IBM's Cell Processor — Not Just for PS3 Anymore

TechFreep writes to tell us that IBM has released a new line of QS20 Blade Servers based on the processor they developed for the Playstation 3. From the article: "Today IBM announced a new line of high-powered QS20 Blade Servers intended for use in seismic research, encryption, digital image rendering and military surveillance applications. Each QS20 will include two nine-cored Cell Processors clocked at 3.2Ghz apiece, which were developed along with Sony and Toshiba for Sony's upcoming Playstation 3 console. As Playstation 3 isn't scheduled for release until November, the QS20 will mark the first application in which the highly-touted Cell will be available to consumers."

Nothing for you to see here. Please move along.

[Doytch]

Aint that the truth. The damn things aren't even being built yet.

Lost in the noise is the dying Sun.

[Anonymous Coward]

Lost in the noise is the fact that the multi-core Cells are IBM's answer to Sun's chip multiprocessor (CMP) -- i.e., Niagara and Rock. Nothing about the CMP is new or unique. Academia has already done 10+ years of research on these beasts, and industry has ready access to the academic results.

IBM already has a CMP. Both AMD and Intel will soon have CMPs. Here, CMP does not mean duo-core; CMP means at least 4 cores.

The window of opportunity for Sun has now closed.

What is ingenious about IBM is the fact that IBM is essentially using the R&D in its consumer-electronics division (that builds processors for game consoles and other toys) to advance R&D in the business-oriented high-performance-processor/high-end-server division. Building electronics for toys has actually strengthened non-toy products.

When will Mattel and Hasbro start selling their own supercomputers?

Re:

[ePhil_One]

When will Mattel and Hasbro start selling their own supercomputers?

Mattel has already tested the computer waters [intellivisionlives.com]

Re:

[Mateito]

When will Mattel and Hasbro start selling their own supercomputers?

Cool. A supercomputer that turns into a robot OR a jet-fighter!

Re:

[jonatha]

What is ingenious about IBM is the fact that IBM is essentially using the R&D in its consumer-electronics division (that builds processors for game consoles and other toys) to advance R&D in the business-oriented high-performance-processor/high-end-server division. Building electronics for toys has actually strengthened non-toy products.

It may be ingenious, but there's only one division (STD, I kid you not, unless they've renamed it again...)

Re:

[Wesley Felter]

Lost in the noise is the fact that the multi-core Cells are IBM's answer to Sun's chip multiprocessor (CMP) -- i.e., Niagara and Rock.

Nope; Cell and Niagara are optimized for totally different uses.

Re:

[raftpeople]

And the cell design began 2 years before the niagara began.

Re:

[God of Lemmings]

What is ingenious about IBM is the fact that IBM is essentially using the R&D in its consumer-electronics division (that builds processors for game consoles and other toys) to advance R&D in the business-oriented high-performance-processor/high-end-server division.

This is nothing new. It has been part of IBM's business model for years to take orders for projects, complete the projects, and then turn around and use the resultant products in their own hardware. Take the powerpc 970 for example; made f

Re:

[Hal_Porter]

But Niagara has 8 cores, each capable of runnning 4 threads. Importantly, they all use the same instruction set and have access to external memory. They all have MMUs too.

http://www.theinquirer.net/default.aspx?article=19 423 [theinquirer.net]

It's an SMP beast.

So you could run a traditional OS on it, and all the cores would be used if needed. On a Cell, there is a single PowerPC core, which is a bit underpowered with no out of order execution for example. The bulk of the processing power is in the SPEs. But these have a diff

Now imagine a...

[Kesch]

Beowulf cluster of these.

Oh, wait. I suppose this [slashdot.org] will be close enough.

Different markets

[Cybert4]

Do you want your nuclear test performed with the same chip that's used to play Spiderman X? There are a lot of decisions with regards to redundancies, ECC, and so on that go into a design. If it was concieved as a chip for the playstation, you could end up with problems later.

But I do hope it works. This kind of thing would be great for neural implants on the way to the singularity.

Re:Different markets

[Anne Thwacks]

Nope. It was concieved as a component for massively parallel processing, but using it in the playstation was the only way to get the volume production that I need to get the kind of reliability I want for my nuclear simulations.

This chip will be to the Pentium what the Pentium is to the 6502 - once the tool chain is understood by software designers. Actually, better, because Cell architecture uses way less gates than pipelining for way more throughput at the same clock speed and feature size. Hell, I might even retire my Sun Niagras.

Sure it won't run Windows very well, but hell, Windows doesn't run very well anyway, and I Don't Care (TM).

Yes, it does run Linux. What about NetBSD?

Re:

[jamstar7]

Nope. It was concieved as a component for massively parallel processing, but using it in the playstation was the only way to get the volume production that I need to get the kind of reliability I want for my nuclear simulations.

OK, lemme see if I get this right...

I buy a PS3. Hook it to the net. Leave it on 24/7. Your nuke simulating supercomputer comes over the net & scams spare cycles off of me a la SETI@Home.

Did I miss anything?

Re:

[be-fan]

I'm not sure if this is a joke or what.

He was implying that spending $1bn+ designing and fabbing an agressive design like Cell is only possible if you have some high-volume consumer product to put it in. Otherwise, it'd just be too expensive.

Re:

[dfghjk]

Hmmm, didn't stop IBM from doing the 970/G5.

Re:

[be-fan]

IBM didn't spend a lot of money designing the G5. They just took Power4, which already had a self-sufficient market in high-end servers, and replaced the SMP fabric with an elastic-IO bus. Then they almost completely neglected to maintain/update it for three years.

Re:

[SanityInAnarchy]

Does it run an entirely open source Linux? I don't exactly trust Sony after how they bastardized Linux for the PS2.

Re:

[be-fan]

Sony is the primary customer of Cell, but IBM's doing the development. They're doing the Linux port, compiler port, etc.

Fully Open Source Linux?

[Lonewolf666]

On the processor? AFAIK yes, the few existing Cell Workstations run Linux. And there are no reports of that linux version being bastardized.

On the PS3? I believe that when it is reported by independent reviewers. Because it would be a radical break from the current business model for console:
1) The console is sold at a loss
2) The console maker recoups the losses from licens fees the game makers have to pay
3) The console is heavily protected against the starting of unlicensed software.

Having an unrestricted

Some games on Linux, then.

[Lonewolf666]

Boot Linux. Insert Linux based game DVD from company X which is not so concerned about cheats, especially in single player mode (players can only ruin the game for themselves). Company X does not pay any license fees to Sony, and I doubt there is a law Sony could use to force them.

Result: Less income for Sony.

And that is what all console manufacturers have tried to avoid so far. Now it is possible that Sony has a different business model this time, or that they are simply not getting it and will have a rude

Re:

[SanityInAnarchy]

Bastardized as in, you HAVE to buy it. Even if you have a spare IDE hard drive lying around, and you intend to download Black Rhino Linux. Even if you have multiple ways to run custom code, so it's not an issue of booting an unsigned disk. Even given all of these, you STILL need their kernel with their custom binary kernel stuff, which means you need to buy the dev kit, which last I checked means some $200, including a hard drive that you shouldn't need (I wanted to boot from the network).

Also: The hard

Re:

[SanityInAnarchy]

They didn't even do that for the keyboard and mosue (it's a really nice keyboard with a built in USB port for the mouse.)

How's that relevant? I can buy any old keyboard/mouse I want, and in this case, I didn't want any, I wanted it to boot from the network and, if anything, I could SSH in.

If you can get hold of an RTE disc and an FFXI HD you can format it and install.

Key word here: Format. You seem to imply that you can share the drive between them, but I've heard just the opposite.

And anyway, why ca

Re:

[SanityInAnarchy]

Games with HD support check for the official Sony drive, but the Linux Kit RTE doesn't.

Garh, that's annoying, but at least it's not Linux being bastardized there. Although swapping between a 160 gig Linux homebrew drive and a 20 gig official drive for playing games would be a hell of a lot more annoying than just partitioning it.

The GPL allows for the sale of Linux distros for money, you know that.

Yes, I do. I also know that anywhere else something like this has been tried, there's usually a very good

Re:

[Anne Thwacks]

The cell processor is nowhere near as powerful as you're making out. Any you know this because?

I worked for many years designing architectures like this for specialised database operations. I know how powerful this architecture is for certain types of work. I suspect you don't.

as for the person suggesting that I would be using it in a SETI type operations: No, I wont, but I think I know someone who might. (But if I told you, I would have to kill you).

Re:

[Ilgaz]

When IBM says "Power" or "Cell" processor, they speak about an architecture standard which is relatively open, check info about it on their new http://www.power.org/ [power.org]

For example while not planning a game console, NEC is also in Power board as well as Toshiba. I can think NEC cares about their supercomputing division and Toshiba cares about their future media plans.

Those Cell processors they use will not be the same thing Ps3 runs. For example while my Quad G5 (PPC970) 2500 Mhz is fine for pro HD work if nece

Re:

[be-fan]

The Cell processors are the same. Cell already supports ECC in the caches and in the XDR memory.

Re:

[dfghjk]

"For example while my Quad G5 (PPC970) 2500 Mhz is fine for pro HD work if necessary upgrades (SCSI,HDI etc) done, a "real" Power5 AIX workstation used in petrol industry, military etc. would show it as a toy. It has near mainframe features packed to a workstation."

What features are those? Name some that make it not "a toy" compared to your PowerMac.

The 970 is a Power4 derivative. Power5 is newer.

I'm actually quite excited

[HatchedEggs]

Not necessarily about this announcement of course as most of us have seen it coming for some time. However, the concept of the core processor is quite strong and I think that it will play an important role in computing in the future.

A little tid bit about all that: http://www.blachford.info/computer/Cell/Cell0_v2.h tml [blachford.info]

Re:

[joe_bruin]

Don't get too excited. The Register [theregister.co.uk] reports that just one of these blades will cost you $18,995.

Re:

[donaldm]

Well back in the mid 1980's some PC's would set you back almost $10k and they only ran DOS while a workstation (full colour graphics) could cost anything from $7k (we got a very good deal from SUN and SGI) to $100k+.

The dual cell machine costing $18k may be quite cheap for what it can do. Not everything revolves around PC's.

If companies want serious number crunching power it is not unusual to see purchases in the millions.

Re:

[fm6]

Attention, would be science and technology writers: don't babble for 3 or 4 pages before you get around to explaining what you're talking about and why it's so kewl. In this case, the Cell is a ordinary Power processor core supplemented by a bunch of on-chip vector processors. Which actually is pretty damned cool, but not as mysterious as Blachford makes it sounds.

Transputer

[master_p]

Let's not forget the Transputer! it was there 20 years before the Cell CPU, and there was parallel programming languages like Occam or Concurrent C.

More info here:

http://en.wikipedia.org/wiki/Transputer [wikipedia.org]

Re:

[Breakfast Pants]

I'm guessing you read about this DRM emphasis on Slashdot.. I remember vividly a series of submissions about it.

"Consumers"

[TCM]

It's not just that I hate this word, but what has it got to do with these blade servers? Are they edible?

Re:

[Mr2001]

This is a pretty standard use of the word. "Consumers" are the complement of "producers", in economics as well as computer science (e.g. a lexer produces tokens which are consumed by a parser).

Producers make a product; distributors and retailers buy it in order to pass it on to someone else; consumers buy it for their own use. The article is saying the QS20 is the first thing people will be able to buy and use that contains a Cell processor.

High availability

[Doctor Memory]

the highly-touted Cell will be available to consumers.

Sure, as long as you're spying on someone else's armed forces, analyzing their armaments via seismograph and taking pictures that you then encrypt to send to your superiors. Bet they'll be backordered for the holidays!

Re:High availability

[jthill]

Cool. What's the working title on THAT game?

Wait...

[UbuntuDupe]

What exactly is a blade server? Is that what they were talking about on that dumbed-down uncreative commercial where the two guys are like "What's a blade server?" "This is." "This is a blade?" "Uh, yeah."

Re:

[kv9]

What exactly is a blade server?

it is a server with blades [ibm.com].

and here's a link to the IBM press release [ibm.com] because it looks like linking to the source has gone out of style.

Re:Wait...

[whitehatlurker]

What exactly is a blade server?

it is a server with blades.

That's begging the question. A blade is a modular computer on a board. You team these up into a chassis and have them communicate together and you've got a blade server.

Re:

[mnmn]

My friend had a blade server eons ago.

He had a Tyan motherboard with CPU daughterboards. He could add upto 4 CPUs but he only had 2.

I wonder if Dell Poweredges lined up and interconnected with Ethernet (tm) will qualify as a blade server.

Re:

[epiphani]

Again, bad explanation.

A blade is a server. Instead of having plugs and wires to connect it to everything else, it has one nice big interface. It slides into an enclosure that provides the interface with everything it needs.

The enclosure will do handy things like handle all that power stuff, provide some integrated administration, hardware management and such, provide cooling and airflow.

Better info

[marleyboy]

Maybe take a peek at the Wiki entry [wikipedia.org] about the cell processor for a good background of what it is capable of.

Re:

[suv4x4]

Maybe take a peek at the Wiki entry [wikipedia.org] about the cell processor for a good background of what it is capable of.

Wow... wow... We screwed up. Boy, are our faced red! Hehehe.
We're switching back to PowerPC in 2007 then.. Yea..

-- Steve Jobs

Re:

[dfghjk]

Please explain (aside from the few more MHz part).

Misleading ... "developed for PS3"

[Anonymous Coward]

I think that stating that the Cell was "developed for PS3" is misleading. The processor, while certainly earmarked for the PS3, was designed all along to go into a full range of computing devices, PS3 was just the most visible (and likely to ship the most units initially). Saying it the other way makes it sound like the chip was developed exclusively for the PS3 and IBM is "just finding uses for it", which is not the case.

I see

[Anonymous Coward]

Clearly this is bad news for Sony.

For some reason I haven't figured out yet.

The plan all along...

[Invisible Now]

This has always been the plan and was a guiding principal in the Cell's design. Amortize the cost of a very usefull (To federal customers) chip over the estimated 40-60M playstations that will use a very similar (But not identical) design. From the beginning the chip was dual-purpose designed with very high speed interconnects and protocols for massive parallel-ism.

$29/chip x 64k chips = more ops per buck than ever - thanks to the world's gamers...

The problem for both PS3 and the NSA, etc is IBM's 10-20% yields. PS3 for Christmas? They better get up the curve fast...

BTW - Anyone remember back to when the Soviets used to buy up Ataris and canabilize their chips for sonobouys?

Re:

[uarch]

The problem for both PS3 and the NSA, etc is IBM's 10-20% yields. PS3 for Christmas? They better get up the curve fast...

Yeah. That would be a huge problem! Good thing yeilds aren't that low.

(You're probably referencing the interview that was on tech sites a few months ago. The write-ups completely misunderstood what was being said.)

Re:The plan all along...

[donaldm]

Even with a 20% yield the cell is going to cost between US$15 to US30. The cell will be made on a Silicon on Insulator (SOI) wafer (300mm diameter) costing approx US1200. In huge quantities (and the for PS3, Wii and Xbox 360 there will be huge quantities) this will bring down the cost of the wafer and thus the overall cost of the chips on SOI will be cheaper.

On an interesting note the Cell chip is approx 250 square mm so taking the 20% option on a 300mm diameter SOI wafer (assume US$1300 for wafer and assembly) the overall cost of the chip is US$23. The fab machines are about $11M each and are capable of producing completed wafers between 25,000 to 35,000 per year. Great if you control the process and IBM, Toshiba and Sony (to a lesser degree) do.

Do a Google search on "Silicon On Insulator" as a starting point.

Re:

[antime]

When you say "Cell chip" I assume you mean an implementation with one PPE and eight SPEs? I would be interested if you have figures for smaller versions, designed for use in less processing-intensive applications (digital TV, DVD players etc).

Re:

[donaldm]

I don't have the figures for a cut-down cell (if such a thing exists), although I would assume that if you only want a PPE (not negotiable) and say 4 SPE's then standard cell chip yields would increase to say 50~70%. Of course if the clock frequency is lower then I would also assume that yields would go up as well. The overall size of the standard cell chip is huge and since you are going to put rectangular chips on a circular wafer then you are never going to get much more than 80~90% yields (reasonable gu

Re:

[frogstar_robot]

BTW - Anyone remember back to when the Soviets used to buy up Ataris and canabilize their chips for sonobouys?

Link?

Now we know...

[Nom du Keyboard]

Now we know why the PS3 Europe release is pushed out into next Spring. With reported 10% - 20% yields for the Cell, there just aren't that many of the buggers around yet.

Re:

[Ilgaz]

In case you aren't joking/sarcasm, PS3 was pushed for Europe because of the BluRay lasers couldn't be produced in timely manner.

http://www.betanews.com/article/Bluray_Causes_PS3_ Delay_in_Europe/1157552739 [betanews.com]

That was why the electronics/optics industry warning them publicly, to end this HD-DVD and BluRay fight in rational way since those lasers are state of the art and not that easy to produce.

How fast *is* it?

[jonatha]

Each QS20 will include two nine-cored Cell Processors clocked at 3.2Ghz apiece

I'm confused. Does that make the damn thing 6.4GHz, 28.8GHz, or 57.6GHz?

Re:

[miro f]

I'm confused. Does that make the damn thing 6.4GHz, 28.8GHz, or 57.6GHz?

no. It's 3.2 GHZ

you don't add clock speeds

Using IBM's Cell Processors in scientific HPC

[rlh100]

I am surprised that no one has mentioned the Slashdot article on the study at LBL on use of cell processors in High Performance Computing:
The Potential of Science With the Cell Processor
http://science.slashdot.org/article.pl?sid=06/05/2 8/047223/ [slashdot.org]

It reference a second article:
Researchers Analyze HPC Potential of Cell Processor
http://www.hpcwire.com/hpc/671376.html/ [hpcwire.com]

This discusses research at Lawrence Berkeley National Laboratory on using the STI Cell processor for scientific computing. From the article quoting the LBL paper:
"Overall results demonstrate the tremendous potential of the Cell architecture for scientific computations in terms of both raw performance and power efficiency,"
and
"We also conclude that Cell's heterogeneous multi-core implementation is inherently better suited to the HPC environment than homogeneous commodity multi-core processors."

The paper went on to say that while the Cell processor was designed for single-precision 32-bit floating performance but with some simple changes to the design it could be optimized for double precision 64-bit floating performance.

This makes a lot of sense if this is the same Cell processor that IBM is using in their blade servers.
Really cheap, really fast 9 core processors!

An interesting read.

RLH

Its not so highly touted

[crossmr]

Chris Hecker for example pretty much says they're crap for game consoles. http://crystaltips.typepad.com/wonderland/2005/03/ burn_the_house_.html [typepad.com]

So, as you know, graphics and physics grind on large homogenous floating point data structures in a very straight-line structured way. Then we have AI and gameplay code. Lots of exceptions, tunable parameters, indirections and often messy. We hate this code, it's a mess, but this is the code that makes the game DIFFERENT. Here is the terrifying realization about the next generation consoles: I'm about to break a ton of NDAs here, oh well, haha, I never signed them anyway.

Gameplay code will get slower and harder to write on the next generation of consoles. Modern CPUs use out-of-order execution, which is there to make crappy code run fast. This was really good for the industry when it happened, although it annoyed many assembly language wizards in Sweden. Xenon and Cell are both in-order chips. What does this mean? It's cheaper for them to do this. They can drop a lot of cores. One out-of-order core is about four times [did I catch that right? Alice] the size of an in-order core. What does this do to our code? It's great for grinding on floating point, but for anything else it totally sucks. Rumours from people actually working on these chips - straight-line runs 1/3 to 1/10th the performance at the same clock speed. This sucks.

I just hope..

[clickclickdrone]

The IBM didn't have a Defender flashback and put in a smart-bomb button on these puppies.

Consumers

[SpaghettiPattern]

QS20 will mark the first application in which the highly-touted Cell will be available to consumers
Consumers?! Do you know how much a QS20 will cost you? These are business systems. You'd have to buy both a blade center and a blade.

Re:

[ZachPruckowski]

It may not matter at all to the PS3. These may only be chips that can't be used in the PS3. In terms of yield, the 7-core, 2.8 GHz Cells in the PS3 should show up a lot on the silicon now, if they're moving from 8-core 3.2 GHz targets to 9-core 3.2 GHz targets.

Re:Playstation 3

[poot_rootbeer]

if they're moving from 8-core 3.2 GHz targets to 9-core 3.2 GHz targets.

I think it's more likely a misstatement in the article than an actual change in what's being produced.

The Cells being produced for the PS3 have one primary processor (PPE) and eight secondary processors (SPE's)--only seven of which need to be functional at 2.8GHz for it to meet Sony's specs--for a total of 9 (or 8 for PS3) 'processors'. Not quite the same thing as 'cores' in the traditional sense of the word.

It's possible that these chips for these new servers are identical on silicon to the PS3 chips, only IBM will not pass them unless all 8 SPE's are functional at the full 3.2GHz clock rate.

Kind of like 386DX vs. 386SX -- instead of throwing away the DX chips that had defective math coprocessors, Intel simply burned out the traces, screened a different label on them, and sold them at discount prices.

Re:

[drsmithy]

Kind of like 386DX vs. 386SX -- instead of throwing away the DX chips that had defective math coprocessors, Intel simply burned out the traces, screened a different label on them, and sold them at discount prices.

Minor correction, that was 486s, not 386s.

Re:

[antime]

It's also how nVidia and ATi do business - if a GPU doesn't pass validation disable the broken rendering units, lower the clock and sell it in budget cards.

Re:

[Yonder Way]

IBM has oodles of fab capacity. Don't forget Apple has dropped off the queue.

Apple was a tiny bit of IBM's production

[SuperBanana]

IBM has oodles of fab capacity. Don't forget Apple has dropped off the queue.

Apple was one of IBM's smallest consumers of PowerPC chips, and always was. The embedded and entertainment market dominates their "queue", and is one of the main reasons the PowerPC series never pushed as hard clock-wise as Intel does; the embedded market sees higher clock speeds as greater power consumption/heat dissipation and more (electronic and thermal) design challenges. When Apple took a hike, IBM didn't shed any tears, and said as much.

I don't have any specific numbers, but I believe Apple's purchases were under 5% of total production. You may say "well, going with Intel was a REALLY stupid idea!" Wrong- before, Apple was "the little fish using embedded-market processors for consumer computers", and the goals didn't match. Now, they're using chips specifically targeted to the markets Apple wants to be in.

Re:Apple was a tiny bit of IBM's production

[Ilgaz]

3% of total PowerPC, yes that is FreeScale official number, if we add the IBM 970 (G5), it will be likely 5%. For people can't believe what they read: Yes, Apple only bought 5% of Power architecture CPUs, in their entire life. If you have a good quality car, you likely have 8-10 Power architecture CPUs.

You can't blame anyone for thinking Power architecture, a living legend or empire is dead after Apple moved away, Apple is run by Steve Jobs who is a publicity genius.

I am sure IBM or Freescale will think twice before doing any business with Apple in the future since Apple attacked PowerPC more than Intel etc. ever dared.

They now claim Xeon based Servers are 5x faster than G5 (yes,PPC970) and their fanboys cheering. Fanboys don't make million dollar server purchases though.

I just wonder what happened to second hand value of those $40.000+ Xserve G5 setups after Mr. Jobs and Apple called them 5x slower. Well, that is what you get for choosing a portable audio oriented company for your server setup instead of IBM or Sun. ;)

I think it is IBM and FreeScale dumped Apple not vice versa.

PS: I own Quad G5 and purchased months after Intel announcement. ;)

Re:

[jdb8167]

What a load of crap. Please cite your reference where Apple claims 5x against a Quad 2.5 GHz G5. On Apple's Mac Pro pages I see: 1.8x, 1.8x, 1.6x, 1.4x and 1.4x on content creation. On SPECInt_rate_base2000 I see 2.1x and on SPECfp_rate_base2000 is see 1.6x.

Apple-Mac Pro-Intel Xeon [apple.com]

Apple dumped PowerPC because IBM couldn't get to 3.0 GHz and they couldn't get to 2.0 GHz with a low power version for notebooks. It isn't complicated and no conspiracy theory is necessary.

Re:

[Ilgaz]

First of all, all this happened because IBM or Freescale didn't give a shit to Apple's constant whining for Mhz while everyone including Intel which introduced this stupid thing isn't using the term anymore. It NEVER MEANT anything anyway.

Secondly, here is your "citation":
"And this award-winning UNIX-based server just got a 5x (1) performance boost over the Xserve G5, thanks to quad-core 64-bit Intel Xeon processing. "

http://www.apple.com/xserve/ [apple.com]

They can decide to become iPod manufacturing, fashion toy sell

Re:

[be-fan]

5x faster is actually pretty close to the mark. Woodcrest (at 3 GHz) runs at a 30% faster clockspeed than the 2.3 GHz G5s in the Xserve. It's about 75% more efficient per clock cycle in integer code (which is what matters for servers), and oh, there are twice as many cores in the new Xserve.

Nobody makes million dollar server purchases using PPC970s. Power4 wasn't a CPU that people bought for the core, which is itself quite mediocre in performance. They bought it because of the huge systems (with gobs of mem

Re:

[Wesley Felter]

The 520 uses POWER5, not PowerPC 970.

Re:

[be-fan]

The iSeries is Power5. The PPC970 is a castrated Power4.

A chronology might help:

Power4: Moderately crappy core, great system architecture/SMP fabric
PPC970: Power4 minus the great system architecture/SMP fabric plus shitty high-latency chipset.
Power5: Redesigned Power4 fixing many instances of crappiness, with even better system architecture.

Re:

[be-fan]

Compared to my 2500 Quad G5, I agree Quad Xeon workstation works 30% faster even with multiple tests. It is not what I talk about, if a company claims their new model is 5x faster, or gives the impression it is not a company to be taken serious.

Why? It's the truth. The previous Xserve was using a CPU that had essentially the same core, and only a 15% higher clockspeed, than the original G5s released in 2003. The new Xserve is using a contemporary CPU (with a brand new architecture), and is using twice as ma

Re:

[be-fan]

Except IBM doesn't have any 3GHz G5s that get over 1000 SPECint/GHz while sucking down only 80 watts.
The top 970MP, which is less than a year old, hits 2.5GHz, with 575 SPECint/GHz, and sucks down 100 watts.

That's actually 2.65x better SPECint/watt on the Core 2 side.

Re:

[renoX]

Well embedded PowerPC are not exactly the same price than a G5, you know..
So it's quite normal than Apple didn't sell a lot of PPC.

What would be interesting to know is what share Apple had of the 'high end PPC', I mean who cares about embedded CPU.. I'm sure the volume of Z80 sold is huge, yawn.

Re:

[Ilgaz]

Low sales of Apple PPC or entire Power family doesn't mean they are "bad" processors, it means people, ordinary home users can't afford them.

Most of PPC high end processors are used in consoles and those things does 1080p HDTV 3d realtime graphics along with 6 channel minimum positional audio.

I am sure MS Xbox 360 sales and Nintendo already quadrupled total Apple G5 sales. PS 3 coming too and as I said in another story, there are people (including me,if price doesn't change) who will buy PS 3 just for some

Re:

[MemoryDragon]

It was not the top5% IBM has moved away from the 970 core which basically was a reduced Power4 core years ago, their current top of the line and has been for more than two years is the Power5 line (which is not the sames as the G5, it is what could have become the G6) Apple did have no choice than moving away, but it is true, that Neither IBM nor Freescale shed a tear of losing Apple. Their core market is a different one nowadays and will be for the upcoming years. I think Intel has a bigger problem, their

Re:

[MemoryDragon]

Ahem, Cell does not have altivec units, the SMP in the cell is an entirely different beast. Altivec is dead, IBM never was very fond of it, and no the SMP from the cell does not even come remotely from that corner, have you ever heard of signal processors, way older than altivec, an entirely different field of processors.

Re:

[antime]

From "Cell Broadband Engine Architecture", page 39:

4.1.1 Optional Features in PowerPC Architecture, Book I (Required for CBEA) The following facilities and instructions are considered optional in the PowerPC Architecture, but are required for the PPE by the CBEA user mode environment.

• Floating reciprocal estimate single A-form (fres)
• Floating reciprocal square-root estimate A-form (frsqte)
• Vector/SIMD multimedia extension

SIMD also has nothing to do with SMP (symmetrical multi-processing).

Re:

[MemoryDragon]

No, but saying IBM was pushed towards SIMD by altivec is a totally false statement, SIMD even existed in the old cray vector computers of the eighties. In fact IBM considers altivec a legacy they are not very eager to support at all.

Re:

[antime]

I don't know if IBM considers SIMD in general obsolete but I do think they don't consider it the best technology for their traditional markets - mainframes and scientific/technical computing. For Altivec in particular it probably doesn't help that it wasn't created by IBM but rather by and for one fairly small customer. If you know otherwise, I'd be eager to read any references you have.

(It would also have been useful to fully read the post you were replying to - you are of course correct in that the SPEs d

Re:

[Ilgaz]

IBM calls it SIMD/Vector, Apple uses "Altivec", that is confusing people, it always did.

Server admins aren't impressed by "Altivec" like end user impressive terms, they want real spec. That is what IBM does.

Apple even used "Velocity Engine" term. Altivec is an option for the chips, if you decide your server will do lots of vector processing etc, you opt in for it.

Altivec is no where near dead, real coders make use of it. It is Apple's fault (hopefully not intentional) not to come up with something which wil

Re:

[Yonder Way]

In my line of work I am root on a number of pSeries machines, 520's, a number of 570's, and more.

It's a shame that Apple computers are the first impression that most people have of the architecture. Slicing and dicing a pSeries 570 gives a much different impression of just what a box like this can do.

Or forget the LPAR; run Gentoo right on the iron. Then

emerge samba

but don't blink or you might miss it.

Re:

[uarch]

Cell BE availability won't be the cause of any PS3 shortages. IBM's been pumping chips out for quite a while.

Look to blueray as the source of PS3 delays.

Re:Not for consumers...

[Bryansix]

I'm not a major corporation but I could order a server from Dell and they would ship it to me. That's consumer enough for me.

Re:

[ben there...]

Are you sure? I once tried to buy a refurbished PC from their Small Business section and they cancelled my order.

Consumers consume, even high end

[Anonymous Coward]

Servers aren't consumer products so I don't see how this gets consumers any closer to a Cell or any sooner.

Depends on which definition of consumers you are using. Taken more literally, even large companies are "consumers". I think the intent of the submitters statement was simply that it would be the first shipping product that would house the cpu. After all, if you were willing to pay the cost, you _could_ purchase it. No different than say a Ferrari Enzo, it's a consumer product, but one in which o

Good Marketing for the Hardware

[twitter]

Servers aren't consumer products so I don't see how this gets consumers any closer to a Cell or any sooner.

The server that runs Fedora core 5 costs $18,000 or so. Like most people, I'm not going to buy a computer that costs so much but it does get me closer to a Cell processor.

Next to a part of a Los Alamos super computer, a $600 PS3 looks like a bargain. Hell, it looks like a bargain next to a "Ready for Vista Ultimate Sucker Edition" computer. I thought the PS3 was interesting as a gaming platform

Re:

[Ilgaz]

Let me give a funny number. PS3 is $600 and can play Blu Ray movies.

Blu Ray "deck" is $999 :)

I have one friend who bought PS2 just because it has excellent DVD playing capabilities (picture quality etc.). It looks like same guy will buy PS3 for Blu Ray. I think the PS2 uses its CPU capabilities to enhance/correct movie.

If they offer WiFi, IMAP etc, why not? I am sure Linux would run too.

$400 less, you get a state of the art CPU, blu ray, real games (bluray has java too) and multimedia. You can "fold@home" t

Re:

[adisakp]

Embedded systems could really use a processor like this. Like my toaster

Sony plans on developing a toaster with a CELL Processor. It will have DRM and only toast Sony brand bread though which will come in an odd size and cost twice as much as any other bread.

Re:

[flithm]

Sony plans on developing a toaster with a CELL Processor. It will have DRM and only toast Sony brand bread though which will come in an odd size and cost twice as much as any other bread.

Don't forget:

- the ToastMan spys on you at night and reports its findings back to Sony
- ToastMax bread soon to compete with Very White Bread (VWB) for dominant market share. Although appearing identical in every way ToastMax is incompatible with VWB. Somehow it holds 60% more peanut butter than VWB, but interestingly init

Re:Does it run

[LnxAddct]

I think you were just being a dumbass, but in case you weren't... according to this article [xbitlabs.com] about the same announcement, IBM is recommending Fedora Core as the operating system to use. So yes, linux does run on it.
Regards,
Steve

Re:

[cmacb]

I think they are similar, but not binary nor instruction set compatible, just "similar" as IBM, Motorola, and Apple colaborated on the design, but I agree that the Intel had a nasty design for assembler programmers. I also think Apple has missed a chance to shine in the server market by focusing exclusively on the lowest common denominator desktop systems (and treating servers as just rack-mounted variations on same).

There are a couple of companies building small and inexpensive desktop systems based on th

Re:

[Surt]

PowerPC was not really evolutionary from 68k. It was a pretty complete redesign, though the early chips had the capacity to run both instruction sets to help ease the transition. Once all the interesting software was converted to PowerPC, the backward compatibility with 68k was dropped.

Re:

[Ilgaz]

As Ex Amiga owner I can say 68k has nothing to do with Power architecture. One is CISC (68k) and other is RISC from beginning.

Motorola joined IBM producing RISC CPU instead of upgrading their 68k line.

I can understand your confusion since Apple and Amiga (via extension board) moved to PPC 603 from Motorola 68k.

I am not a programmer but RISC chips are famous for their reduced registers and commands, to run instructions in less cycles.

Better give you pointer written by real programmers:
http://en.wikipedia.org [wikipedia.org]

Re:

[Tet]

If I am getting my history correct, isn't this chip basically the evolution of the motorola 68000?

No. It can trace some lineage back to the m88k, though[1], which was an even cleaner design than the 68k. If pushed to pick my favourite CPU of all time, I'd probably have to go for the m88k. It was an absolute joy to work with.

[1] As well as to the IBM POWER chip, of course.

Re:Mac OSX?

[geekoid]

"Anyone else expect to see a Mac OSX Xserve running these babies in 2008?"

nope.

Re:

[rootmonkey]

MacOS wouldn't have any problem running on the PPU portion of the Cell. It will take alot of effort to get the OS to effectively utilize the SPU's though. The PPU is awfully slow and has very poor memory access rates. So yes MacOS could run on it, but you wouldn't want it to. Since the PPU is powerPC it wasn't a huge effort to get Linux running on the Cell. The added work was to expose the SPUs through the SPU file system. To the kernel the SPUs just looks like a block device.

Re:

[CarpetShark]

I think the real message here is, if you don't buy the PS3, the terrorists win.

Quite true, if by "Terrorists" you mean microsoft ;)