Super Cell (CELL insider talks to Forbes)

[Npl]

EE and Cell are one and the same only separated by time.

I hope Sony wont place an EE into the PS3 - Quick, someone tell them about the Time-Paradoxon

 

[one]

Now on to some mathmatical conjecture. 32 chips for 16TFLOPs. That is 500GFLOPs per chip. Note that the CELL engineer says "This figure is "probably a p.r. exaggeration," the Cell engineer says" but "future workstations containing racks of 32 chips" will likely reach 16TFLOPs". So it would seem 500GFLOPs CELLS are the future but not the present.

The article also states that the current chip is in the 250GFLOPs range which with the "future" comment and putting the numbers together seems to confirm the first CELLS will be in the 250GFLOPs range.

So my guess is that the first CELLs are going to be 250GLFOPs and that a shrink to 45nm will bring about 500GFLOPs chips that will make up the 32chip 16TFLOPs workstations.

My interpretation of the lines I emphasized is like this;

Since he calls the 16TFLOPS figure a PR, it seems he regards it as a theoretical number calculated by a simple multiplication (x FLOPS * y * z) like you did. He says "(the future workstation) will be able to attain this speed", so instead of a PR, it's safe to assume that he talks about a performance figure rated by some benchmark test. Now, think about an actual supercomputer. BlueGene/L DD2 beta-System (0.7 GHz PowerPC 440) has 32768 processors and achieves Rmax 70720 GFLOPS, when its theoretical peak (Rpeak) is at 91750 GFLOPS. In the case of BlueGene/L, 80% in the theoretical performance can be actually utilized (and the realworld performance is even lower). So I don't think "500GFLOPs (theoretical performance) CELLS are the future". Also, you should note the Cell itself is a supercomputing cluster and the interconnect in the chip matters.

Now onto the second paragraph about the 65nm process. I have no idea why you bring up the 45nm process here, when he suggests the performance increase of Cell brought by a process shrink between 90nm and 65nm. How can they increase the speed? A higher frequency is one way, but in the context of the paragraph, it's unlikely that he talks about the clockspeed of Cell. You have to scale it up by increasing PEs. The first Cell, 1 PE in the 90nm process is likely to have 250GFLOPS theoretical performance. It seems the 65nm process Cell with an approximately 2x die-size can have 2 PEs though you have to control heat and power consumption.

Anyway, with an anonymous insider, this article can be a hoax, but not a bad try I think, especially when this guy dismisses the 16TFLOPS figure

 

[passerby]

The number of new Cell topics/hype per day is getting to the point of being as annoying as DM's anti-Sony trolling back then.

 

[Acert93]

Good points One.

I guess I am reading the 65nm part differently:

One surprise, says the engineer, would be if Cell lives up to the rumor going around the development team that the consortium is on its way to production using advanced 65-nanometer technology, in which transistors are squeezed even closer together than the 90-nanometer production process that Intel uses and that the Cell consortium has claimed to have been using so far. The denser a chip's transistors, the more powerful the chip can be. "For Intel, it would be a big shock," promises the engineer.

I thought the CELL was targeted at 65nm to begin with. If Sony is on track for making 65nm CELL processors for the PS3 by early 2006 what is the point of talking about 90nm CELLs? Since this guy is in the know is there a chance that his comments about CELL are assuming the 65nm process? His "surprise" seems pretty confident.

I think we both are making some assumptions when readin this. Obviously your assumption is that the 250GFLOPs is for a 90nm chip. You may very well be correct. You know a lot more about this than I do. But with the CELL engineers comment about STI being on course for 65nm and with the goal for 65nm CELL ships for the PS3 launch (which I have heard quite a few times here from Sony advocates) I am not sure we can assume 90nm = 1 PE and 65nm = 2 PE. Why not a huge/expensive 90nm = 1 PE with a 65nm 1 PE chip being right around the corner to get an affordable chip?

And you can see where my assumption is. I am assuming that 65nm is the target process for CELL. There is also this flip side: Just like the EE for the PS2, CELL may debut at 90nm and be a big, expensive chip and a 65nm shrink will only be a cost reduction move for mass production. I find this scenario a strong possibility.

Maybe I am missing something (I do not fallow CELL news as closely as some). One, is there any information noting how many PE will fit on a typical 90nm die or is it all conjecture at this point? Btw, you post some pretty good stuff and share a lot of good info. Keep it up!

 

[one]

The number of new Cell topics/hype per day is getting to the point of being as annoying as DM's anti-Sony trolling back then.

Don't worry, in a few days starting Feb 6, we'll be surrounded by the hype matrix... or sobering up to the ISSCC revelation

After that, the real hype Juggernaut will roll out to the PS3 Premiere in March, I promise

 

[Phil]

Riiiiiiight Ken. I think MS and Nintendo need to take lessons from this guy. While this is GREAT stuff for the mainstream press and mainstream consumers to get them excited it really turns me off. CELL is an AWESOME piece of technology and deserves hype and attention. But this is the entire EE thing all over again. While CELL is gonna kick some serious butt, I think the Xenon design is not gonna be some wimp either. And it seems the MS offering from what we know will be pretty flexible and be a very good general computing device. So PS3 is brute power and Xenon looks to be more general purpose. Mark it now: Whoever creates the platform that is the most flexible, accessible, and affordable is going to reach the most developers. I want to know what Sony and MS are doing in this area. I could care less about the PR.

I think you're misguided in the belief that these comments are entirely regarding the PlayStation 3. What I think Ken is refering to is CELLs ability to bring massive computation power through a CELL network (first locally, in the future perhaps even on a WAN scale). PS3 is just the groundwork and not CELLs only purpose.

I thought the CELL was targeted at 65nm to begin with. If Sony is on track for making 65nm CELL processors for the PS3 by early 2006 what is the point of talking about 90nm CELLs? Since this guy is in the know is there a chance that his comments about CELL are assuming the 65nm process? His "surprise" seems pretty confident.

Following various slides by Sony held presentations, I thought it had been clear that 90nm was always the targeted process for prototype CELL based workstations. The PlayStation 3's implementation of CELL (the Broadband Engine) however is speculated to be targeting the 65nm process [as hinted in the Broadband patents and Sony's milestones?]. I think people are getting mixed up with CELL (an architecture) and the Broadband Engine (a CELL implementaiton).

 

[Druga Runda]

the best part will be that in real life, emulating X86 code it will be barely faster than Via Epia

put two of them together and you get a dual Via Epia performance...

now that will be funny...

of course it will fare a lot better when properly progreammed for... but not much use in "real life"

 

[one]

I thought the CELL was targeted at 65nm to begin with. If Sony is on track for making 65nm CELL processors for the PS3 by early 2006 what is the point of talking about 90nm CELLs? Since this guy is in the know is there a chance that his comments about CELL are assuming the 65nm process? His "surprise" seems pretty confident.

Well, all the Cell presentations at the ISSCC 2005 talk about the 1st-generation CELL Processor in the 90nm process. And the paragraph refers to Intel and the 1st generation Cell like

than the 90-nanometer production process that Intel uses and that the Cell consortium has claimed to have been using so far

and it seems Intel is supposed to be 'shocked' at the prospect of the Cell in the 65nm process. Isn't it the most straight interpretation?

The CPU in the PS3 is surely targeted to the 65nm process, judging from the fab plans at the Sony/Toshiba side. But it's very risky to mass-produce CPU with a new architecture in a new process technology. The risk is doubled. So they starts the first production of Cell architecture CPU at IBM fabs, with the relatively older, well-known 90nm process, possiblly for prototype workstations. Currently, the 90nm fabs in Sony/Toshiba are all filled with the production of the chips for PStwo, PSP and PSX and as you know the shortage of PStwo/PSP indicates they have no more capacity in the 90nm lines.

 

[zidane1strife]

and it seems Intel is supposed to be 'shocked' at the prospect of the Cell in the 65nm process. Isn't it the most straight interpretation?

What'd shock intel would be a 45nm implementation, though highly unlikely, if they managed to do that, it would cause quite a commotion

 

[nelg]

Q. How can they increase the speed?

A.

With the power-thrifty Dual Stress Liner (DSL) technology announced in December, Advanced Micro Devices and IBM have taken a big step toward ensuring that Moore's Law applies to more than just the chips that run supercomputers.

LINK

 

[Acert93]

Cell will make possible a transformation in entertainment like that from novels to movies,"says Ken Kutaragi

Ok Phil, explain to me how this quote is true and not just a load of pure B.S. hype?

Btw, broadband market penetration is still to small for mainstream so you alienate more than half of consumers by going through this route. We have to wait for development, deployment, and then adoption of broadband technologies that are "last mile" before we can even start to consider the sharing of information on the scale to do the things he is hyping. Same applies to local area networks with TVs and DVD players with CELL because of market penetration issues. So please explain how this quote is true. How is CELL going to overshadow everything else to such a degree that stuff not using CELL feels like a novel compared to a movie (a horrible comparison because most novels kick movies square between the legs).

I have never seen a good explaination to back this hype up. CELL is going to be powerfull--I have never denied that. But Ken is talking multitudes of degrees and as I have said before that power is not everything--the creative process takes time, money, and knowhow. Easing development, making it quicker, and saving money will have a profound impact on a products quality. So, until you or someone else can explain how such rediculous hype can be taken seriously I will file it away under "Ken Hype" that never comes true. Ball is in Sony's court to demonstrate that what they are saying is true. Until then you can try to demonstrate why we have no reason to file this under the heading of "Meaningless Sony Hype that never comes true but confuses the mainstream consumer". I am all ears.

 

[one]

How is CELL going to overshadow everything else to such a degree that stuff not using CELL feels like a novel compared to a movie (a horrible comparison because most novels kick movies square between the legs).

Kutaragi is an engineer so I assume his Sci-Fi hype always tends to have something (very tiny though) in common with reality. It's an analogy. Convert all of them into bits and bytes, then you get the picture

 

[MfA]

He used to be an engineer, now he is in management.

EE was shipped as presented BTW.

 

[one]

EE was shipped as presented BTW.

You mean, as presented at ISSCC '99 in Feb. 1999? It was shipped with the same architecture, with the same process technology (0.25-micron CMOS process technology with a 0.18-micron gate length) as presented, but shipped with higher clockspeed (250Mhz -> 300Mhz), though it was disclosed that EE had 300Mhz in Mar. 1999 at the PS2 unveiling event.

 

[Acert93]

How is CELL going to overshadow everything else to such a degree that stuff not using CELL feels like a novel compared to a movie (a horrible comparison because most novels kick movies square between the legs).

Kutaragi is an engineer so I assume his Sci-Fi hype always tends to have something (very tiny though) in common with reality. It's an analogy. Convert all of them into bits and bytes, then you get the picture

I think I know what he intended--and Kutaragi has every right to be excited about CELL. But I think technical fields are no place hyperbole. You can quantify the performance and expectations of a chip so exaggerating the "This chip can do X" statements (note that I did not say performance) really misleads consumers.

I hope my cinicism about certain PR comments and hype buildup are not confused for my feelings toward CELL itself. I think the PS3 CELL (and the primary way most of us will get to know CELL over the next 3 or 4 years) is going to be the most powerful CPU in consoles/set top boxes. From a numbers standpoint it will even overshadow the desktops, but I have a feeling that comparing CELL to a general purpose CPU is like apples-to-oranges. In that regards comparing the PS3 CELL to the Xenon CPU is hard because from what we know they appear to be tackling the issues from different angles. The fact CELL may be doing the vertex shading for the PS3 also needs to be factored in when comparing these systems. How they are used is as fundamentally important as to what they can do. CELL looks to be great and my guess is that for sheer power nothing MS or Nintendo does will compare. But total system design is very important, and in that regards system memory is going to be a significant factor in how all this plays out. For that reason alone I am skeptical of most hype.

He used to be an engineer, now he is in management.

EE was shipped as presented BTW.

It is not that EE is different than the release specs, it is the fact the exaggerated hype (i.e. statements about what the EE technology could do with its performance, not the performance itselt) and the details in the small writing that aggrevate people. Ken's quote is basically the same stuff we heard with the EE. EE specs were dead on, what the specs could do was slightly exaggerated. Hype is good for a products success and I recognize that, but that does not mean I have to like it. But when my friends see the PS3 demo shots in their game magazines (I still cannot believe late 20 year olds get those, but they do) and hear all the hype they are not gonna shut up about it until they get their PS3 Sony is very very very good at marketing.

Btw, I have a feeling STI is going to have some great looking stuff for us come next week. I am excited! Man I want to fast forward to E3. Or better yet fall 2006!

 

[Megadrive1988]

Phil wrote:

I think people are getting mixed up with CELL (an architecture) and the Broadband Engine (a CELL implementaiton).

totally agree, Phil. I try my best *not* to get it mixed up, and help others understand this as well.

 

[Sonic]

I just want to make a note in the sense of what Acert93 said. The platform with the most flexibility may not necessarily get the most developers, the platform with the highest number of machines sold will get the most developers.

 

[phat]

EE had lousy performance considering the area it took, the VUs were too limited in both performance and flexibility to justify the size IMO. A spruced up SH-4 would have offered the same performance with easier programming and a smaller die size.

No possible way. Most of the EE's transistors were devoted to FMACs and FDIVs, taking area away from the execution pipeline and instruction decoder, which resulted in a really difficult to program VLIW instruction set. EE was a floating point monster for the time. In order for the SH-4 for be just as powerful and easier to program, it would have to have been larger. You simply need transistors to get work done.

 

[MfA]

All told less than 10% of the original EE was taken up by floating point units.

Probably higher on later models, cause that first EE's implementation must have truely stunk given the subsequent shrinks. Still doubt it even got near 20%.

 

[London Geezer]

Phil wrote:

I think people are getting mixed up with CELL (an architecture) and the Broadband Engine (a CELL implementaiton).

totally agree, Phil. I try my best *not* to get it mixed up, and help others understand this as well.

Yep. Microwave Cell implementation won't be a 1TFlop monster. Unless you need a grill in your oven too, in which case they could push the GHz and go for a BBQ set.