Support NeoGAF

[gofreak]

This popped up on the Rambus website, from their 2005 Developer Conference last week:

http://www.rambus.co.jp/events/Main1_2_SCE_Suzuoki.pdf

Basically quite a detailed technical analysis of the design choices behind Cell - difficult enough to understand in places even if you were in the audience, I'm sure, and even harder just reading the slides, but if anyone can distill some of the finer detail, feel free..

Also, perhaps more interesting for many here, it included a new demonstration, as mentioned in the PDF. Had a demonstration movie, and then this accompanying slide:

Benchmark Profile
•1 SPU
•Frame Buffer
–Resolution1024x1024
–ColorRGB=8:8:8bit
•Data Software Cache
–Texture Tile16x16x4RGBA (single float x 4)
–Frame Tile16x16x4 RGBA (single float x 4)
–Vertexdynamic double buffer (size varies)
•Operation
–One overlay per object
•Newton Equation
•Collision
•Translation & Lighting
•Pseudo drop shadow (no self-shadowing)
•Bilinear texturing
•Rasterization

Basically some software rendering and physics on one SPE. I doubt the individual components of this demo could be too complex given it appears it was constrained to one SPE, and it was doing so much stuff on its own, but I guess they were intending to make a point Unfortunately there's no more detail on the demo out there yet as far as I can see, specifically performance detail - would be really nice to have. No one here happened to be in attendance?

 

[Straightballin]

Basically some software rendering and physics on one SPE.

Were they medicore physics or full blown because if one SPE can handle software rendering and Full blown physics thats pretty amazing.

 

[gofreak]

Were they medicore physics or full blown because if one SPE can handle software rendering and Full blown physics thats pretty amazing.

Like I said, it seemed to be an "all-in-one" demo, so I doubt there was any big focus on one specific area. In fact I'd perhaps wager to the "unaided" eye, so to speak, it may not look so impressive (if this was on just one SPE). I think as much as the demo itself may be nice it's even nicer to emphasise how a SPE can also juggle many tasks at once...there seems to be somewhat the impression that you go and "dedicate" SPEs to specific tasks and write them off for use with others, but as something like this demonstrates it can be switched between lots of different things for a single frame, so to speak.

There's been continuous, positive, suggestion about Cell's strength with physics though.

 

[Phoenix]

It is interesting to see their thought process 'after the fact' since they based a lot of decisions, apparently, on writing actual applications and designing around the bottlenecks as opposed to just making something that looks good on paper - or so that is what they appear to be saying based on many of their performance graphs.