End of Cell for IBM

[patsu]

Absolutely, I'm not disputing that at all. I've stated all along that CELL makes sense in a console or for HPC.

I was merely pointing out why CELL has problems breaking out of those niches, and therefore enjoys less than stellar commercial succes (the context of this thread).

Well... that is a different debate altogether since it involves economics, marketing, finance, business goals, and so many other factors.

Even regular CPUs (without SPUs) fail to take on the general purpose CPU market. The SPUs may not be the bane of the Cell architecture.

 

[Gubbi]

Isn't this more a problem of OS design than Cell's design? Take your Windows 3.1 example. Cooperative multitasking is ancient history, right? But if we had Cell back then, 8 hardware cores would mean 8 active processes and no need to task switch given the workloads of the period.

Are you wearing rose tinted glasses or have you never experienced the pain of Win 3.10 ?

When you listen to music streamed over http, you network driver calculate checksums for each packet, your anti-virus software scans each packet, your WinAmp decodes each packets, then it pushes it to the Windows audio layer, which ultimately pushes it to the audio device driver. Notice that the these actions are entirely sequential in nature, allocating a core for each step would be an unsightly waste.

The idea of pinning a OS process to a single core just ends up idling most of your silicon, - and would result in lower performance, think rar-ing large amounts of data for backing up, applying filters in photoshop and compiling large software projects; You want to use every available resource for this.

Cheers

 

[Shifty Geezer]

When you listen to music streamed over http, you network driver calculate checksums for each packet, your anti-virus software scans each packet, your WinAmp decodes each packets, then it pushes it to the Windows audio layer, which ultimately pushes it to the audio device driver. Notice that the these actions are entirely sequential in nature, allocating a core for each step would be an unsightly waste.

In a Cell computer with idealised OS, the audio data comes into a SPU, the checksum is calculated, a seperate core can run security on all data passing over the ringbus if you really wanted, and a 3rd runs the system audio blending all audio channels into an audio stream.

The idea of pinning a OS process to a single core just ends up idling most of your silicon, - and would result in lower performance, think rar-ing large amounts of data for backing up, applying filters in photoshop and compiling large software projects; You want to use every available resource for this.

Perhaps it's due to this matter that I disagree with that explains why I feel Cell etc. can be successful PC processors. My processor is 95% idle 95% of the time. Compressing a huge amount of data will be limited more by storage speed then processing speed. A Cell core could handle this with ease. Applying filters in Photoshop will be realtime on a Cell optimised platform on a couple of SPEs in most cases, and anything more significant will still only be a matter of seconds. And how many people compile large projects at home?

So sure, it wouldn't be optimal, and if running a task like an encode, perhaps a good percentage of cycles will be wasted as they're reserved for other systems. It'll still be a very viable platform though, comfortable to use and very fast in the heavy workloads. At the end of the day, a Cell PC running a Cell-targetted OS would feel every bit as responsive and flexible as an x86 Windows PC and still be able to number crunch supremely well.

 

[JardeL]

Another rumour related to this ;

We can officially reveal in this world exclusive that SCEI has officially chosen IBM’s currently in development POWER7 architecture for it’s PlayStation 4 system, currently scheduled for a 2012 worldwide release.

IBM shall debut POWER7 for the server market in the summer of 2010. The PlayStation 4 shall use a cost effective version of the architecture custom designed for Sony’s specific needs.

The only information FGNOnline currently has available regarding specifications and performance is that the PS4 implementation of the chip shall use 6-8 cores, 24-32MB shared L3 Cache, Quad threading per core, and a double precision performance approaching 200GFLOPS.

The CELL chip which powers the PS3 is capable of 15 double precision GFLOPS. This would give the PS4 a performance leap of over 10 fold over it’s predecessor.
Interestingly, the primary reason IBM cancelled development on the CELL based PoweXcell 8i is because of SCEI’s change of stance. The company wanted to move away from the exotic architecture employed for the PS3 to a more traditional architecture favored by third party developers.

As displayed by it’s choice of processors, the server based POWER7 and the TBDR based PowerVR 6, SCEI once again has sights set for performance leadership in the next round.

[ http://ps3clan.nl/2009/11/sony-kiest-voor-ibm-power7-cpu-playstation-4/ ]

 

[Carl B]

LOL

 

[deepbrown]

Another rumour related to this ;

[ http://ps3clan.nl/2009/11/sony-kiest-voor-ibm-power7-cpu-playstation-4/ ]

The bit you underlined is untrue...they aren't cancelling the PoweXcell 8i...they're using it now. They are cancelling its successor.

And apparently IBM has come out to deny the original report - stating they will be making new chips in the Cell line...

And this website can't exclusively confirm this CPU...they don't stop reporting rumours.

Plus I'd put the PS4 release in around 2013-14.

 

[Shifty Geezer]

Can the Internet please learn to avoid FGNOnline and other make-believe websites? Any "exclusive reveals" that aren't backed up with...anything whatsoever aren't worth the electrons they're sent with. It's bad enough opinions are being formed on misrepoted sites, without the added embarassment of trying to explain away random burblings.

BTW - A clue to if they're making it up or not. It can't be official if it isn't coming from an officially sanctioned source, which is almost certainly only to be the company's own PR department.

 

[Carl B]

And apparently IBM has come out to deny the original report - stating they will be making new chips in the Cell line...

To my knowledge IBM hasn't come out to deny that at all... or rather, to claim that the architecture will see future variants.

 

[slider]

To my knowledge IBM hasn't come out to deny that at all... or rather, to claim that the architecture will see future variants.

I'd always (naively) assumed that the PS4 would simply involve a relatively simple upgrade of Cell. # of SPUs with an enhanced PPU. Caches and LS and their implications didn't really come into my thinking (probably cos the implications are beyond me!).

I know things are still up in the air on the future of Cell but I can't help but wonder if the PS4, whenever it comes out, will have a simple "upgrade" Larrabee rumours notwithstanding. Could Sony commission something like that at reasonable cost?

Of course I have no idea how simple/complicated a job it is to bump the number of SPUs etc. but given the investment made by Sony's first parties I hold out hope for Cell remaining. Plus the more exotic an architecture the more excitement I have for it... :|

Dammit, I need some HW speculation to get my motor revving. Not the baseless rumours but speculation based on something substantive. Dammit, dammit.

 

[deepbrown]

To my knowledge IBM hasn't come out to deny that at all... or rather, to claim that the architecture will see future variants.

http://www.driverheaven.net/news.php?newsid=344

We managed to get hold of an IBM spokesperson an hour ago and they said that only one CPU development cycle is being 'halted' which is the successor to the current PowerXCell-8i cpu. IBM have said they are planning to work on other CPU's in the Cell Processor 'family'

Not the best source - but neither is Heiss Online in the first place that brought this IBM cancelling Cell thing.

 

[deepbrown]

Can the Internet please learn to avoid FGNOnline and other make-believe websites? Any "exclusive reveals" that aren't backed up with...anything whatsoever aren't worth the electrons they're sent with. It's bad enough opinions are being formed on misrepoted sites, without the added embarassment of trying to explain away random burblings.

BTW - A clue to if they're making it up or not. It can't be official if it isn't coming from an officially sanctioned source, which is almost certainly only to be the company's own PR department.

I know - tis annoying. They'd get the same amount of traffic if they just said it was a rumour.

 

[Carl B]

http://www.driverheaven.net/news.php?newsid=344

Not the best source - but neither is Heiss Online in the first place that brought this IBM cancelling Cell thing.

The difference is that Heiss Online was quoting the head of IBM's deep computing division - and then later got further clarification to boot - and DriverHeaven is simply quoting a PR person on the extent of what has officially been canceled. I've spoken to IBM PR people before, and believe me, anything from them is like nothing at all.

Not only that, but a part of me is worried that Kotaku and DriverHeaven and all these folk spoke to the 'same' PR person... which is to say that only one of them did, and they're all simply quoting each others findings and claiming that they also contacted IBM in tandem.

 

[patsu]

LOL

I don't even know whether the appropriate reaction is to COL or LOL.

 

[Weaste]

The double precision part is making my sides hurt for sure. 200GFlops of double precision power.

 

[Arwin]

Another rumour related to this ;

[ http://ps3clan.nl/2009/11/sony-kiest-voor-ibm-power7-cpu-playstation-4/ ]

Yah, the same source for this rumor I presume, lol!

“The PlayStation 4 shall use a high end variant of the Series 6 line. Performance, specifications and features are at this time unknown. The Series 6 shall receive an official announcement from IMGTEC sometime in 2010, with initial models targeting the smartphone and netbook sector.
IMGTEC’s PowerVR technology uses an advance technique called TBDR which can outperform a competing IMR product from nVidia/ATi by 3-5 fold whilst maintaining equal die size and price point. TBDR was the primary reason the SEGA DreamCast was capable of such astonishing graphical feats as early as 1998.
Interestingly SCEI has also chosen IMGTEC as the graphics provider for their next generation PSP. That particular product shall however be using the Series 5XT.”

 

[Florin]

http://www.driverheaven.net/news.php?newsid=344



Not the best source - but neither is Heiss Online in the first place that brought this IBM cancelling Cell thing.

Generally, Heise is first class reporting.

 

[Lazy8s]

Eventually, the winning designs in the mobile sector -- META someday -- will spread out to supercomputing, whereas mini-Cell never would've stood a chance in the mobile space.

 

[draconian]

Wow, if this rumor is true...

Does this mean that the ps4 will most likely not be backwards compatible with the ps3?

Because of diminishing returns of graphics. This gen's graphics will, I think, compete well with even next gen's graphics.

And because of this Sony pretty much lost next gen. This generation's game will have long legs and people will still want to play uncharted or alan wake on next gen systemts.

 

[semitope]

You should read the rest of the thread. All i am seeing is that the 32spe 4PPE cell is no longer going to be made, not that cell is completely off for IBM. I would like to see a form of cell in the future ps. Seeing the things they are doing with the current 8 spe cell it would be great to see something like a 4+Ghz PPE 5Ghz spe x 64 cell processor in the future ps. :smile: (are varying clock speeds possible?)

I believe sony would be quite secure with improved cell variants for at least 2 more playstation generations with ati dx11 graphics as GPU.

There is mention here of NLM and whatever other GPGPU but why would sony need to take yet another risk with their playstation line? Its time to mature what they have built and I dont see the need to integrate everything yet. The cell can already do some work for the gpu and dx11 gpus can be used to do some tasks outside of graphics so I am thinking work could be done to build an archictecture where even though both are separate they could be made to work in a system in which either could help the other (without bandwidth limitations and with low latency memory access both ways). Basically a system where the GPU can help the CPU and the CPU can help the GPU but neither compromises in being able to carry out its own traditional task.

/end fantasy

 

[draconian]

This is what I would like for sony to do.

Keep cell architecture.
go from single core ppe to 3 core ppe. Increase on die cache from 512kb to 1 mb. Keep the spus the same # but increase the sram from 256kb to 512 kb. Maybe even give some spus access to outside ram(I'm dreaming aren't I?).

Then simply increase ram from 512megs to 4 gigs. and get a better backwards compatible gpu from nvidia.

That's it. Maintain backwards compatibility and everyone will be happy.