Presentation of a 6GHz second gen Cell Processor at ISSCC

[Farid]

http://news.com.com/2100-1006_3-6146309.html

The first-generation Cell Broadband Engine chip, co-developed by IBM, Sony, and Toshiba, has just appeared in Sony's PlayStation 3 game console and can run at 4GHz. The second-generation chip will run at 6GHz, according to the ISSCC program. In addition, the new chip will have a dual power supply that increases memory performance--a major bottleneck in computer designs today.

It's not entirely, if at all, related to consoles, but that's the forum where people discuss Cell topics, so the thread is better in here than in the hardware forum.

 

[London Geezer]

http://news.com.com/2100-1006_3-6146309.html


It's not entirely, if at all, related to consoles, but that's the forum where people discuss Cell topics, so the thread is better in here than in the hardware forum.

Cool... I didn't understand how the lack of a dual power supply is a major bottleneck on current CPUs though? Anyone with the time to explain?

 

[Bohdy]

Will it include the DP modifications, or is that not for the "mainstream" models of Cell?

 

[Shifty Geezer]

I doubt this is a 'mainstream' Cell processor as the most mainstream applications would benefit from a smaller, cheaper CPU. I guess this is more for highend, and DP will improved. In fact, DP will probably be improved in all future iterations as, IIRC, it's not a substantial extra cost for substantial gains.

 

[Jawed]

Ooh, that's interesting, making 100 GFLOPs DP rather easier to achieve (perhaps it'll end up 200GFLOPs DP). Haven't read this yet, is this a 65nm version?

Jawed

 

[ADEX]

Cool... I didn't understand how the lack of a dual power supply is a major bottleneck on current CPUs though? Anyone with the time to explain?

I don't know but I think it's something to do with the memory circuits demanding a certain voltage and everything else thus has to sit at that voltage. A lot if IBM 90nm stuff seems to run at 1.1v or higher and I wonder if this is why.

If you can put memories on a different voltage the logic might then be able to run at a lower or higher voltage allowing allowing lower power and or higher performance.

Could be something completely different though.

--

As for the DP version it isn't due until 2008. It'll be going into supercomputers and the like so it'll probably not be cheap. They'll need a new SPE design for it though as it's going to have full speed DP (i.e. 100 GFLOPs).

They can alter the existing Cell to enhance the DP performance without making huge changes so I'm wondering if they have done this with the 65nm version. In the article linked the other day the Cell in the future PS3 has a 6.4GHz memory bus, if they keep the bus full width that'd double memory bandwidth.

 

[one]

Why is it featured now?
http://www.beyond3d.com/forum/showpost.php?p=886865&postcount=57

 

[Carl B]

It's good to see that Cell is enjoying a revision schedule seemingly no less aggressive than that normally seen on the desktop associated with the 'big' architectures. There have been some reports that Cell products have been selling very well for IBM, and I wonder if this is in any way reflective of that.

Anyway, I'll be interested to see when Toshiba and Sony begin their CE, low-power push. Maybe CES will hold some clues.

 

[Capeta]

The first-generation Cell Broadband Engine chip, co-developed by IBM, Sony, and Toshiba, has just appeared in Sony's PlayStation 3 game console and can run at 4GHz.

Why is it only running at 3.2GHz then? Are they waiting to unlock a secret clock generator? Are there any 4GHz CELLs that are in actual use out in the wild?

 

[Shifty Geezer]

I think they mean the Cell chip can be clocked to 4 GHz, and not the the Cell in PS3 can be. The intial announcement of Cell had it clocked at 5 GHz. That'll probably be a select few Cell chips, making them very costly. For PS3 and Blades, a lower clocked varient is being used for economies I think. A 4 GHz flavour may appear at 65nm, but Ps3 will remain at 3.2 GHz.

 

[3dilettante]

Why is it only running at 3.2GHz then? Are they waiting to unlock a secret clock generator? Are there any 4GHz CELLs that are in actual use out in the wild?

The CELL in the PS3 has a lower power consumption ceiling. The higher clock speeds come with higher power output.

It could scale to over 4Ghz, but the power output would be more than double what it is now.

 

[Megadrive1988]

I would concider a DP-oriented CELL at 6 GHz to be a 1.5 generation CELL processor not a true 2nd generation CELL. but opinions can differ.

2nd generation CELL, IMO would be something that appears around 2010, with TFLOP performance, 32 SPEs, etc.

 

[Titanio]

BusinessWeek mentions in its IBM CES report that STI are expected to unveil a new version of Cell that is 'about' twice as fast as the original next month:

http://www.businessweek.com/technol...6.htm?chan=top+news_top+news+index_technology

I'd be somewhat suprised if that actually does refer to clockspeed, and we see a 6Ghz Cell product formally unveiled then. Talking about 6Ghz processors at ISSCC is one thing (in the same way they talked about 4Ghz Cell processors previously there while the fastest actual product we've seen to date has been ~3Ghz).

I do hope that is true, though. Whilst the road of increasing clockspeeds looks murky, I hope we can get more out of that path too going forward rather than solely relying on parallelisation, and this would be a nice little jump in the given period.

 

[Carl B]

Just the way that article is worded, I definitely expect that it is refering to the 6GHz Cell announcement at ICCSS and nothing more.

 

[Shifty Geezer]

Just, please, someone stick the damned thing in a TV and make the original worthwhile before worrying about next-generation chips!

 

[Carl B]

Shifty you know the thing's got to get to 65nm before it's going into a TV. (unless we're using otherwise 'wasted' 90nm chips, the expense would just be too great)

Sony and Toshiba supposedly have bulk process Cell plans on 65nm, so let's see what happens. Maybe Archie will chime in on this thread with some insights if he has them.

 

[Blazkowicz]

2nd generation CELL, IMO would be something that appears around 2010, with TFLOP performance, 32 SPEs, etc.

and have at least one OOO general purpose cores.

 

[Shifty Geezer]

Shifty you know the thing's got to get to 65nm before it's going into a TV. (unless we're using otherwise 'wasted' 90nm chips, the expense would just be too great)

Sony and Toshiba supposedly have bulk process Cell plans on 65nm, so let's see what happens. Maybe Archie will chime in on this thread with some insights if he has them.

If that's true, why did they say they'd have Cell TVs out in 2006? Were they expecting 65nm to be out earlier? What's bothering isn't just the lack of Cell sets, but any real showcased prototypes. At least other speculative technologies (SED TVs) get shown every year! Wouldn't they have a Cell TV prototypr around by now showing wonderful Cellness and just waiting for a 65nm shrink to go into mass production? Why not sell that at stupid money as a premium range device?

 

[Gubbi]

If that's true, why did they say they'd have Cell TVs out in 2006?

That was january 2005.

Today Broadcom offers HDTV-on-a-chip for $30. That includes almost all you need to receive, decode and display a HDTV stream (including full HD-DVD/BluRay support).

So while Toshiba might want to be leading edge technologically they also want to turn a profit, and here cutting cost on the BOM is essential.

Cheers

 

[popper]

That was january 2005.

Today Broadcom offers HDTV-on-a-chip for $30. That includes almost all you need to receive, decode and display a HDTV stream (including full HD-DVD/BluRay support).

So while Toshiba might want to be leading edge technologically they also want to turn a profit, and here cutting cost on the BOM is essential.

Cheers

when you say "That includes almost all" is that real HD though as in UK BBC mbaff Mpeg4-AVC/H.264 or if you must the VC-1 (Divx with bells on) or just old hat US Mpeg2 HD? or is that part covered in another sister chip.

i thought all the cell cores that failed the PS3 spu test would be the ones found in your new HD tv set, its not as if you need all the spe's for decoding multiple AVC streams or other fancy GUI effects you might find in new TV firmware...