PDA

View Full Version : Detail Layout of PS3 Cell Architecture and On-Chip Communication Interconnect.

Nerve-Damage
12-29-2006, 03:55 AM
Forgive me if this was posted already... :oops:

Some nice tid-bits within this file: PS3 Cell Architecture and On-Chip Communication Interconnect . (http://www-static.cc.gatech.edu/classes/AY2007/cs8803hpc_fall/lectures/22_Cell.ppt)
acousticvan
12-29-2006, 04:44 AM
Forgive me if this was posted already... :oops:

Some nice tid-bits within this file: PS3 Cell Architecture and On-Chip Communication Interconnect . (http://www-static.cc.gatech.edu/classes/AY2007/cs8803hpc_fall/lectures/22_Cell.ppt)

Nerve, I saw this post of yours on beyond3d but I can't reply because I am not their member yet.

The cell has never been said to be out of order execution. It was in-order all along.
Nerve-Damage
12-29-2006, 05:03 AM
Nerve, I saw this post of yours on beyond3d but I can't reply because I am not their member yet.

The cell has never been said to be out of order execution. It was in-order all along.

Yeah, I must have misread this awhile back…

Sometimes I can make mistakes…well lots of them anyhow. :troutslap
lips
12-29-2006, 06:23 AM
:clapping: good read. Low latency lookups and high bandwidth access is definitely welcome.
Siraris
12-29-2006, 07:34 AM
I didn't know that there was that substantial of a difference between the EE and Cell. I'm surprised any developers are complaining about Cell when they had to work on PS2 for the past 7 years. I mean Cell has 512kb of L2, while the PS2 didn't even HAVE L2.

Bleh
Garfunkel
12-29-2006, 08:17 AM
I didn't know that there was that substantial of a difference between the EE and Cell. I'm surprised any developers are complaining about Cell when they had to work on PS2 for the past 7 years. I mean Cell has 512kb of L2, while the PS2 didn't even HAVE L2.

Bleh

they are similar in many ways but development for the processors has many differences, my guess is that if you have experience with the EE then you will find CELL a challenge but not a bad challenge.

And yeah the added L2 cache would make it a Brazilian times easier.
acousticvan
12-29-2006, 05:47 PM
Nerve, where has the thread about a developer talking about RSX on avsforum gone in the console technology section on beyond3d?
cliffbo
12-29-2006, 06:04 PM
Nerve, where has the thread about a developer talking about RSX on avsforum gone in the console technology section on beyond3d?

http://www.beyond3d.com/forum/showthread.php?t=35566
acousticvan
12-29-2006, 06:45 PM
http://www.beyond3d.com/forum/showthread.php?t=35566

Not this one cliffbo.
The other one, just up last night and its gone now.
The developer he said he's been the business for 12 years.
cliffbo
12-29-2006, 06:49 PM
Not this one cliffbo.
The other one, just up last night and its gone now.
The developer he said he's been the business for 12 years.

sorry can't help you there fella
acousticvan
12-29-2006, 07:24 PM
sorry can't help you there fella

My fault, it's still there but the title got changed.
lips
12-29-2006, 08:00 PM
heh. read that one too. Seems a poster who programs gpu code was irrate because he cannot dedicate his alus to pixel shading. I can see how being forced to create low latency results can be irritating if your resources are not dynamic. too bad no one can tell him to take a chill pill and allow the team to balance out actual available resources.
cpiasminc
12-29-2006, 08:55 PM
Yeah, I must have misread this awhile back…
Perhaps you were thinking of the papers of XeCPU, which said something like they "had to drop OOOE" for various reasons. But even XeCPU was never really intended to be OOO at any point. The "dropping" of OOOE had to do with paper designs of the original research CPU that Microsoft looked at for 360. MS just said they would like some adaptation of that work, but that was long after IBM had done a fair bit of research on the matter.

I don't know of any source which suggested anything about Cell being OOO or even *having* OOOE. Asynchronous execution of whole job contexts, sure, but that's just the nature of multi-threaded code on multiple CPU cores.

Seems a poster who programs gpu code was irrate because he cannot dedicate his alus to pixel shading.
juh? The biggest thing I saw him complain about was the vertex attribute read rate and the support for various pixel formats, the former of which hurts bad when the vertex format is really big, and the latter being more of a nuisance at the tool-chain end.
lips
12-30-2006, 05:55 AM
juh? The biggest thing I saw him complain about was the vertex attribute read rate and the support for various pixel formats, the former of which hurts bad when the vertex format is really big, and the latter being more of a nuisance at the tool-chain end.

eh? I though he did mention something about not enough shaders. Well, I have been wrong before. In my defense, later posts do contend his points. I'm just glad I am not a graphics end optimizer as a coder, I often find tackling gpu core issues feels tantamount to the pot and stove analogy.
nwo504
12-30-2006, 06:24 AM
grrovy