Agreed. I enjoy reading intelligent discussion as much as everyone else but this thread has become a beast. At least it's pretty easy to follow the tangents that created this monster.
Consoles & Propoganda
- Thread starter aselto
- Start date
- Status
Agreed. I enjoy reading intelligent discussion as much as everyone else but this thread has become a beast. At least it's pretty easy to follow the tangents that created this monster.
scooby_dooby
Legend
I don't think that was FUD. The DirectX10 group seems to think that integer instruction is more important for games and made some hardware requirement for it in DX10...I think. Even when speaking about DX10 the group talked down console flop performance on both systems in favor of integer. I wouldn't be surprised if that is where Nelson got that information from.
http://www.extremetech.com/article2/0,1697,1982101,00.asp
Last edited by a moderator:
scooby_dooby
Legend
Hmm..this pot seems a little still....gotta side with Chef on the MN document (surprise!), because as full of crap as it was, it's basic premise has proven true, that Xenon can hold it's own against CELL despite the specs being much lower, why? Because it is better at general purpose processing.
The systems are a wash, which is all the document has said, in the end that seems to be the case. So, somewhat validated in hindsight imo.
The systems are a wash, which is all the document has said, in the end that seems to be the case. So, somewhat validated in hindsight imo.
Uh huh. Ok, well I'll go ahead and give you my take on it as well ok?
False - the Cell processor is capable of delivering greatly elevated levels of performance in running 'game code' relative to the XeCPU; there is no task that cannot be moved to the SPEs given the proper optimization
The majority of present unoptimized code is of this nature, but there is no contraint that confines "game code" to this paradigm. When code is optomized for the Cell, it will run faster than the 'messy' code left unchanged for sake of ease and cross platform compiler-reliance.
The SPEs have local store instead of cache, which is faster but does put an extra burden on the programmer. 'General purpose' computing I would contend does not exist at all - it is a blanket statement to refer to unoptimized and legacy code, and it's use would refer most often to sclaer code that works best on an OOE processor. DSP stands for digital signal processor - these are *extremely* limited in what they are capable of handling, and MajorNelson is purposefully dismissive of the SPEs by likening them to the chip in you CD player. The SPEs are capable of performaing any task in the world of computing - they will not always be the best candidate, but with optimized code they will sometimes be the best candidate. And this includes many of the functions one might flipppantly put under the term 'game code.'
General purpose, here we go again...
Vector math and dot products... can you explain to me why these are important Chef? And after doing so, can you explain to me why the processing power of Cell in vector math being on the SPEs is a bad thing?
Here is my problem with all of this, people who don't understand the underpinnings of the technology read something like this MajorNelson article, and... it sounds knowledgable, right? But yet every single thing you're saying is true is either false or a patent misdirection on MS' part.
You said this was false, so good yes - this entire dot product hype is the red herring of the generation.
Here's how MajorNelson works it - he tells you that Cell has its FP operations performed on the SPEs; he then informs you that geometry work has moved mainly to GPUs in the modern era - the subtle hint you're supposed to get is that the FP power of the SPEs is good at geometry - and *only* good at geometry - but hey 360 has that covered via the GPU. And you would be left with a grave misconception thinking that the SPEs are there only for geometry work, because there are many other tasks they can be utilized for that you can be wel-assured that the Xenos cannot go back and help the XeCPU on.
How in he world are you saying that this is true? Tell me then - what type of game programming is the PS3 about?
The above two paragraphs are the most ridiculous in the entire bunch - again he's there with the 'general purpose' BS and again he's there with calling the SPEs DSPs. There's just no way to even describe what a false front he's presenting here, and how ignorant the reader that believed it would have made themselves afterward. By the way, Cell's SPEs are not confined to a streaming floating point model Chef, just in case you were wondering. Nelson seems obsessed with using the word streaming as often as he can.
They are certainly not a wash... Cell grinds the XeCPU under its heel. Even in his charts MajorNelson brings up Integer performance - he informs you that te XeCPU has three times the performance of Cell in this regard. Now, did you know that the situation is actually reversed, and that Cell actually has over twice the IntOp potential of the 360? Hey - with MajorNelson as a guide, how could you?
The bolded part is the only part of that that's true, and as such we'll leave it at that.
The MajorNelson document is the equivelent of 'professional' docs that say smoking doesn't cause cancer or global warming isn't happening. It is 100% misdirection. 100%. And what people like myself have to deal with is every month someone starting another thread in the tech section asking about "general purpose" computing, or "Cell is only good at floating point," or "Floating point doesn't matter in game code" type stuff.
Yes, it makes me crazy.
Ok here we go again - general purpose computing.
Scooby, let me ask you, what's a "general purpose" computing task you associate with gaming? Word procesing, spreadsheet software, what?
scooby_dooby
Legend
Obviously I'm no console programmer, but I simply define it as less specialized, generally more flexible and more easily suited to a wider variety of tasks.
Anytime you have specialized hardware like the SPE's, you lose flexibility.
What other explanation would you give for the fact MS's 3 cores are able to prouce similar, and even better results than the 7cores in the PS3 at this point in time? Is it not the fact that the PPE cores are more robust, more flexible, and easier to work with?
Okay guiys, I'm just going to step in here with a polite suggestion - don't write any more! We've had these debates before. Perspectives are very different and not going to change. If you enjoy the discussion, by all means go ahead. But if it were me (and why I haven't contributed) I'd say all that time spent typing responses can be put to other uses like...heck, playing the game consoles we own? I wouldn't spend any time trying to convince Scooby that SPE's are 'general purpose' or trying to convince xbd that float-maths isn't that important. Opinions on these matter are set in stone, given that reams of previous debate hasn't shifted anyone's opinion even an inch.
So of course feel free to carry on, but sometimes I have to remind myself how the time I spend on forums in debates that go nowhere, I'd rather be spending doing other stuff, and I thought you guys might appreciate a little reminder too.
I wouldn't spend any time trying to convince Scooby that SPE's are 'general purpose' or trying to convince xbd that float-maths isn't that important. Opinions on these matter are set in stone, given that reams of previous debate hasn't shifted anyone's opinion even an inch.So of course feel free to carry on, but sometimes I have to remind myself how the time I spend on forums in debates that go nowhere, I'd rather be spending doing other stuff, and I thought you guys might appreciate a little reminder too.
:smile:
Is the original statement factually incorrect or misleading? Big difference. He's not going to explain how one could rethink their programing habits and in fact from what I've seen since the inception of this document is most of this rethink is post the original article! So no, I don't believe this part was false. Misleading yes. False - no.
Again why would MN attempt to provide examples of rethinking game coding to fit cell in this document especially when at the time, many methods were still undiscovered?
Agreed. See above.
Is this statement False?
It isn't a bad thing, obviously MN will spin the facts to paint the picture which shows xb360 in the best light possible given the circumstance of the timing of this article(!)
Misdirection of course. Timing & circumstance.
:smile:
misleading ...of course
At the time of the writting ...
I know.
Whoa! I never said cell is a wash with xcpu! I said the ps3 is basicly a wash with xb360 as a system. Big difference. Also you never answered my question: Are there not circumstances where xcpu would outperform cell?
??? You're saying that current multiplat games are not roughly equal on ps3 and xb360 and that most unbiased devs have not called the systems a wash overall?? c'mon xbd! That's ehrm ... questionable.
Sorry you have to deal with these idiots in the tech forum ... I try not to cause too much trouble there.
Last edited by a moderator:
Okay guiys, I'm just going to step in here with a polite suggestion - don't write any more! We've had these debates before. Perspectives are very different and not going to change. If you enjoy the discussion, by all means go ahead. But if it were me (and why I haven't contributed) I'd say all that time spent typing responses can be put to other uses like...heck, playing the game consoles we own?
So of course feel free to carry on, but sometimes I have to remind myself how the time I spend on forums in debates that go nowhere, I'd rather be spending doing other stuff, and I thought you guys might appreciate a little reminder too.
Thoughtful debate is healthy isn't it? At least that's my my Mum tells me
I disagree Shifty I think we've added quite a few new members on this forum since the article and a good honest reflection is helpful IMO.
Plus I do enjoy a good conversation.
Last edited by a moderator:
(This'll be my only contribution to this conversation!)
What MS have done in being misleading is no different to to what a lot of Sony's marketting is that you complain about. A common hatred of E3 '05s trailers is based on people being mislead (in the main, although things got messy with different spokespersons giving different info as the matter was pressed) rather than outright lied to.
On the whole, there are few outright lies in marketting. There is truth stretched beyond all recognition, but still founded on a legitimate POV; and there's hyperbolic PR fluff which is meaningless. That goes for all and every company. That's the rules of the game. I want the rules changed, so all we get are proper facts and figures allowing for intelligent analysis, but that'll never happen and companies must be expected to keep trying to outplay each other with their PR gumpf. That means Sony will release visualisation of what devs think they'll be able to do with PS3, and MS will release pseudo-tech docs 'educating' the public on the way machines work, and everyone who isn't careful will get sucked in to some misunderstanding.
I can't believe we're digging up Major Nelson's articles again.
MAKE THE HURTING STOP.
MAKE THE HURTING STOP.
Ok, a while ago "general purpose" was discussed at length in a thread that wondered whether a modern OOE processor might not have been a better choice for these consoles.
Read this four-page thread: http://forum.beyond3d.com/showthread.php?t=38592
And come away with a better understanding of what that term means, and why Major Nelson is just throwing up smoke when he uses it. Essentially though it boils down to there being *no* game code run on Cell that need go unoptomized as the gen progresses. Major Nelson is doing things like implying that "general purpose" code like animation can't run on Cell... but if you change the way you do your animation, then voila! - suddenly it runs and it runs extremely well to boot. And that's why I hate the term, and it misleads those that read it.
Anyway, read the thread, it is *the* place to gain an understanding of the term, and saves me a lot of typing here.
Read this four-page thread: http://forum.beyond3d.com/showthread.php?t=38592
And come away with a better understanding of what that term means, and why Major Nelson is just throwing up smoke when he uses it. Essentially though it boils down to there being *no* game code run on Cell that need go unoptomized as the gen progresses. Major Nelson is doing things like implying that "general purpose" code like animation can't run on Cell... but if you change the way you do your animation, then voila! - suddenly it runs and it runs extremely well to boot. And that's why I hate the term, and it misleads those that read it.
Anyway, read the thread, it is *the* place to gain an understanding of the term, and saves me a lot of typing here.
scooby_dooby
Legend
Ok, a while ago "general purpose" was discussed at length in a thread that wondered whether a modern OOE processor might not have been a better choice for these consoles.
Read this four-page thread: http://forum.beyond3d.com/showthread.php?t=38592
And come away with a better understanding of what that term means, and why Major Nelson is just throwing up smoke when he uses it. Essentially though it boils down to there being *no* game code run on Cell that need go unoptomized as the gen progresses. Major Nelson is doing things like implying that "general purpose" code like animation can't run on Cell... but if you change the way you do your animation, then voila! - suddenly it runs and it runs extremely well to boot. And that's why I hate the term, and it misleads those that read it.
Anyway, read the thread, it is *the* place to gain an understanding of the term, and saves me a lot of typing here.
It's all semantics though, what is general purpose code? is there such a thing? who really cares? I think it's a silly phrase as well, I think we can all agree on that, and move on sinse it really clouds the discussion.
The fact is the PPE are better general purpose cores than the SPE's, they are more flexible, they are better suited to a wider variety of code. I think it was always implicit, that you could rewrite code to work on the SPE's, but that was at the heart of the argument, rewriting would take extra time, and come at a cost.
Now, if this isn't the case, why are PS3 xplatform developers having such a hard time matching the performance of Xenon with CELL? Is there another explanation?
The SPEs are semi-specialized; specialized in that they favor certain programming models, but general in that they can be applied to any task. It's not the processor that is the limiting factor in their case, but the programmer (for better or worse). They are however quite flexible. The XeCPU has an advantage in that it more closely resembles the processors most (non-PS2) programmers are used to working with, and that MS' tools support makes re-compilation from PC quite painless. But the cores themselves are not going to keep pace with well written SPE code, and although easy to wrap ones head around, their theoretical performance peak is actually harder to reach than the SPEs using optimized code.
Well, read that thread to gain insights into GP.
That's not true though. Again, the PPE cores are better towards models of programming, but not the tasks themselves. Change the method, and tap the SPEs - this will become an increasing part of what happens on the PS3 this gen.
First of all, are PS3 devs having a problem matching the XeCPU with Cell? I'm not hearing that - to my knowledge, the only deficiencies so far are more graphics-related... a GPU rather than CPU issue. Again, the XeCPU benefits from its approachability and its tools - but the actual architecture is no crowd pleaser among devs; ask the 360 devs here their thoughts. Cell will pull away, trust in this. Now whether it will be hamstrung and to what extent if it needs to be helping RSX out w/geometry vs Xenos, I don't know. But it's like this, if it was 2 consoles, Cell and Xenos vs XeCPU and Xenos, the former system would show its merit over the later in a big big way a couple of years after their respective launches, and those differences would be "game code" related.
Cross-platform usually means a quick port, at least recently. It's the same reason that Quake 4 on the 360 was so bad, the port from the x86 was poorly done in a hurry. So you take a 360 game and do a quick port, using no SPUs, and you get bad results on the PS3. If you have two dev teams working on the same game, each knowing their target platform you get better results (or do a good port, see Oblivion).
There is no debate that a PPU is more general purpose (ie. it would run Word and Firefox better with less coding), the question is how much code in a game is general purpose?
Both Sony and MS would have loved to have a top of the line Intel or AMD for their CPU, but they could not afford the price or heat, so they both made compromises.
there was that one slip of the lounge last year by the Assassin's Creed team that mentioned the branching in Xenon allowing "better" AI than the PS3 version... quickly recanted.
from what I've read wrt to Xenon/Cell comparisons I think the fact that nobody is talking about it does not mean that it is not an issue somewhere along the line in some game code (AI in that example).
Perhaps now devs are using the SPEs to pick up that slack. I (of course) realize that the SPEs allow the Cell to be SUPERIOR to Xenon cpu but... they are being leveraged in many, many instances to do some of the things the 360 does in other parts of the system; essentially giving us what many feel is the ultimate result... near parity between these two systems.
- Status
Similar threads
