Support NeoGAF

[McFly]

Looks like Cell is already at version 3:

# Comment by Barry Minor — December 2, 2005 @ 12:21 pm

Juice,

I used a UP Cell bringup system.
3.2 GHz DD3.1 Cell processor with 8 good SPEs, 512 MB XDR Memory, 100 Mb network.
Pixels were 128 bit, 32 bit float per color channel.
All rendering parameters were the defaults set by the Cg program with the exception of the window size which was increased to 1024×1024.

Posted by one on B3D --> http://www.beyond3d.com/forum/showpost.php?p=641649&postcount=15

Would be interesting to see a picture and some specs of this new cell. Maybe this is the one with a 1:2 dp flop performance. Anyone with more informations about this?

Fredi

 

[gofreak]

I doubt it's the one with improved DP performance. They were only talking about working on that a short time ago, I doubt they'd already have silicon available like this.

Mystery! I wonder what's going into PS3.

 

[BlueTsunami]

Mystery! I wonder what's going into PS3.

Thats what i'm wondering to. I didn't even know what their was going to be another revision to Cell, besides DD2.1.

 

[Eric_S]

IIRC it was the PPE that was at version 2, no idea what the Cell was at though. But lets assume that the blogger is right for a moment.

So what could have changed?

We've heard rumors that the intiger pipelines on the SPE's have doubled in preformance, wich, plus some minor powermanagement tweaking, could very well account for the change in version numering (assuming that this isn't some missunderstanding or missprint).

Other possible things that could have happended are:

increased cache sizes (unlikely unless they've managed to layer transistors... wich in turn is about as likely as pigs flying in the now icy wastelands in hell),

PPE tweak (not shure what could be done now except a minor power management tweak or the like, anything else would propably mean a big redesign or swiching to another core. And as much as I'd like to see the PA6T in there, it aint gonna happen),

Upgraded memor subsystem (Don't know really, I suppose that it could be relatively "easy" to swich to a XDR2 memory controller as that part of the design looks pretty modular, a pincount increase is very unlikely due to the incresed production costs it brings with it).

 

[gofreak]

We've heard rumors that the intiger pipelines on the SPE's have doubled in preformance

We have?

I think PPE tweaking is probably the most likely explanation. That seemed to be the biggest change from 1 to 2, so further tweaking could account for 3.

 

[Eric_S]

Bleh, missread the old GAF thread about that one. Apparently it was about how the Cell could do double the amount of intiger operations as opposed to floating point operations. :-/

Skimming though threads == bad

 

[McFly]

This is where the original comment was posted: http://gametomorrow.com/blog/index.php/2005/11/30/gpus-vs-cell/

Fredi

 

[gofreak]

Yeah, the guy is an IBM-er, so it's not BS or anything.

Also interesting was his claim that they'd used Cell to render in realtime at 1080p for the TRE demo. Although on second thought I guess that makes sense (if they were getting >= 30fps at 720p, at least 15fps should be possible at 1080p, which would just about constitute realtime).