Last we heard it was still alive. Who knows what happened since.Pistolero said:Oh Gosthunter is the elephant in the room nobody seems to notice. That was an amazing engine pusher, with superb lighting and textures...
Thanks for correcting me. I've also read that Eight Days was canned...Who knows though...
garrickk said:A while ago, when we were first learning of the Cell architecture, I thought it was postulated that eventually a library would be developed to allow one of the SPE's to function as a branch predictor for the PPU - allowing the PPU to more efficiently crunch "general purpose" code.
I'm sure a lot of this is predicated on how efficiently the system memory <-> Cell cache operates, how well Cell cache <-> SPE cache flows, the depths of the cores, etc. The raw processing of the SPE utlized for this may be squandered, but hey, there are plenty of them for most applications, right? I mean, if you the raw throughput of the PPU is increased and the loss of a single SPE isn't missed. What about people using Linux, etc?
I'm not sure where I read this theory - EETimes or perhaps just an ArsTechnica forum post from a member who sounded technically sound? For all I know, it's not at all feasible, or actual software engineers got their hands on the hardware and realized it was either too difficult to implement or offers little benefit. Just curious if anyone has heard anything.
loosus said:Does the processor in the Wii share any similarities with those of the 360 and PS3? Or, are they just entirely different beasts despite being made by IBM?
garrickk said:A while ago, when we were first learning of the Cell architecture, I thought it was postulated that eventually a library would be developed to allow one of the SPE's to function as a branch predictor for the PPU - allowing the PPU to more efficiently crunch "general purpose" code.
I'm sure a lot of this is predicated on how efficiently the system memory <-> Cell cache operates, how well Cell cache <-> SPE cache flows, the depths of the cores, etc. The raw processing of the SPE utlized for this may be squandered, but hey, there are plenty of them for most applications, right? I mean, if you the raw throughput of the PPU is increased and the loss of a single SPE isn't missed. What about people using Linux, etc?
I'm not sure where I read this theory - EETimes or perhaps just an ArsTechnica forum post from a member who sounded technically sound? For all I know, it's not at all feasible, or actual software engineers got their hands on the hardware and realized it was either too difficult to implement or offers little benefit. Just curious if anyone has heard anything.
Wow the ghosthunter game looks amazing for a PS2 game.SolidSnakex said:No that's Sony London, they're currently working on Singstar, and atleast had Eight Days in development if its still around. Cambridge is Ghosthunter and Primal
Ghosthunter
If you're talking about the Wii, yeah. If you're talking about the 360 and PS3 you're wrong. These systems are the closest any 2 systems have ever been in the same generation.MikeB said:For this generation there's a much greater performance and specs gap, if compared to last gen.
I thought the PPE had 512KB of cache and EACH SPE had 256KB of cache - all running at the full clock speed.alske said:The SPEs don't have cache. It's all programmer controlled, which is both an advantage and a disadvantage. I can't find anything in IBM's documentation on how deeply the SPEs are pipelined. They state that the PPU is around 20 stages deep. I kind of doubt the SPEs are that deep.
A lot of the design decisions made in Cell seem to be fairly similar to Itanium. Cell doesn't seem to have speculative loads, but it is massively parallel, doesn't have out-of-order execution and seems to rely heavily on programmers optimizing their code. Documentation on both architectures seems to indicate that a lot of this programmer optimization is supposed to be compiler aided, and at least in Cell's case, that magic compiler doesn't seem to exist.
garrickk said:I thought the PPE had 512KB of cache and EACH SPE had 256KB of cache - all running at the full clock speed.
BTW, jonabbey's post above I think nailed it. The branch prediction in the SPE is sort of impossible without the SPE knowing precisely (or at least efficiently) which branches the PPE is making. I assumed that the PPE branch prediction was so meager and perhaps "obvious" that the SPE could guess/know what the PPE was doing and take the "wrong" branch everytime. Even if that was the case, there is a lot of overhead, multiple branches, and other problems. I hadn't really thought it through.
Pistolero said:Oh completly forgot about the Cambridge guys...Aren't they the ones responsible of The getaway ? The fact that they're helping propagating technical excellency through the veins of Sony studios won't tie their hands I hope. The first Gateaway was quite impressive, if short of some solid gameplay mechanics.
GoW III will come to stab every other self-imposed god once everyhting is said and done..
ROFL said:Getaway was done by my old mates Team Soho (or 'Sony London' - ugh). They released some very impressive early target renders yonks ago for a PS3 Getaway but I've not heard or seen anything since...
alske said:The SPEs don't have cache. It's all programmer controlled, which is both an advantage and a disadvantage. I can't find anything in IBM's documentation on how deeply the SPEs are pipelined. They state that the PPU is around 20 stages deep. I kind of doubt the SPEs are that deep.
A lot of the design decisions made in Cell seem to be fairly similar to Itanium. Cell doesn't seem to have speculative loads, but it is massively parallel, doesn't have out-of-order execution and seems to rely heavily on programmers optimizing their code. Documentation on both architectures seems to indicate that a lot of this programmer optimization is supposed to be compiler aided, and at least in Cell's case, that magic compiler doesn't seem to exist.
jonabbey said:I believe the SPUs have variable pipeline length, depending on the operation. The longest operations take 7 cycles. The SPUs are also dual-issue, with certain operations on one pipe and the rest on the other.
The SPUs have branch hinting, at least, and they will pre-fetch the hinted branch.
And, yes, the compilers aren't anything close to magic, but one nice thing about the SPU architecture is that you can focus your optimization in a very well understood environment, without variable memory access latency, for specific discrete tasks running uninterrupted.
If you had to do that kind of optimization image-wide for a multithreaded environment running on the PPU, say, you'd have a much harder time achieving near-optimal performance.
ROFL said:Getaway was done by my old mates Team Soho (or 'Sony London' - ugh). They released some very impressive early target renders yonks ago for a PS3 Getaway but I've not heard or seen anything since...
http://en.wikipedia.org/wiki/The_Getaway_(video_game)
<The Getaway 3
A third Getaway title is in development (with a working title ) of The Getaway 3 or The Getaway Working Title by Team SOHO for the PlayStation 3. A technical demo featuring Piccadilly Circus was demonstrated in May 2005, but this was not directly from the game. It has been confirmed that a third game would again be set in London; however, rumours have been circulating that the game could also feature a city in Europe, more than likely Amsterdam given the most recent trailer.>
I have looked at assembly output of C compiler generated code for Itanium and it is not too shabby.alske said:I totally agree. I was merely pointing out that both Intel and IBM promised compilers for their exotic chipsets, and that they never really materialized. Granted programming an SPE is probably much simpler then hand generating IA-64 assembler (yuck).
..
is the polygons that made it that impressive.PistolGrip said:Wow the ghosthunter game looks amazing for a PS2 game.
Any info on Ninja Gaiden Sigma, DiRT, Tekken 6, Metal Gear Solid 4, Haze, Turok, Hot Shots Golf 5, etc regarding SPE usages?
Hammer24 said:Don´t you just hate it when your absolutely perfect speculational fanboy thread gets totally ruined by actual dev´s and programmers?
chubigans said:Wow, you're an idiot.
I take informational posts like the ones of alske f.i. over pointless fanboy/troll baiting posts anyday. If that makes me an idiot, so be it.
MikeB said:@ Hammer24
I find the comments provided by expert PS3 developers most informative, that's the main reason why I started this thread.
Why not try to add something useful yourself, instead of trolling?
Hammer24 said:Then why do you post PR speak? The forum posts here are way more believable than those.
Simple: I am no programmer nor dev - I couldn´t add anything substantial to a technical topic like this. But I enjoy reading them.
Don´t you notice that you scare the experts away with your PR quotes?
dogmaan said:Mike B uses the PR quotes as evidence of SPU Usage, not to scare dev's away, or for PR purposes
PistolGrip said:Wow the ghosthunter game looks amazing for a PS2 game.
Hammer24 said:Don´t you just hate it when your absolutely perfect speculational fanboy thread gets totally ruined by actual dev´s and programmers?
Hammer24 said:.....I couldn´t add anything substantial to a technical topic like this.
MikeB said:@ Orlics
No they don't because game engines need to be redesigned to take advantage of the SPEs. More development time equals more costs, cross platform development is more difficult as the XBox 360 nor PC provide anything truly equivalent to a SPE. The Cell isn't just a multi-core CPU, the SPEs act more as independent processors than within a multi-core design.
I will post a new message within which I will compare the XBox 360 and PS3 to greater depth. I am sure it will wind up some fanboys though.
PPU alone can't run code that's made for the Xenos
MikeB said:This comment seems to indicate that "The Darkness" does not use the SPEs:
"It depends on the type of engine you are doing. In The Darkness they have pretty similar performance, but that is very intentional from our side. We need the two platforms to perform similarly, and therefore we cant design features that would take advantage of the difference of the two platforms. To my knowledge the PS3 has untapped potential in its seven SPUs"
Source: Edge online
http://www.edge-online.co.uk/archives/2007/02/you_interview_s_1.php
From a HS dev's blog, regarding "Atomic Cache Unit":alske said:Each SPE has 256 kb of memory that is as fast as cache. It is not cache though. It's completely under programmer control. Check the diagrams on page 24 of the Cell Broadband Engine Architecture. It appears that there is a shared cache for all the SPEs, but that is likely for when you try to pull in data over EIB.
The ACU (s) are a part of each SPU that allow atomic updates to occur very quickly. It appears fairly simple each SPU had 512 bytes of cache (yes contrary to what you might have heard SPU do have a tiny bit of cache). 512 bytes is divided into 4 128 byte lines.
MikeB said:This comment seems to indicate that "The Darkness" does not use the SPEs:
"It depends on the type of engine you are doing. In The Darkness they have pretty similar performance, but that is very intentional from our side. We need the two platforms to perform similarly, and therefore we cant design features that would take advantage of the difference of the two platforms. To my knowledge the PS3 has untapped potential in its seven SPUs"
Source: Edge online
http://www.edge-online.co.uk/archives/2007/02/you_interview_s_1.php
MikeB said:This may be of relevance to Ratchet and Clank: TOD or the recently announced sequel to Resistance Fall of man.
An interesting read, Insomniac regarding their new igPhysics system.
Introducing SPU shaders:
http://www.insomniacgames.com/tech/articles/0907/files/spu_shaders_introduction.pdf
Benefits of SPU Shaders to igPhysics
? Pipeline well defined and completely SPU-driven
? SPU processing completely asynchronous
? Data well-organized and well-defined.
? No (or minimal) PPU intervention
BTW, other lowend and higher end Cell based products are being planned, not for nothing are Cell processors being further developed, clocked lower as well as clocked higher than the version specced for the PS3..
cool info !MikeB said:
MikeB said:With regard to Half-Life 2 The Orange Box, I expect to see no miracles,
Microsoft millionaire from the singletasking CLI-only MSDOS era and co-founder of Valve, Gabe Newell:
"I think the PS3 is a waste of everybodys time. Investing in the Cell, investing in the SPE gives you no long-term benefits. Theres nothing there that youre going to apply to anything else. Youre not going to gain anything except a hatred of the architecture theyve created."
BTW, other lowend and higher end Cell based products are being planned, not for nothing are Cell processors being further developed, clocked lower as well as clocked higher than the version specced for the PS3. Developing with the SPEs in mind surely provides longterm benefits as the PS3 will be around for quite a while and a PS4 will almost certainly use a higher specced Cell processor as well.