Wow! IBM is worried that PS3 processors may"break"

[soonersfan60]

This is an odd thing for IBM to say:


Regarding to the Cell itself, specifically concerning the fact that it will feature eight cores, of which seven will be used by the PlayStation 3, Reeves noted that should one of these cores fail: “It’s just like a reliability failure on your TV or DVD recorder. If it’s within warranty, you send it back. If it’s not, your game doesn’t work anymore.”


Wow. I mean, I realize that things can break, but for the developer of the PS3 processor to discuss this just seems odd.

Here's the article:

http://www.gamasutra.com/php-bin/news_index.php?story=9999

[camoor]

This is an odd thing for IBM to say:



Wow. I mean, I realize that things can break, but for the developer of the PS3 processor to discuss this just seems odd.

Here's the article:

http://www.gamasutra.com/php-bin/news_index.php?story=9999

Where did you get that he is worried? Also where does he say "break" (you put it in double quotes, that indicates that it is his word and not just yours)

[mtxbass1]

Note to self: Buy the extended warranty.

[TimPV3]

Isn't the reason for 8 cores should one of the seven not work there's the 8th one to fall back on?

And the comment seems like he was pointing out an unlikely possibility or was atleast asked what would happen if one failed. It just seems like he's saying if a hoard of elephants stampedes through your house you'd have to call your insurance company but if you don't have any you're fucked. While that's what you would have to do, it's unlikely elephants will stampede through your house.

[Mookyjooky]

note to self, never buy this thing.

[Metal Boss]

This doesn't make it seem like they are concerned, merely pointing out the cause & effect of a certain possibility.

I think it's even more odd the staggering number of 360s that have failed since launch, hopefully the PS3 is abit better quality at that price tag.

[Strell]

People.

High tech shit is high tech shit.

And for the most part it's all equally liable to break.

Your car's AC might die tomorrow. That won't stop you from crankin' it up.

[epobirs]

Isn't the reason for 8 cores should one of the seven not work there's the 8th one to fall back on?

No, the Cell as designed is intended to have 8 functioning SPEs in addition to the PPC central core. The problem is that the Cell is a massive hunk of silicon to accommodate all that transistor real estate at 90nm. It proved nearly impossible to manufacture with all 8 SPEs working at the specified speed. (This was after they cut the intended clock rate by almost 15%.)

But the defect rate was such that if you planned for one of the SPEs to be defective, although which one could only be determined in testing, then you could get almost OK yields if you only needed 7 of the SPEs to actaully work in the PS3.

At E3 2005 Sony happily touted the 8 SPEs. By the following E3 they had retreated to 7 because it would otherwise mean a price tag far higher for the PS3. If you have a wafer that costs $5,000 (just a for instance) and only 20 out of 100 expected chips are usable, those 20 chips were very expensive. Those aren't the numbers Sony is dealing with but it's the same math. Severely low yields on 8 SPE Cells would have meant the chip alone cost as much as the announced retail price for the PS3.

Sony and IBM are already working to produce a 65nm version of the Cell that should have much better yields and lower costs. Potentially it could reliably deliver all 8 SPEs for use but by then Sony intends to have sold around 10 millions PS3s, so developers will be forced to ignore the additional SPE under most circumstance to retain compatibility with the first generation PS3s.

One of the SPEs is entirely reserved for OS functions, similar to the functionality in the 360 that extends to all games. Thus developers actually have only 6 SPEs at their disposal.

[BKPartisan]

So if one cell breaks it could cause... *giggle* Massive Damage???

[Metal Boss]

So if one cell breaks it could cause... *giggle* Massive Damage???


That's only if the giant enemy crab from ancient japanese history gets ahold of it.

[Strell]

So if one cell breaks it could cause... *giggle* Massive Damage???

Oh come on. You know that requires flipping the PS3 on it's back.

And with that round bloated alien shit they have going on, it theoretically is not possible.

That was not coincidental.

[epobirs]

The quote really isn't indicative of any big worries. The same thing could happen in a PC. If your processor lost its floating point unit it wouldn't affect the OS itself but a lot of apps would crash that expect an FPU as standard. Pretty much any software that requires a Pentium and will not run on a 486 is like that. The 486 was the first generation of x86 CPU that had the FPU integrated but with the 486SX it remained optional and developers could not assume it was there. Since the first Pentium there hasn't been a mainstream PC CPU that lacked dedicated floating point math hardware.

Keep in mind that the 486SX came into existence because Intel had a certain defect rate on the FPU and offering a cheaper 486 that lacked the FPU was a way to rescue those defective but otherwise functional units from going in the trash. Demand for the 486SX became so great that Intel to start making them that way on purpose.

[jkam]

I'm not really worried about high tech things breaking. Yes it does happen but with Microsoft's problems with the 360 and Sony's track record I'm not exactly looking forward to buying either system for awhile. It's obvious that Sony always plays the game of we'll work out all the kinks out once the system is out already. They love to push tech that just isn't ready yet. Oh you have to buy a 2nd PS2 GREAT! How about another PS3 EVEN BETTER!

[soonersfan60]

Where did you get that he is worried? Also where does he say "break" (you put it in double quotes, that indicates that it is his word and not just yours)

I put the word in quotes to denote a euphamism. I did not intend to mean it as a direct quote of the IBM guy, but I can see where it was taken that way. As far as him being "worried," you don't talk about something unless it is on your mind. I mean, I don't see Motorola introducing a new phone and telling people that it might break and that you'll be out of luck if you don't buy an extended warranty. Best Buy employees might say that, but I just thought it was odd for the manufacturer to being saying that. IBM doesn't even get any of the extended warranty money, so why is this even something he would be thinking about. Usually manufacturers discuss their products in glowing terms. Sure electronics/hi-tech stuff can break, but do you hear other manufacturers talking about it? Most of them try to pretend that it doesn't happen (like MS ;) ).

[Graystone]

Thats not a huge deal, thats just like with any eletronic device. If one thing breaks it could shut down the whole system. All I got to say is if your dumb enough to pay $600 for a PS3. At least justify yourself and buy the extended warranty.

[KaneRobot]

I have the solution. We need EIGHT MORE CORES.

[Kayden]

Eight cores? In one chip? At 90nm?

Wiskey-Tango-Foxtrot

I must have been sleeping for the past 2 years. I'm not going to touch a PS3. The chip is either going to need water cooling or it'll be the size of a loaf of bread.

Honey, whats that smell? Oh, its just the MeltyPlasticStation.

[graf1k]

Eight cores? In one chip? At 90nm?

Wiskey-Tango-Foxtrot

I must have been sleeping for the past 2 years. I'm not going to touch a PS3. The chip is either going to need water cooling or it'll be the size of a loaf of bread.

Honey, whats that smell? Oh, its just the MeltyPlasticStation.
These aren't 8 symetrical full-on CPU cores. There one main core and six much less powerful cores that handle individual things like audio or physics and because they can be used in any configuration it is supposedly more flexible than a traditional CPU.

From Wikipedia:



Cell is a shorthand for Cell Broadband Engine Architecture, commonly abbreviated CBEA in full or Cell BE in part. Cell combines a general-purpose POWER-architecture (http://en.wikipedia.org/wiki/IBM_POWER) core (http://en.wikipedia.org/wiki/Multi-core_%28computing%29) of modest performance with streamlined coprocessing (http://en.wikipedia.org/wiki/Coprocessor) elements which greatly accelerate multimedia (http://en.wikipedia.org/wiki/Multimedia) and vector processing (http://en.wikipedia.org/wiki/Vector_processing) applications, as well as many other forms of dedicated computation.

[Quackzilla]

I seriously doubt IBM would make a fragile computer chip.

Their hardware is like fucking tanks, why would they change?

[camoor]

I put the word in quotes to denote a euphamism. I did not intend to mean it as a direct quote of the IBM guy, but I can see where it was taken that way. As far as him being "worried," you don't talk about something unless it is on your mind. I mean, I don't see Motorola introducing a new phone and telling people that it might break and that you'll be out of luck if you don't buy an extended warranty. Best Buy employees might say that, but I just thought it was odd for the manufacturer to being saying that. IBM doesn't even get any of the extended warranty money, so why is this even something he would be thinking about. Usually manufacturers discuss their products in glowing terms. Sure electronics/hi-tech stuff can break, but do you hear other manufacturers talking about it? Most of them try to pretend that it doesn't happen (like MS ;) ).

Yes, but the whole thing seems as if it's taken out of context. I highly doubt the IBM rep sought out the reporter and expounded on his own - it sounds like he was answering a specific question with a straightforward answer.

To me your post read like a Michael Moore documentary - it's not a flat-out lie, but a critical observer could percieve a bunch of selective editing under a sensationalistic and misleading title.

[Apossum]

To me your post read like a Michael Moore documentary - it's not a flat-out lie, but a critical observer could percieve a bunch of selective editing under a sensationalistic and misleading title.


QFT.

[Chris in Cali]

To me your post read like a Michael Moore documentary - it's not a flat-out lie, but a critical observer could percieve a bunch of selective editing under a sensationalistic and misleading title.


Indeed, although I'm glad it didn't read like a Bush adminisration statement because then it would have most definitely been a flat-out lie.

[Skylander7]

All these worries over an expensive piece of plastic and silicon....

Am I the only one who flung an old generation 1 Genesis off a roof, brushed off the dirt & grass, smiled for barely missing a pile of dog shit.. then took it inside, plugged it up, and played X-Men? Now that's the good ole days.. when consoles were built like Tonka Trucks.

In all seriousness though, c'mon. I'm not a consumer all into the console wars (because whoever wins a console war is the one who gains marketshare, thus eliminating the possibility of quality games on another platform), but the Cell is going to raise a few power bills (not only to keep the fucking thing running, but to keep the AC on to cool off your house or apartment). This whole MS vs. Sony thing is starting to feel like Bush vs. Kerry with a Nintendo Nader.. only this time, it's as if somebody dumped money in Nader's campaign fund and he kicked the donkey and elephant in the balls.

[Vinny]

I seriously doubt IBM would make a fragile computer chip.

Their hardware is like fucking tanks, why would they change?

Because... they worked with Toshiba and Sony on this thing?

And this thread is so pointless, yet humorous. We all know that every 'may' break, why is it news that the Cell 'may' break? Is it not allowed to break or something?:lol:


Am I the only one who flung an old generation 1 Genesis off a roof, brushed off the dirt & grass, smiled for barely missing a pile of dog shit.. then took it inside, plugged it up, and played X-Men? Now that's the good ole days.. when consoles were built like Tonka Trucks.

Yeah? Well get over it. Those glory days are gone with... these days rather than hearing a customer service rep say "enjoy your new system" you hear them ask "would you like to purchase our extended warrenty plan with that?". Well, that's the technological advancements we've made in 15 years time.:-(

[PrivatePixel]

As far as him being "worried," you don't talk about something unless it is on your mind. IBM doesn't even get any of the extended warranty money, so why is this even something he would be thinking about. Usually manufacturers discuss their products in glowing terms. Sure electronics/hi-tech stuff can break, but do you hear other manufacturers talking about it?

I think you jumped the gun on what Tom Reeves (VP of semiconductor and technology services at IBM) was saying / alluding to in his interview with Electronic News. Tom was talking about problems and challenges producers involved with chip manufacturing faced in general; the Cell was used to illustrate an example that IBM faced, and was not the focal point of the interview (which can be read in its entirety here (http://www.reed-electronics.com/electronicnews/article/CA6350202.html?industryid=21365)).

epobirs offered some great insight into the workings of the Cell's SPEs, which was brought full circle by Tom Reeves. Fact is, regardless of who manufactures the chips in any console, defects will be encountered. What Reeves did say was that larger chips (such as the Cell CPU) resulted in lower yields due to the increased likelihood of defects occurring; with logic redundancy* built into the chips, they can potentially double their yields. What you mostly end up with are CPUs with eight, seven or six operational cores. Those with six or fewer operational cores are set aside and will not find their way into a PS3; only those with seven will. The really good batches (all eight cores operational) will be used in applications such as medical imaging, aerospace and defense. A CPU with six or less operational cores aren't totally useless; it's a matter of IBM finding the right applications for them, i.e. in non-critical environments.

* duplication of critical system components to increase reliability, intended as a backup or fail-safe

All chips, regardless of manufacturer, come with a reliability failure rate inherent in them. Theoretically speaking, if a point circuit fails in one of the SPEs, it will likely knock out that core. With redundancy and the use of electronic fuses (instead of laser fuses), plus the presence of the eight SPE as a spare, a hit to one of the cores will not affect performance.

Here's the quote from Tom Reeves that is the heart of your contention:
It’s just like a reliability failure on your TV or DVD recorder. If it’s within warranty, you send it back. If it’s not, your game doesn’t work anymore. You’ll always have choices about how reliable you want to make a chip with burn-in. Most chips that go into the consumer marketplace on things such as camcorders or DVD players aren’t burned in. But you can add burn-in and improve reliability 5x to 10x. It's extra cost. Certainly, a company like Sony adds that in.
In the same way that PC manufacturers employ a 24-48 hr. burn-in period for newly-built PCs, that same methodology is also being adopted by Sony to ensure that the Cell CPU leaves the factory with all seven SPEs working (remember: the eighth is a spare and redundancy is built into all of the SPEs).

I would highly recommend reading the articles on the Cell's architecture and how the PPE and SPEs work interconnectedly on technical sites such as Ars Technica to better understand what Tom was trying to get at, instead of using a sound bite as the gist of your argument.

[epobirs]

These aren't 8 symetrical full-on CPU cores. There one main core and six much less powerful cores that handle individual things like audio or physics and because they can be used in any configuration it is supposedly more flexible than a traditional CPU.



This is inaccurate. THe full spec Cell is nine distinct cores in terms of processing elelments that can operate with some independence and could be useful if produced as a chip unto itself. In fact, the Cell is very much like a 70s supercomputer design, except everything is on a single slab of silicon rather than each core being a separate box full of board with a high speed (for the era) interconenct running between them.

1 PowerPC based core and 8 SPEs are found in the Cell design but that has proven to difficult to produce in quantity at the needed price. Since one SPE is expected to fail in most units, this is found and disable in testing then the resulting chip put through a burn-in process to determine its reliability at the intended clock rate.

When they say 8 cores they mean the PPC central controller and 7 SPEs.

[Kayden]

So whatll they do if all 8 SPEs are good? Just disable one at random?

[Z-Saber]

Eight cores? In one chip? At 90nm?

Wiskey-Tango-Foxtrot

I must have been sleeping for the past 2 years. I'm not going to touch a PS3. The chip is either going to need water cooling or it'll be the size of a loaf of bread.

Honey, whats that smell? Oh, its just the MeltyPlasticStation. Post of the year!

[epobirs]

So whatll they do if all 8 SPEs are good? Just disable one at random?

No, those 1 in a 100 (or whatever the real ratio is) chips go into high end applications being developed around Cell. Boxes that sell for many thousands of dollars, even tens of thousands of dollars. The sort of high end applications that were once the realm of companies like SGI but have since largely fallen to PC based solutions sometimes enhanced with an application tuned FPGA-based accelerator.

A $2500 Cell with a bunch of RAM on a PCI-E board may sound expensive but if it can match or beat a $10,000 GPGA solution, it's a good buy.

A lot of this is much ado about nothing. The PS3 is a console, which means everything on the spec is either there for the developers to exploit (or reserved for the OS and still required) or the machine is a loss. If developers want to do the work of creating software that deals with varying hardware resources the PC is wide open with no royalties to pay for getting a disc made.

Thus, if an internal element of a console goes bad the whole machine is a loss or in need of repair. On a PC there are minimum requirements but software will adjust to things above that. If someone was writing software for a Cel board living in a PC, they could design it to get along with less SPEs and offer improved performance if there are more than the usual. Which means the software would be capable of supporting future Cells that might offer more SPEs. Which is how PC applications should be done.

But it isn't what consumers or developers want from consoles. They want the performance to be exact and highly predictable. What the consumer gets at home should be exactly what the developer got in his studio and exactly what they showed on X-play.

Which means a broken SPE is effectively a dead PS3. The same could be said for any of the three PPC cores in the Xbox 360. Or any of the major elements in either machines GPU. A GPU could have a bad vertex shader unit on one of its pipelines and if it numerically is a pipe unlikey to be assigned to vertex shading, the machine may run a lot of games without anything going wrong. But if a game tries to talk to that specific shader, the whole box might as well melt in a smoking puddle.

[doraemonkerpal]

So if one cell breaks it could cause... *giggle* Massive Damage???

:rofl:

excellent posts by privatepixel & epobirs! :)

[graf1k]

This is inaccurate. THe full spec Cell is nine distinct cores in terms of processing elelments that can operate with some independence and could be useful if produced as a chip unto itself. In fact, the Cell is very much like a 70s supercomputer design, except everything is on a single slab of silicon rather than each core being a separate box full of board with a high speed (for the era) interconenct running between them.

1 PowerPC based core and 8 SPEs are found in the Cell design but that has proven to difficult to produce in quantity at the needed price. Since one SPE is expected to fail in most units, this is found and disable in testing then the resulting chip put through a burn-in process to determine its reliability at the intended clock rate.

When they say 8 cores they mean the PPC central controller and 7 SPEs.
Right, but in my understanding 1 SPE is not equal in processing power to a traditional Intel/AMD x86 class processor at the same frequency. So the Cell with 8 SPEs is NOT equal in processing power to a 8-core AMD/Intel chip of the same speeds. That's all I was saying in my original post. And as I understand, the SPEs still need the main PPE to organize and control the SPEs.

Again, from Wikipedia:



To achieve the high performance needed for mathematically intensive tasks such as decoding/encoding MPEG (http://en.wikipedia.org/wiki/MPEG) streams, generating or transforming three dimensional data or undertaking Fourier analysis (http://en.wikipedia.org/wiki/Fourier_analysis) of data the Cell processor simply marries the SPEs and the PPE via the EIB to give both access to main memory or other external data storage. The PPE which is capable of running a conventional operating system has control over the SPEs and can start, stop, interrupt and schedule processes running on the SPEs. To this end the PPE has additional instructions relating to control of the SPEs. Despite having Turing complete (http://en.wikipedia.org/wiki/Turing_complete) architectures the SPEs are not fully autonomous and require the PPE to initiate them before they can do any useful work. Most of the "horsepower" of the system comes from the synergistic processing elements (SPEs).




I do agree with what you said about the SPEs not being anything to worry about. Hell, a capacitor gets bent the wrong way in a PS3 or 360 and the whole thing is fucked. This is nothing new.

[soonersfan60]

To me your post read like a Michael Moore documentary - it's not a flat-out lie, but a critical observer could percieve a bunch of selective editing under a sensationalistic and misleading title.

What an insult. :bomb: If you bothered to click on the link, you would see that I didn't selectively quote anything. That is exactly what Gamasutra.com reported. Now maybe they did, but I don't know. Even still, I think what the IBM guy said was funny. Everyone is piling on Sony, their exectives keep putting their feet in their mouths, and now their allies aren't really helping their cause much. Yes, things break and electronics are more complicated than ever, but it just doesn't sound reassuring for the company that made the thing to talk about how it might break. It may be honest and true, but it doesn't exactly inspire confidence among people who might be sitting on the fence. Just the way he said this I find it to be very amusing. (...Buy the warranty or you might end up with a broken piece of hardware that can't play games...:bouncy: )

[epobirs]

Right, but in my understanding 1 SPE is not equal in processing power to a traditional Intel/AMD x86 class processor at the same frequency. So the Cell with 8 SPEs is NOT equal in processing power to a 8-core AMD/Intel chip of the same speeds. That's all I was saying in my original post. And as I understand, the SPEs still need the main PPE to organize and control the SPEs.

Again, from Wikipedia:




I do agree with what you said about the SPEs not being anything to worry about. Hell, a capacitor gets bent the wrong way in a PS3 or 360 and the whole thing is fucked. This is nothing new.

I never suggested they were on par with a X86 core but the Cell is not a general purpose CPU. It's designed with some specific specialties in mind. The Cell can even suck at some run of the mill integer operations compared to a much less expensive x86 CPU. (Especially since the Cell's PPC core lacks the branch prediction functions that have long boosted performance on big desktop/server processors.) But nobody is buying a PS3 to run Excel and what integer loads it sees will be handled quickly just by the brute force of the clock rate. High end system incorporating one or more Cells will have conventional X86 or PPC CPU to deal with the more mundane tasks efficiently. For those things the PS3 is required to deliver, the Cell will offer far more performance than any existing x86 could on it's own.

You might recall at E3 2005 Sony ran a bunch of demoes where the Cell was the sole renderer of some nice and complex graphics in realtime. Any x86 a year later would crawl on those demoes because they required the kind of multi-threading DSP type functionality the Cell or a serious GPU offers. Those demos were a holdover from the early days of the PS3 design cycle when the machine had no dedicated GPU. It was going to be a box of Cells (3 or 4) and RAM which developers could allocate in any way that suited their project.

Once it became apparent there were going to be serious yield issues (not to mention problems with the multiprocessor support that still haven't been resolved) the need for a dedicated GPU became inescapable. Pretty soon they were in meetings with Nvidia, which is why Nvidia participation was announced so late in the game.

The SPEs need a controller but that doesn't mean they all have to reside on the same slab of silicon. The SPEs are for the most part DSPs in their heritage. A DSP is rarely used on its own without a general purpose CPU running the show. In most of those exceptions the DSP has a lightweight CPU, like a stripped down ARM core, to serve that purpose.

THere isn't really anything new here about bringing an external function onboard to have the fastest possible interconenct with the shortest possible wire lengths. The first hardware shaders for use with Renderman were entire boards that installed into SGI boxes. The GeForce 3 took what had a bit less than a decade earlier been a $100K machine the size of a small refrigerator (no small coincidence since active refrigeration was used for cooling) and reduce it down to a single PCI/AGP board.

Even earlier, when the Amiga was first being demoed under the name Lorraine, each of the custom chips was a colelction of wirewrap boards the size of a shoe box. By the time the A1000 shipped each was just one chip on the motherboard.

[epobirs]

A real question is why this IBM guy felt compelled to state the obvious and provide grist for the mill. There is a reason why marketing departments often like to keep engineers isolated from the press.

[graf1k]

Right. My point wasn't that YOU said that. I was just restating what I said in my original post. You quoted my post and called in inacurate so I was just proving my original point. The person I originally quoted seem to take the cell as an eight-core CPU that would be the size of and have the thermal output of a x86 class eight-core CPU.

[BKPartisan]

Hmmm, would it be possible to inflict Massive Damage in a zero-g environment?

Oh come on. You know that requires flipping the PS3 on it's back.

And with that round bloated alien shit they have going on, it theoretically is not possible.

That was not coincidental.

[Kayden]

Hmmm, would it be possible to inflict Massive Damage in a zero-g environment?

Superman did it in 4. 8-)