paying for ultra-premium when really we just want really-good"."near" was much shorter.

His argument is that faster chips might appear in the future, but that's completely moot if eMMC chips at the current bus speed can be acquired.

UHS-I Class 3 cards are already being produced which exceed the bus speed of the OMAP5 which'll run you about $50 for 16GB. Unfortunately I'm having trouble finding prices for equivalent eMMC chips. We need these prices in order to make a comparison. Get those numbers and the argument becomes perfect, we can prove that uSD cards are fast enough at a reasonable price point, or using historical records we can estimate when they will be fast enough at the right price. Without actual numbers to compare it's just conjecture: at some point in the future the speed/cost ratio may be sufficiently better that it's worth waiting, but we don't know that for sure.

I explained that right after the part you quoted. The average user can get a cheaper cost due to the cost of production lowering, which is likely to happen, because EvilDragon doesn't have to feel constrained to picking a larger storage size with microSD since those users who do want more could replace it themselves.

And I still questioned whether the development cost for uSD would really be cheaper than eMMC. I suspect that it isn't. Sure, maybe end product is cheaper because it uses a cheaper part, but only for those that opt for the inferior part; plus it has nothing to do with development and the difference in production cost is paid by the end user anyway so it has nothing to do with costs incurred by EvilDragon.
edit: I think I mucked up the terminology of the UHS speed types. Regardless, microSD cards at 90MB/s read speeds are costing about $50 for 16GB.

 

[Natsu]

Assuming roughly equal costs? ID really C which.

 

[WizardStan]

Toshiba produces a line of eMMC chips that can clock up to 200Mhz (slightly higher than what the OMAP5 can support, check out THGBM5G7A2JBAIM), quick price lists from chinese resellers list the 16GB chip at about $20, down to $16 when bought in the thousands.

So for the same size, eMMC gets double the speed for less than half the cost of the best uSD cards.

You'd have to get a significantly slower or smaller uSD card to beat a $16 eMMC that is as fast as it can possibly get. Personally, even if you could manage to get a 16GB uSD card with 90MB/s transfer speed for free, I'd still pay the $16 for the double speed eMMC.

 

[Exophase]

Here's a benchmark of another Toshiba eMMC available as an add-on module for Odroid-U2:

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

128MB/s read speed real world, 64GB module costs $80: http://www.hardkernel.com/main/products/prdt_info.php?g_code=G135393548479

That price surely includes some overhead for HardKernel's PCB/assembly, card reader, and margins. And HardKernel's purchase orders would probably be even lower volume than ED's. I expect that part can be had for $60 at the most. Highest speed 64GB uSD cards cost around $100.

Personally I don't trust listing on alibaba or other Chinese sites, so I wouldn't go with quotes there being representative unless there was hard evidence that they could ship and honor the price. I've seen too much stuff there that was blatantly impossible.

 

[Jonney]

Would it be possible to use a m.2 SSD for internal storage, or would something like that be too expensive?

 

[Grench]

Using Exophase's Toshiba example as a benchmark...

Toshiba has that feature in several sizes.

http://www.toshiba-components.com/memory/emmc.html

The size point ED has thrown around is 16GB, which would be a Toshiba THGBM5G7A2JBAIM.

It prices out at $21. https://parts.arrow.com/item/detail/toshiba/thgbm5g7a2jbaim?gclid=CL2h_p3Bx7wCFeJj7Aoda14Agw#RRgQ

The nearest microSD alternative I could find is:

http://www.amazon.com/Toshiba-micro-Exceria-Memory-Class10/dp/B00AIH3CUA

It is a 95MB/s (rated) Toshiba microSd card for $17.85.

It is a bit slower, but can be upgraded at the user discretion up to 32GB at that speed at ~$40 retail.

http://www.amazon.com/Toshiba-Micro-EXCERIA-Memory-Extreme/dp/B008YRZ6UK

http://www.toshiba-memory.com/cms/en/products/microsd-cards/exceria/exceria-type-hd.html

So... getting as close to apples to apples as possible:

Option A: 16GB of 128MB/s eMMC is about $21 wholesale. It is faster, but once it's on the board, it can never be upgraded.

Option B: 16GB of 95MB/s microSD is about $18 retail + socket cost of roughly $2 or less. It can be upgraded currently to 32GB at 95MB/s for $20 more ($40) and there is an 8GB version for ~ $10. Real world sustained speed isn't known.

To me, on this type of device, the ability to easily replace or expand the capacity of the primary OS 'drive' is more important than the speed bump we're talking about.

It will still boot up in a fraction of the time that the Pandora does.

Like the Pandora, in general use there isn't really any reason to power it down since it will sleep/standby for days or weeks.

So, I would prefer to have an internal microSD slot. At that point I wouldn't be dissapointed if ED used the $10 8GB Toshiba Exceria microSD card as the 'shipping' card, knowing that I could swap it for whatever suited my purpose or needs.

 

[comradekingu]

If the question is speed, im seeing removable eMMC as more of a solution, as the bridge in between is a considerably lower cost of the total. Keeping in mind that a 64GB eMMC solution is much much faster than a 16GB one.

Even unpopulated (as the case may be) you end up spending just the tiniest amount on everyone, and get in return the ability to upgrade once eMMC is cheaper, its extendable, servicable, and a super duper warranty on nand failing, making it much more of a peace of mind to even use the nand.

i have a hardkernel u3 board, and the eMMC solution is tiny in comparison to a usb port.

Sacrifice stereo sound if it comes down to it, i think it makes all the sense in the world.  And then we dont have to have the woes of another non systemic package system.

 

[Ligachamp]

If I remember right, gta04 said, the OMAP5 can only handle one eMMC and one SD slot, a second SD slot has to be added with a separate usb-sided controller. So why should we "loose" one SD port, if we add uSD instead of eMMC? There would be only one SD port anyways if they don't add the controller.

The conversion between eMMC and SD isn't so complicated, so we could use uSD via the eMMC port.

About the possible speed using this kind of system I really don't know much about. But I don't think the max possible speed would decrase a lot (on bus side).

EDIT: In my opinion we should take the uSD slot. User modification would be as easy as possible, and who really minds ~30MB/s more or less if we're already talking about ~100MB/s standart speed? Anyways it would be fast

 

[WizardStan]

The eMMC port IS an SD port. There's no loss going from one to the other, it'd be the same port internally regardless.

And the OMAP5 has 4 SD ports: one eMMC, 2 SD, and one WLSDIO port.

 

[Ligachamp]

The eMMC port IS an SD port. There's no loss going from one to the other, it'd be the same port internally regardless.


And the OMAP5 has 4 SD ports: one eMMC, 2 SD, and one WLSDIO port.

So what was gta04 talking about? I'm sure he was talking about troubles handling 2 SD and the NAND directly with the OMAP...

 

[WizardStan]

Hrm... a 16GB uSD card... my quick search turned up cards for $50, but for only $18, that's the same range as the eMMC.

It's still half(ish) the speed, but without the huge cost difference that may be an acceptable compromise to be able to upgrade it later. Gonna need to think a bit.

So what was gta04 talking about?

No idea, I don't remember seeing what you're talking about. Got a link?

 

[Neelix]

The eMMC port IS an SD port. There's no loss going from one to the other, it'd be the same port internally regardless.


And the OMAP5 has 4 SD ports: one eMMC, 2 SD, and one WLSDIO port.

So what was gta04 talking about? I'm sure he was talking about troubles handling 2 SD and the NAND directly with the OMAP...

I think I remember the post you are referring to, but IIRC  I think he was actually referring to one of the other SOC's being proposed and comparing it to to OMAP.

- Neelix

 

[God Ginrai]

So what was gta04 talking about?

No idea, I don't remember seeing what you're talking about. Got a link?

Same here. I don't remember seeing gta04 mention anything about this subject.

-God Ginrai