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

[vcoleiro1]

^ $79 is for 64GB not 16GB, and includes the extra mini board, pre-installation of software and Hard Kernels handsome markup - Exo pointed that out.  If you look at the prices ED gave for eMMC you will note 16GB is about a 1/5 of the price of 64GB. So scaling down you should get a 16GB eMMC 160MB/s chip for around the $19 price point or less.

Also, as Exo pointed out, that was just A test of the eMMC , not conclusive and it could run faster in other tests or circumstances. So even though that test showed 128MB/s , in more extensive testing it could reach higher to the claimed 160MB/s.  I'm sure not all tests on uSD would result in top speed either.

Secondly here's a comparison between fast 64GB eMMC above and 64GB uSD

160MB/s 64GB eMMC as per above for $79

95MB/s uSD for $122 -> http://www.ebay.com/itm/Genuine-SanDisk-64GB-Extreme-Pro-SD-64G-SDXC-Class-10-C10-UHS-I-633X-Card-95MB-s-/321297993509?pt=Digital_Camera_Memory_Cards&hash=item4aceda4f25

Also, I have seen other similiar speed (160MB/s) eMMC , its cheap.  

eMMC that is much faster than uSD can be bought for cheap .

Update, I have just got an email from Kingston re: my enquiry on the 16GB eMMC ED quoted as costing $19 - ie has to be one of the KE4CN4K6A and KE4CN4A5A options.  They both run at 166MB/s  

 

[_wb_]

^ $79 is for 64GB not 16GB, and includes the extra mini board, pre-installation of software and Hard Kernels handsome markup - Exo pointed that out.  If you look at the prices ED gave for eMMC you will note 16GB is about a 1/5 of the price of 64GB. So scaling down you should get a 16GB eMMC 160MB/s chip for around the $19 price point or less.

Prices don't work like that. You can't just "scale down" like that. You say we "should" get that size/speed/price combination, I'm not so sure. E.g. this one: http://uk.rs-online.com/web/p/flash-memory-chips/7610367P/ is listed for 23.10 GBP with moq=100, so about $38. OK, so let's say ED can somehow magically cut that price in two. Its speed specs are not that easy to find, but I found them here: http://www.wpgholdings.com/procurement/index/zhtw/160_Kingston_7123 and it turns out it reads at 28MB/s and writes at 14MB/s. I'm not quite convinced...

An 8GB 95MB/s uSD costs $11.35 (shipping included, moq=1): http://www.amazon.com/Toshiba-micro-Exceria-Memory-Class10/dp/B00BJW6CL4

Can you show me an 8GB or larger eMMC that beats that price and speed, simultaneously?

Bigger sizes are more expensive (today), but if it's upgradeable, size is not that important. It would be perfectly fine to ship with an 8GB card. Or let the customer choose whatever size/speed/reliability/price

 

[vcoleiro1]

You obviously missed my update.  The 16GB eMMC ED quoted at $19 reads at 166MB/s  - I enquired and got a reply from Kingston .

The choice is 95MB/s uSD or 166MB/s eMMC for the same price point.  Both will get cheaper over time

If you go with uSD, you will be stuck with 70% slower speed forever because of the 100MB/s uSD OMAP5 bus limit.

Not to mention uSD gives you worse latency

Going with eMMC still gives you the option of expanding your storage in future via the 2 SD slots.  SD cards have 4 times the capacity of uSD cards anyway.

 

[Natsu]

Or let the customer choose whatever size/speed/reliability/price

Well, if it's uSD only then initial speed would have to be the same.

Going with eMMC still gives you the option of expanding your storage in future via the 2 SD slots. SD cards give 4 times the capacity of uSD anyway.

But then we will be stuck with the same internal capacity.

 

[_wb_]

You obviously missed my update.  The 16GB eMMC ED quoted at $19 runs at 166MB/s .  

The choice is 95MB/s uSD or 166MB/s eMMC for the same price point.  Both will get cheaper over time

If you go with uSD, you will be stuck with 70% slower speed forever because of the 100MB/s uSD OMAP5 bus limit.

Not to mention uSD gives you worse latency

Going with eMMC still gives you the option of expanding your storage in future via the 2 SD slots.  SD cards give 4 times the capacity of uSD anyway.

Can you stop editing your posts? It's a bit annoying.

I am sceptical about that 166MB/s claim from Kingston. Are you talking about reading or writing? Or maybe just the theoretical maximum of the interface they use? What about random access speed? Are these vendor claims -- claims I can't even find publicly stated anywhere -- somehow backed by independent benchmarks?

The most cutting edge eMMCs from SanDisk (http://www.sandisk.com/products/embedded/inand/inand-extreme/) have a sequential read/write speed of "up to" 150 / 45 MB/s.

Their cheaper, but still quite decent eMMCs (http://www.sandisk.com/products/embedded/inand/inand-ultra/) read at 120 MB/s and write at 30 MB/s. I think the OMAP5432 dev board uses one of those, a 4GB one.

You do know that Kingston is not the most cutting edge nor the most reputable company in the NAND business, right? Their main business is buying Toshiba/Samsung flash memory in bulk and rebranding it as Kingston, aimed at the lower-end section of the market. It would be quite surprising if their eMMCs would beat those of SanDisk, and then they don't even brag about it on their website. That smells fishy, don't you agree?

 

[vcoleiro1]

It's typical read speed, not max.   That's what it says in the doc.

No I don't agree on the fishy thing, I've heard good things re Kingston.  I'm sure ED did the research and has chosen accordingly.

It's not the only example of cheap  eMMC I've seen that's that fast, I've seen a number.  Exo was also trying to say this to everyone from early on.   

 

[_wb_]

It's typical read speed, not max.   That's what it says in the doc.

No I don't agree on the fishy thing, I've heard good things re Kingston.  I'm sure ED did the research and has chosen accordingly.

It's not the only example of cheap  eMMC I've seen that's that fast, I've seen a number.  Exo was also trying to say this to everyone from early on.   

Yeah right, typical, not max. So you're saying it's even faster than 166MB/s? This is getting better and better. What document are you referring to anyway?

ED hasn't chosen yet, those Kingston prices were just an example he put in that thread to clarify that higher capacity comes at a cost and size/price is not linear.

You have seen a number. Big deal. Anyone can claim numbers. I want to see actual benchmark results, not some fishy claims. If just "seeing numbers" is enough for you, then maybe I should sell you a 512TB eMMC that does 31337 GB/s random access reads, for only $199.99. Get it now, before stock runs out!

 

[vcoleiro1]

^  _wb_

Exo gave you an example , and now I've pulled up the actual Kingston eMMC ED was talking about.

I contacted Kingston, pretending to be a company by the way, just to get this real data.  The document is the specific data sheet for that eMMC. Its a 42 page pdf.  It clearly states that the typical read speed is 166MB/s - I would presume that's the max speed . 

As I have said, I have seen other examples, and test of the speed.  It is what it is.  Both myself and Exo have told you this now.   I don't know what further to tell you. 

$19 is for that 16GB eMMC with 166MB/s : see for yourself http://octopart.com/ke4cn4k6a-kingston-24567956

$19 by the way is for 1, not for 1000's as ED would order, it should be a lot cheaper.

I think at this point now , its' probably prudent to let ED decide via actual tests.

 

[_wb_]

I contacted Kingston, pretending to be a company by the way, just to get this real data.  The document is the specific data sheet for that eMMC. Its a 42 page pdf.  It clearly states that the typical read speed is 166MB/s.

Do you mind sending me that PDF (via pm, or attach it to a post)?

What write speed do they mention? What about IOPS? And what does "typical read speed" even mean?

Why and how on earth would Kingston make an eMMC that is significantly faster than any other brand, yet the others (like SanDisk) are bragging about their speed and IOPS while Kingston doesn't even mention any of those numbers anywhere on their website?

Maybe that makes sense to you, but it seems really weird to me.

 

[WizardStan]

Prices don't work like that. You can't just "scale down" like that. You say we "should" get that size/speed/price combination, I'm not so sure

The chip in the ODroid module is THGBM5G9B8JBAIE, a 64GB module. While I can't find any retailers selling it, let's just assume that the $79 is the maximum price.The 16GB version of the chip, same controller, same specs, same everything except the size, is the THGBM5G7A2JBAIM which a casual google search finds for $17-$22 which is pretty comparable to the fastest microSD cards. The 32GB version, THGBM5G8A4JBAIM, can be purchased from retailers for $34-$44 (eg, double that of the 16GB and about half that of the 64GB); it as well uses the same controller and same everything with the only difference being size.

The THGBM5G* series is worth investigating, the price per GB is comparable to microSD cards, and we know it has a minimum sequential read speed of 128MB/s but whether that's a limitation of the chip or the ODroid we don't know.

 

[comradekingu]

vcoleiro stop quoting sequential read speed like its the metric that matters?

To tunnelvision that much you had to conveniently forget how the 23 dollar (retail) samsung uSD plus handidly beat the toshiba card overall. Because, as ive been saying, compromises are made to get sequential read speed high. People buy cards based on it, also, its good for photography.

A paper from kingston on the performance of kingston products isnt trustworthy.

Also, 32, or as the case may be 16GB is slower than that of comparable 64GB emmc. Hardkernels markup is not "significant". It is a custom pcb with a slot in solution and a uSD adapter+preflash atop http://en.chinaflashmarket.com/pricecenter/emmc metal at a current going rate of 40 dollars for 64GB 4.5 emmc. Time to market is a few dollars, and also its the 1x price against something that probably is quoted at a 1000 minimum order quantity, certainly not 1pc.

Also, kingston does not make chips, they buy them from others and put the kingston name on them.

 

[Exophase]

HardKernel's markup isn't their added BOM cost, it's their margin. You don't think they're running a charity, do you?

On the subject of IOPS, I'm going to contend that random writes (which high capacity NAND flash technology sucks at because of the huge block sizes) are much less important than random reads. You don't normally need to update a huge number of files in one quick burst, or update many random bits of a big file. And since writes are usually not blocking (ie, the program making them can fire and forget) a high latency is typical not a big problem. Reading from random places in a file is a lot more normal, for example reading from a large ROM or CD image, and it's much more likely that the program will have to wait for data to be returned.

 

[_wb_]

So eMMC and µSD have roughly the same price for the same speed and size. Maybe eMMC is even slightly cheaper and slightly faster. Still, is it worth adding an item to the BOM (raising the amount of pre-production investment needed), increasing the base price, losing upgradeability, and losing an opportunity to diversify shipped internal storage without complicating pcb population?

 

[Exophase]

How is eMMC adding an item to the BOM vs uSD? In the latter case you're adding the uSD cage.

 

[_wb_]

How is eMMC adding an item to the BOM vs uSD? In the latter case you're adding the uSD cage.

OK, but we need a SIM slot anyway and the extra cost of a SIM/µSD combo is pretty much irrelevant compared to the cost of an eMMC.

 

[Exophase]

OK, but we need a SIM slot anyway and the extra cost of a SIM/µSD combo is pretty much irrelevant compared to the cost of an eMMC.

Assuming that footprint doesn't cause problems. There isn't a clear answer for this yet.

I also think the current SIM slot placement is problematic for this, if the uSD slot isn't accessible to the user without disassembling the Pyra that's a real pain if the user is expected to be able to buy a Pyra with configurable uSD size (or no uSD). But I think the SIM slot needs to be made accessible too - expecting people to take the device apart to install a SIM card doesn't seem acceptable to me. But here's a question: is it possible to make such a combo card accessible if it's behind the battery? Or do you want it accessible via the side of the unit?

 

[Natsu]

I really don't think they would build it without easy access to the SIM/µSD.

 

[vcoleiro1]

The fact of the matter is,  cheap eMMC is significantly faster than uSD , even in random reads -  that hardkernel real test showed that eMMC ran at 128MB/s for reads. 

Also, in the eMMC option , the SD slots don't disappear, they are still there and can be used to use bigger storage in the future.  uSD adds the ability to get 1/8 more storage in the future than you will be able to get with the 2 SD slots (SD seems to always give 4 x more storage size options).  That's not worth the downsides of uSD IMO

Again, lets just let ED do  some testing on solutions available.  Who knows , there may even be another solution, or he may even go with a different SOC.   We are just pissing in the wind at the moment, if he chose a different SOC for example, options could free up to add more things for example.

 

[_wb_]

The fact of the matter is,  eMMC is significantly faster than uSD for cheap (even in random reads - see the hardkernel real test , that eMMC ran at 128MB/s for random reads). Also, in the eMMC option , the SD slots don't disappear, they are still there and can be used to use bigger storage in the future.  uSD adds the ability to upgrade to 1/9 more storage in the future than you will be able to get with the 2 SD slots (SD seems to always give 4 x more storage size options).  That's not worth the downsides of uSD IMO

You keep insisting that internal and external storage are the same thing. They're not. Or are you volunteering to package all of Debian into PNDs I can put on my external storage? If there would be no difference between internal and external storage, then we could ditch the internal storage completely and just use the 2 SD slots for all our storage needs. And put the base OS on a fast USB 3.0 stick.

So please stop saying that a fixed-size eMMC is no problem because there are plenty of external slots and ports. I'm exaggerating and making a bit of a caricature here, but to me that argument sounds like saying that we don't need a replaceable rechargeable battery, a non-replaceable non-rechargeable battery suffices, because if the battery gets empty (and that probably will not happen because you can get very great batteries very cheaply nowadays), you can still use the device by simply plugging it into a wall socket or by simply taping an external battery to your Pyra.

 

[vcoleiro1]

^ Not everyone is going to want 200GB of debian packages that don't even fit on a uSD card today and may never.  As others have said, they have loaded debian with as little as 1GB . Only a small number of people are ever going to want 200GB of those packages, and as I said there is no guarantee uSD will advance enough to give you that anyway.

As I said, Lets just let ED decide , we may get a different SOC, he still hasn't ruled out the Snapdragon 800. That should bring other options if it comes about.