Details trickle out on CELL processor...

[Brimstone]

When did we start listening to /. anyways? "Data Plane Processors" -- has anyone heard this? I surely haven't. Their thread on Cell is a riot BTW.

I just did a quick search on google and found this of instrest.

Data Plane Processing with Configurable Architectures

It read a lot like the "CELL" concept but more limted because ARM isn't a desktop/workstation CPU. I really just glanced over this. Maybe this guy is telling the truth.

 

[McFly]

Their thread on Cell is a riot BTW.

Not only the Cell thread, /. has became a riot in much every thread lately, but there's mostly one or two people that really know something.

Fredi

 

[j^aws]

Ok, back to the topic.

From: http://games.slashdot.org/comments.pl?sid=131807&cid=11006251

Actually, CELL is based around the 970. Expect about 80-90% performance compared to an equivalently clocked 970. Where it goes nuts is that there's a number of vector units attached that are basically "VMX on steroids" to quote one of the main guys at IBM behind this. The vector units (or Data Plane Processors as they're calling them) can also communicate between each other as well as with the central core. The workstations are actually headless server blades, each of which will have 2 CELL's on them and they'll be running Linux.

This stuff isn't bullshit, it was all disclosed Thursday at the Australian Game Developers Conference. I didn't sign a NDA so it's all good. I also fondled a PSP =]

Fredi

How come he has all this info but no one asked the obvious question, how many freaking FLOPS has it!?

 

[Spidermate]

@spidermate..Is the TV the new PC?

LAS VEGAS - The high-definition TV will be the future of the next-generation broadband-networked entertainment hub, predicted Sony President and Chief Operating Officer Kunitake Ando in his Thursday morning Consumer Electronics Show (CES) 2003 keynote.

In the narrowband age, the PC was the only gateway to the network, Ando told attendees. But in the new digital age, the TV is being reborn as the center of broadband entertainment, he said.

Going forward, PCs, like Sony Vaios; game consoles like PlayStation 2; and consumer devices, from DVD camcorders, to handheld Sony Clies, all will link up to smart TVs, like Sony's Wegas. Users will be able to use large-screen TVs to display all forms of media, which they can select with the flick of a finger, in Sony's world view.

Source: extremetech.com

I believe that'll be the Cell base television I was talking about earlier.

 

[j^aws]

Some random thoughts...

How high do PPC 970'c clock?
Could the PUs be clocked at half the S|APUs like one of the patents?
Could a fat PU like the PPC 970 imply scalar ops are done there and not on the S|APUs?

 

[V3]

anyways? "Data Plane Processors" -- has anyone heard this? I surely haven't.

Its call Broadband Engine, so data plane works should be given.

 

[j^aws]

anyways? "Data Plane Processors" -- has anyone heard this? I surely haven't.

Its call Broadband Engine, so data plane works should be given.

I don't get this...are you saying they used the term purely for an analogy to a programmable DSP processor?

 

[passerby]

This stuff isn't bullshit, it was all disclosed Thursday at the Australian Game Developers Conference. I didn't sign a NDA so it's all good. I also fondled a PSP =]

If this is true, I think news and titbits will be flying through the developer community now? Anyone heard anything to verify this?

Anyway, I did take a look at the AGDC site. Here's their schedule:
http://www.agdc.com.au/conference/schedule_thu.php

2 subjects caught my eye:
Keynote
Jason Rubin, ex - President, Naughty Dog, Inc.
Fear: An Appropriate Response to the Business of Next Generation Game Development

Platforms
Randy Moulic , Research Staff Member and Senior Manager, IBM T.J. Watson Research Center Super Computing "Cell" Converges Action Movies into Gaming Worlds

 

[j^aws]

Well, if they used the PPC ISA, it would be nice for APPLE and OSX!

 

[V3]

I don't get this...are you saying they used the term purely for an analogy to a programmable DSP processor?

To name it for the function it does, its a pretty good name, don't you think ? You can call DSP a sound processor too, if it does the sound processing.

 

[j^aws]

I don't get this...are you saying they used the term purely for an analogy to a programmable DSP processor?

To name it for the function it does, its a pretty good name, don't you think ? You can call DSP a sound processor too, if it does the sound processing.

Well, that would seem consistent with the other analogy of VMX on steroids for the S|APUs then...

 

[V3]

plus from that patent

[0014] In accordance with this modular structure, the number of PEs employed by a member of the network is based upon the processing power required by that member. For example, a server may employ four PEs, a workstation may employ two PEs and a PDA may employ one PE. The number of APUs of a PE assigned to processing a particular software cell depends upon the complexity and magnitude of the programs and data within the cell.

They are probably not making it up. Though might not bet telling the whole truth either.

 

[j^aws]

plus from that patent

[0014] In accordance with this modular structure, the number of PEs employed by a member of the network is based upon the processing power required by that member. For example, a server may employ four PEs, a workstation may employ two PEs and a PDA may employ one PE. The number of APUs of a PE assigned to processing a particular software cell depends upon the complexity and magnitude of the programs and data within the cell.

They are probably not making it up. Though might not bet telling the whole truth either.

I read the two CELL comment from that guy as per blade per workstation...hmmm...

EDIT: They would need 4 blades to match 'The Registers' claim of 2TFlop per prototype WS...

 

[one]

16TFlops is the official number for 1 rack.

1 rack = 42U dualprocessor modules = Cell (1PE@90nm) processor * 2 * 42 = 84 Cell processors = 16TFlops

1 Cell (1PE@90nm) = 190GFlops

Broadband Engine (4PE@65nm) >= 760GFlops

 

[j^aws]

16TFlops is the official number for 1 rack.

1 rack = 42U dualprocessor modules = Cell (1PE@90nm) processor * 2 * 42 = 84 Cell processors = 16TFlops

1 Cell (1PE@90nm) = 190GFlops

Broadband Engine (4PE@65nm) >= 760GFlops

Huh...I'm getting 256 GFlops per CELL (1 PE with 8 S|APUs) at 4GHz. Where did you get 190 GFlops?

 

[Jov]

Maybe because it's cheaper for Sony to have one cheap cell chip that can be used for many different tasks by writing software for it, than have many, many different specialized chips that have to be redesigned for every new task.

Fredi

And that's why I said: SONY is trying to to use these EE/GS to save money by not buying other companies ASICs or designing their own.

Isn't the EE/GS designed by Sony? If Sony can save money using the EE/GS in other devices why do they need other companies ASICs or reinvent the wheel again?

How do you know its an overkill?

 

[j^aws]

16TFlops is the official number for 1 rack.

1 rack = 42U dualprocessor modules = Cell (1PE@90nm) processor * 2 * 42 = 84 Cell processors = 16TFlops

1 Cell (1PE@90nm) = 190GFlops

Broadband Engine (4PE@65nm) >= 760GFlops

Huh...I'm getting 256 GFlops per CELL (1 PE with 8 S|APUs) at 4GHz. Where did you get 190 GFlops?

Right...you've assumed all 42U used for 84 CELLS...I don't think it will work like that if it comes in a rack. They'll prolly have other systems, i/o, cooling etc filling up the extra space.

 

[one]

16TFlops is the official number for 1 rack.

1 rack = 42U dualprocessor modules = Cell (1PE@90nm) processor * 2 * 42 = 84 Cell processors = 16TFlops

1 Cell (1PE@90nm) = 190GFlops

Broadband Engine (4PE@65nm) >= 760GFlops

Huh...I'm getting 256 GFlops per CELL (1 PE with 8 S|APUs) at 4GHz. Where did you get 190 GFlops?

Right...you've assumed all 42U used for 84 CELLS...I don't think it will work like that if it comes in a rack. They'll prolly have other systems, i/o, cooling etc filling up the extra space.

Really? I think those servers from IBM aren't filled with those extra things other than self-contained blades themselves. Since it's not clear at this stage that 16Tflops number is theoretical or real, each Cell may have higher processing power when not interconnected.

BTW reading /. I noticed the other guy than the person quoted above, attended AGDC

Moulic from IBM stated in his presentation at the AGDC that the Cell processor was to ship with a full SDK that contained solutions to much of the management aspects associated with writing and running parallel processes.

You can also take the "bull by the horns" and control the hardware directly if need be.

...

The primary goal of IBM with regards to the Cell processor was that it be scalable first as a rack mounted solution. Therefore the Cell Processor Workstation (CPW) will be available first up as individual form factor boards, containing a CPP, several DPPs and other small components for I/O etc.

...

Moulic confirmed in his talk at the AGDC that the based OS shipping with the first Cell Processor Architecture will be Linux.

...

I was there also. They did not disclose the number of DPPs on each unit.

 

[j^aws]

16TFlops is the official number for 1 rack.

1 rack = 42U dualprocessor modules = Cell (1PE@90nm) processor * 2 * 42 = 84 Cell processors = 16TFlops

1 Cell (1PE@90nm) = 190GFlops

Broadband Engine (4PE@65nm) >= 760GFlops

Huh...I'm getting 256 GFlops per CELL (1 PE with 8 S|APUs) at 4GHz. Where did you get 190 GFlops?

Right...you've assumed all 42U used for 84 CELLS...I don't think it will work like that if it comes in a rack. They'll prolly have other systems, i/o, cooling etc filling up the extra space.

Really? I think those servers from IBM aren't filled with those extra things other than self-contained blades themselves. Since it's not clear at this stage that 16Tflops number is theoretical or real, each Cell may have higher processing power when not interconnected.

BTW reading /. I noticed the other guy than the person quoted above, attended AGDC

Moulic from IBM stated in his presentation at the AGDC that the Cell processor was to ship with a full SDK that contained solutions to much of the management aspects associated with writing and running parallel processes.

You can also take the "bull by the horns" and control the hardware directly if need be.

...

The primary goal of IBM with regards to the Cell processor was that it be scalable first as a rack mounted solution. Therefore the Cell Processor Workstation (CPW) will be available first up as individual form factor boards, containing a CPP, several DPPs and other small components for I/O etc.

...

Moulic confirmed in his talk at the AGDC that the based OS shipping with the first Cell Processor Architecture will be Linux.

...

I was there also. They did not disclose the number of DPPs on each unit.

Of course they would be self contained with a base spec. I'm thinking the extra space for more disk arrays, UPS etc to fill up the space. The main reason really that 190 GFLOPS is way less than 294.4 GFlops @ 4.6 GHz per Cell. Why down clock that much was my main reason when you've spent billions reaching that high clock!

 

[one]

As for 2Tflops(?) workstation reported in hands of developers, you can assume it's different from IBM/Sony 'Cell Processor Based Workstations' which are available as products for non-gamedev-licensee users.

In this presentation picture from E3 2004, you see 'Cell Based Game Development System' (possibly available from Sony rather than from IBM) beside the workstation.