Alternative distribution to optical disks : SSD, cards, and download*

[Blazkowicz]

dunno. I've found this year-old shoot out between 500GB to 750GB 7200 rpm drives, that gives an idea of performance, those are probably dual platter drives.
http://www.tomshardware.com/reviews/notebook-hdd-750gb,2832.html

access times are 14 to 17ms, a figure not much different than in the late 90s.

7200 vs 5400 may be rendered moot if you only look at sequential transfer rates, this gives a theoretical +33% but a drive will twice the density is faster by sqrt(2). so a slower rpm, more dense disk is equal to a faster rpm, less dense one, for simple sequential copying.

here's a little video similar to the HDD vs bluray one, but with a laptop
http://www.youtube.com/watch?v=o9f8fKn40kk
unscientific, with different capacities.. but we can see the perf divide widen a lot in the more stressful scenario.

 

[eastmen]

http://semiaccurate.com/2012/02/22/sandisk-develops-the-worlds-smallest-128gb-flash-chip/

it uses 19nm and comes in at 170mm2 .

Prices are going to continue to drop.

I also apologize for not being able to respond to everyones posts tonight. Way to busy of a day, prob tommorow after work

 

[Gubbi]

Contract price for TLC NAND is $0.50 / GB. I'd be very surprised (and disappointed) if we don't see at least 64GB of solid state storage in next gen consoles.

Cheers

 

[-tkf-]

Contract price for TLC NAND is $0.50 / GB. I'd be very surprised (and disappointed) if we don't see at least 64GB of solid state storage in next gen consoles.

Cheers

This article from Anand sheds some light on why it isn´t so easy.

On the other hand, we're concerned that the cut in prices is done at the expense of endurance. One advantage often heard about buying an SSD is that SSDs are a lot more reliable than hard drives. In terms of P/E cycles, that is probably true with current MLC NAND. However, there have been quite a few widespread firmware issues, such as SF-2281 BSOD and Intel 320 Series 8MB bugs. Those have been fixed, and we may finally be looking at SSDs which have good performance, adequate endurance, and are more or less trouble-free. However, TLC will require new controller logic, and new logic may result in additional firmware issues.

Putting an unproven technology in millions of consoles is, as this gen showed us, pretty optimistic.

http://www.anandtech.com/show/5067/understanding-tlc-nand

 

[pcchen]

Putting an unproven technology in millions of consoles is, as this gen showed us, pretty optimistic.

http://www.anandtech.com/show/5067/understanding-tlc-nand

TLC is not good with endurance, but if all you need is some sort of semi-permanent storage, it's probably good enough. For example, some MP3 player and mobile phones use TLC as they don't really need to write into their storage that frequently.

If a console uses TLC then I suspect that it may have to be a two tier system. That is, a large portion uses TLC but that's mainly for game installations (i.e. read-only data). A small portion is used for game save data and other temporary storage needs.

 

[-tkf-]

TLC is not good with endurance, but if all you need is some sort of semi-permanent storage, it's probably good enough. For example, some MP3 player and mobile phones use TLC as they don't really need to write into their storage that frequently.

If a console uses TLC then I suspect that it may have to be a two tier system. That is, a large portion uses TLC but that's mainly for game installations (i.e. read-only data). A small portion is used for game save data and other temporary storage needs.

I am sure something could be hammered out with TLC, but while price is lower it´s without any doube the least proven technology out there and even with the the known Nand solutions there is plenty of challenges..

So add to the extra cost the extra risk of screwing it up.

 

[Acert93]

Taking that form of risk aversion last gen would not have seen:

* Xenos. Unproven shader architecture.
* eDRAM. Sporting logic on a new process node with dense eDRAM and a discreet bus requiring things like tiling.
* Cell. Large. Hot. Heterogeneous. Totally new microarchitecture.
* XDR. Hardly a volume product.
* Blu Ray. The definition of pushing a product as fast as possible out of R&D to market.
* Motion Sensors. While older technology was it well established how these devices would survive mass production into a commodity device intended for hours-per-day abuse by children?

If a product is in mass production and can be tested designers can use it in a closed box. It is more of a question of trade-offs. Per storage there is nothing to prevent it to be on a removable drive or for it to be replaced by such in case of failure. If a device has a 5 year projected use and you get 1k writes; assuming a 64GB drive that is 50% filled by volume a smart controller would make it last how long? The projections I have seen for MLC (3k-5k) is decades iirc.

A bigger issue of TLC I would think is reduction and you likely will be looking to transitioning to MLC or another technology so there may not be much of a cost drop for a couple years on the storage, assuming we don't see some kind of tiered storage (which makes more sense IMO).

 

[-tkf-]

Taking that form of risk aversion last gen would not have seen:

* Xenos. Unproven shader architecture.
* eDRAM. Sporting logic on a new process node with dense eDRAM and a discreet bus requiring things like tiling.
* Cell. Large. Hot. Heterogeneous. Totally new microarchitecture.
* XDR. Hardly a volume product.
* Blu Ray. The definition of pushing a product as fast as possible out of R&D to market.
* Motion Sensors. While older technology was it well established how these devices would survive mass production into a commodity device intended for hours-per-day abuse by children?

If a product is in mass production and can be tested designers can use it in a closed box. It is more of a question of trade-offs. Per storage there is nothing to prevent it to be on a removable drive or for it to be replaced by such in case of failure. If a device has a 5 year projected use and you get 1k writes; assuming a 64GB drive that is 50% filled by volume a smart controller would make it last how long? The projections I have seen for MLC (3k-5k) is decades iirc.

A bigger issue of TLC I would think is reduction and you likely will be looking to transitioning to MLC or another technology so there may not be much of a cost drop for a couple years on the storage, assuming we don't see some kind of tiered storage (which makes more sense IMO).

A bunch of excellent points, but afaik those that designed those "risks" had great knowledge and would imho have a better chance of knowing how it would playout, HiDef war excluded. Or in other words, i don´t put them as risky as putting TLC in consoles from the get go, is there even a SSD based on that now? (i know, i should google it)

 

[pcchen]

A bunch of excellent points, but afaik those that designed those "risks" had great knowledge and would imho have a better chance of knowing how it would playout, HiDef war excluded. Or in other words, i don´t put them as risky as putting TLC in consoles from the get go, is there even a SSD based on that now? (i know, i should google it)

TLC is actually not that new, so I don't know why you think it's more risky than other new technologies. The first TLC flash chip was sampled in 2010 IIRC. OCZ wanted to make SSD with TLC flash, but apparently haven't been able to do so. For a generic SSD it's really very difficult to put something with a few hundred erase cycles in actual use while maintaining a reasonable lifespan.

However, if your data usage pattern is relatively static, a few hundred erase cycles should be more than enough.

 

[Acert93]

The Anand article says, "Just to make this clear, TLC isn't anything new. For example Hynix had a 32Gb 48nm TLC die in 2008. This is because TLC was originally used for devices like USB flash drives, where its poor endurance would be negligible."

 

[-tkf-]

TLC is actually not that new, so I don't know why you think it's more risky than other new technologies. The first TLC flash chip was sampled in 2010 IIRC. OCZ wanted to make SSD with TLC flash, but apparently haven't been able to do so. For a generic SSD it's really very difficult to put something with a few hundred erase cycles in actual use while maintaining a reasonable lifespan.

However, if your data usage pattern is relatively static, a few hundred erase cycles should be more than enough.

Isn´t that a problem? You are asking the Console manufactors to bet on their users not using the storage "that much". If we are lucky and console manufactors decide to include a browser, just that usage over a few years might be enough. As a user gets more games the need for re-installs grow as the size would have to be small on the flash to keep the prices down.

Yes it´s all wild guessing since we don´t know how much endurance would be needed and how much can be delivered. But consider that just getting SSD´s reliable enough and giving them enough of a life span has been a challenge even for intel. I would consider it a expensive challenge if Sony/Microsoft were to do it themselves. They need something off the shelf which apparently isn't there. .

 

[-tkf-]

The Anand article says, "Just to make this clear, TLC isn't anything new. For example Hynix had a 32Gb 48nm TLC die in 2008. This is because TLC was originally used for devices like USB flash drives, where its poor endurance would be negligible."

And they still use them for that? Still makes it new in the context we are discussing.. right?

 

[Acert93]

It doesn't look like TLC Flash is being used in a major way right now in SSD. But I think you are creating an artificial barrier between the risks I mentioned when you noted, "those that designed those "risks" had great knowledge and would imho have a better chance of knowing how it would playout" because (a) TLC is old tech and (b) they know how a SSD based on such would play out--more so than, say, putting eDRAM and logic on a die with functional tiling and a discreet bus to a dedicated GPU (aka Xenos) or how all the Blu Ray (especially the software layer) were going to shake out looking from the perspective of a 2006 launch from sitting in 2005.

Yes, the drive still needs to be made and the concern is obviously is a $0.20-$0.40/GB savings over MLC worth the investment and tradeoffs? It would seem so much simpler to go with 16GB or the like of SSD with a better lifecycle/speed and then a large (500-750GB for the time frames we are looking at) single platter 7200rpm HDD behind it. But I wouldn't be too concerned about an TLC solution; these are closed boxes after all and a company like MS or Sony should be able to nail down their own drivers and software for such. I mean Sony was able to not only fab Cell but to get all the libraries and SDK out for a completely new microarchitecture. If Sony/MS had a storage design chosen now (which should be one of the easier parts of these consoles) they essentially have 2 years of time for a end 2013 launch to ensure it is working properly.

Edit: all that to say that if they decided that saving hundreds of millions by going with TLC fit their requirements I am sure they could make it work. It wasn't much different from MS telling AMD/IBM their chips in the 360 weren't getting discreet memory pools but a shared pool and they had to make it work with the memory controller on the GPU. Or that they wanted eDRAM but were going to have to imbed the ROP logic on the die. If it saves millions and or meets your performance criteria you make it work. TLC may not meet those standards, so I dunno. But I wouldn't count it out because it isn't in the PC space--the reasons it isn't in the PC space make a lot of sense from a marketing perspective as it would be perceived as unreliable (already an SSD concern, if not an unnecessary one for the casual user). In a closed box console you are not looking at the same concerns from a marketing perspective, just a functional one. Taking a tech that is cheaper/inferior in some ways but meets your design goals is all that matters. Same criteria why SPEs aren't a good desktop solution but were arguably a good console one. But I personally don't know, but worth keeping in mind as moving data off a slow Blu Ray into a HDD w/ SSD memory could solve a big issue.

 

[Blazkowicz]

they would need some way to update the firmware. would be funny if that becomes a convoluted attack vector.

no, really TLC is made as a crap-grade flash and is useless for the caching of streaming from optical we were all talking about.
your idea of small SSD + HDD may be done, it could be both devices combined i.e. what we call a hybrid hard disk.

possible SKU differenciation here, you could have a bare 500GB hard disk or a 1TB + flash one.

 

[eastmen]

they would need some way to update the firmware. would be funny if that becomes a convoluted attack vector.

no, really TLC is made as a crap-grade flash and is useless for the caching of streaming from optical we were all talking about.
your idea of small SSD + HDD may be done, it could be both devices combined i.e. what we call a hybrid hard disk.

possible SKU differenciation here, you could have a bare 500GB hard disk or a 1TB + flash one.

which is why they should ditch optical and just go strait to flash carts .

Yes its more expensive per game than optical , but in terms of what is needed in the actual system its alot cheaper and savings are found in other parts of the supply chain also.

You should be able to do 16 gigs of TLC flash for less than $5 per 16 gigs in 2013

 

[Shifty Geezer]

The latest rumour is the next xbox will forgo an optical drive.

Seems kinda unclear to me though. Could potentially be an XB360+, with cards instead of disc at an entry level price with Kinect/Kinect 2. Or a download only version of XB360? Or, if the lower performance rumours are to be believed, could be a proper new box but the choice of cards is limiting storage which in turn is maybe making MS think they don't need as much power in there as there won't be be the assets to use all the latest power.

 

[AlphaWolf]

I really don't see how it would make any sense for the 360. Releasing games on carts and disc at the same time isn't going to be cost effective or net any feature bonus to consumers. And IGN's lower performance rumor is stupid even from a low performance perspective. (Not the cheapest, or the lowest power consumption, just the slowest option...)

 

[Rangers]

The latest rumour is the next xbox will forgo an optical drive.

Seems kinda unclear to me though. Could potentially be an XB360+, with cards instead of disc at an entry level price with Kinect/Kinect 2. Or a download only version of XB360? Or, if the lower performance rumours are to be believed, could be a proper new box but the choice of cards is limiting storage which in turn is maybe making MS think they don't need as much power in there as there won't be be the assets to use all the latest power.

I dont see how it necessarily means lower performance. You'd assume they'd shoot for 16GB carts or something as the standard to start out with which should be enough and is an ~2x increase on 360 (as well as being a not too shabby ~1/3 50GB max Blu Ray). Current top PC games are in the 10-12Gb range from what I've seen.

And saving room, heat, wattage, and cost on an optical drive could in fact lead to a more powerful system everywhere else. Also presumably fast flash seek times could lead to better performance.

All wild, wild speculation at this point though.

 

[Rangers]

I worked up some "interesting" numbers on GAF (even though I imagine this type of analysis has been beat to death already), Sony has shipped 589m PS3 games so far, assume 1b by the time the system is EOL for a system of only middling popularity. Even at $2 per flash cart vs nominal disc cost, that's 2 billion dollars in costs over the gen that need to be absorbed between publishers, consumers, and platform holder in some way. That's obviously pretty huge. Also I have no idea if $2 is realistic for carts or if reality is much higher.

But wait, a commodity DVD drive (best case) costs..?? Over, 100 million pieces of hardware for PS3 and 360 probably. Newegg has OEM DVD drives for ~20. An isupply 360 teardown I looked up seemed to peg the DVD drive BOM at $21, which seems a little high, but was presumably done in 2005.

If you peg the DVD drive cost at $10, you can save a billion dollars over 100m pieces of hardware, half of our supposed flash cart loss. If it's closer to 20, then you're at 2 billion.

That assumes the price wont be higher if you assume Blu Ray instead of DVD, which even next gen I expect will have some type of premium in cost over a DVD drive.

Plus presumably you can save on reduced warranty repairs as you eliminate failing drives, you can possibly save in other areas too, less heat/cramping =less chance of RROD type failures? More room=cheaper cooling system?

All just throwing darts at a wall.

Edit: guess flashcard readers aren't "free" either, whoops.

 

[Brad Grenz]

Yeah, but those numbers only work out if you chose the tiniest number you can imagine for the flash costs. And even then it barely works out. If it's $5 and not $2 all of a sudden you're down 3 billion instead of maybe breaking even. And even $5 is extremely generous to the argument. It's not hard to see it actually being more like $8-15. In that situation the cost is just going to be passed along to the consumer. I don't know about you, but I'm not eager to go back to the age of $85 games.