Cell Programming Examples presentation webcast going on now

slide17jw.jpg


slide26cm.jpg


slide31bv.jpg


slide47dq.jpg


slide53ow.jpg


slide66pv.jpg


slide74jf.jpg


slide86yw.jpg


slide99kx.jpg


slide105xj.jpg


slide118al.jpg


slide150on.jpg


slide131mp.jpg


slide145rz.jpg
 
Yay, they asked my question (IBM libraries on PS3 linux distros)..but he obviously didn't read the latest Kutaragi interview ;)
 
This lack of audio thing renders it all useless for me.

Is there any indication of when the appropriate documentation will explode onto the net? Is it at the end of the presentation?
 
xbdestroya said:
This lack of audio thing reners it all useless for me. :(

Is there any indication of when the appropriate documentation will explode onto the net? Is it at the end of the presentation?

This was asked, but crap it, I can't remember the answer. Was typing at the time. He said the whole presentation should still be on the web after anyway, so maybe we can go over it again..
 
Anyone have any good Cell questions? They're taking them still..

Doh, nevermind, it's over. I'll get working on the rest of the slides.
 
Titanio said:
xbdestroya said:
This lack of audio thing reners it all useless for me. :(

Is there any indication of when the appropriate documentation will explode onto the net? Is it at the end of the presentation?

This was asked, but crap it, I can't remember the answer. Was typing at the time. He said the whole presentation should still be on the web after anyway, so maybe we can go over it again..

He stated that the documentation would be available "Soon" - when pushed further he clarified this as "Over the Summer".

V.
 
jvd said:
slide17jw.jpg



Does this mean we will need to download new exes for ps2 and ps1 games :?:

Considering the fact that the slide references "Porting legacy code" rather than "Running legacy code" I'd say that it is entirely unrelated to emulation.

Nice try though ;)

V.
 
I've added the rest of the slides including the FFT stuff and performance comparisons. Some of it may require explanation :p
 
slide118al.jpg

That's the slide that is most tell-tale to me as far as practical performance in game-land. Most of the tasks we'd send to individual SPEs would be pretty small-scale micro-tasks like that. And it pretty much shows that in order to get some real performance out of CELL, you're basically going to have to hand-code it.
 
ShootMyMonkey said:
slide118al.jpg

That's the slide that is most tell-tale to me as far as practical performance in game-land. Most of the tasks we'd send to individual SPEs would be pretty small-scale micro-tasks like that. And it pretty much shows that in order to get some real performance out of CELL, you're basically going to have to hand-code it.

What do you mean by handcoded? Any code obviously has to be written by the programmer. The programmer has manual control over all the SPEs, obviously.

Someone actually brought up a point that Cell was designed for games and so forth, but if they were making a chip for more common general tasks, would they reduce the number of SPEs and increase local storage, or what design decisions they'd make differently. He said it was hard to tell, performance always differs from application to application on every chip. He also mentioned that beyond the set of "ideal" workloads for SPEs, others can be "twisted" to get high performance out of SPEs.

edit - also, the webcast is going to be archived and will be available from here later tomorrow if anyone wants to listen again or pick up stuff they missed: http://www.gginternet1.co.uk/powerorginbarcelona/GGLoadPage.asp
 
Back
Top