Xenos - RSX - What was left out?
- Thread starter blakjedi
- Start date
Vysez said:
Maybe you should direct that troll statement at the respondee instead of the responder.
BTW I don't need to prove anything especially SPEs not being able to do GP work. Heck ANY processor can do GP work with enough effort. There are GPUs being used to do GP work.
Question is, will it be as good as something design to do it better? The concensus from developers here seems to indicate SPEs aren't that great at GP code unless YOU can prove otherwise.
No, only the definition of GP workBrimstone said:
Before we can answer if XeCPU is 3x better at GP work (or is that GP code??) than Cell, we need a clear definition - which MS didn't provide, so we're left guessing. Research has shown new algorithms can take GP operations and run them extremely effectively on non-GP (streaming) hardware. GP code, with buckets of conditional branches will never be good on streaming hardware, but if you can rewrite your processors to structure data fetching etc. to work with in-order steaming, your processor can do GP work.
So (as I've said before) 1) SPE's can do GP work. MS totally discounted them as though they have no input, which is totally untrue. There's no info on how well or badly they can do GP work, but they can do it. 2) Even if the SPE's are poor at GP work, there's 7 of them. If a SPE handles GP at 25% efficiency as a PPE, XeCPU has 300% one PPE, Cell has 265% one PPE. Even at only 10% efficiency per PPE, that's 170% one PPE for Cell, which is over half XeCPU. In short, the 3x figure has to be pretty way-out-there.
I think SPEs not being able to do GP work is a strawman argument. Like I said even GPUs can do GP work, but that doesn't mean much does it? Arbitrary numbers like 25% for each SPEs don't mean much either. How about I give an arbitrary number of 15% for each SPE at GP?
Oh and regarding the leaked document, I think someone had a problem reading it and understanding it hence coming to the wrong conclusions.
Oh and regarding the leaked document, I think someone had a problem reading it and understanding it hence coming to the wrong conclusions.
PC-Engine said:BTW I don't need to prove anything especially SPEs not being able to do GP work. Heck ANY processor can do GP work with enough effort. There are GPUs being used to do GP work.
The whole point of a "general purpose" processor is that it can do anything, it's not designed specifically for any one thing, and thus isn't particularly brilliant at any one thing. In mapping "general" apps to GPUs and the likes, the general trend seems to be that there has been big performance improvements.
PC-Engine said:
There's a set of tasks that fit naturally on SPEs, and then beyond that, tasks have to be cast or mapped to them. It is a very open question as to what can and cannot be done with them. I think it'll be a function of developer effort.
I'd also like us to concretise what exactly we're talking about. People seem to be hiding behind this nice nebulous term "general purpose processing". With SPEs an open question, how can we ascertain the margin of advantage Xenon has with such tasks anyway? Furthermore, is performance with tasks heavily reliant on this as important if they're only taking a smaller fraction of CPU time anyway? When you optimise, you don't start with things that take the smaller amounts of time, you start with the largest.
Indeed, he's feeding your troll.PC-Engine said:Maybe you should direct that troll statement at the respondee instead of the responder.
But let's quit with this or else the whole thread will get the "treatment".
Nobody said programing SPEs was as easy as programing PPC cores, actually most people, that includes me, in this very thread, say it's obviously a more complex task.PC-Engine said:
But, and that's what you decide to ignore, it's not an impossible task.
The fact that it's hard to code for an SPE doesn't make its raw power null and void only because other types of cores are easier to work with.
PC-Engine said:
Let's try to keep 8 years old rhetoric out of the thread, please.What is the difference between integer processing & branching vs "GP code"? Allard says they don't run any general purpose code but that most games are made up of 80% general integer operations or operations such as branching. I thought thats what general purpose code is.
Neither CELL or Xenon are GP processors, but the PPE cores seems to be more GP than the SPEs. Xenon has 3 PPEs, CELL has 1 PPE and 7 SPEs.
Stop making up BS man. Nobody here said SPEs cannot do GP work, heck even a GPU can do GP work. Do you even read what other people post?
Well last time I check weren't you the one to demand proof while having ZERO proof of your own?
All we have now is the peak ops/Flops performances for all the cores cited (PPCs/SPEs). And, therefore we can make projections and conjectures only based on these datas.PC-Engine said:
Now what we need is some insights from people who worked, a long time, with both architectures if what you want is to determine the real world integer/ floating point performances of thoses two CPU. But for that you'll have to wait a few years.
Drop the attitude for a moment PC-E, it doesn't do you any good.PC-Engine said:
Better check yourself.PC-Engine said:Well last time I check weren't you the one to demand proof while having ZERO proof of your own?
edited: No need to continue this discussion.
Sounds like that 3x performance was bs when you take into consideration that general performance of the cpu dropped drastically. That's why I hate pr spin like that. In fairness though the games are looking better on beta than alpha, ie. pgr3 but who knows how the games are running.
ralexand said:
it's interesting that some of the posts in this thread are saying the SPEs will do great (read: better than the Xenon PPCs at GP code) AFTER devs have been given sufficient time to learn to optimize them.
With the next breath they say that Xenon has now been "proved" (based on this document) to be inferior and "uhh-ohhh big problems with the XCPU", without giving anyone enough time to figure out how to work with them.
Tap In said:
I'm not sure where you got that idea.
No one is saying that. A SPE won't be better than a PPE for "general purpose" processing.
My point, at least, is that they're not entirely dismissable when it comes to "general purpose" stuff either.
Tap In said:
The contrast between MS's PR re. this and the apparent realities of development deserves highlighting. They were the one's putting the spotlight on "general purpose" performance and what they would deem to be ill-advised Sony design choices - yet their own would appear not to be a million miles apart. It's just funny that a key "strength" of Xenon as claimed by MS really isn't really a significant strength at all in absolute terms.
You're right, I don't know the facts and that's why I stated sounds like which is based on innuendo and rumors that have come out. I certainly hope the rumors are false for the games sake. I would hate to think that all that money and research was spent for something that's a step backwards.PC-Engine said:
Titanio said:
well maybe all were not saying better than a PPC core but my point was that they are wiling to give time to optimize SPEs but the Xcpu appeared to be "dead in the water". (my words
)What's being talked about here sounds pretty much just like what anand was saying in the pulled GP article of both Cell and Xenon. Given time I would imagine both CPUs will have solutions to these very early issues.
Tap In said:Titanio said:
well maybe all were not saying better than a PPC core but my point was that they are wiling to give time to optimize SPEs but the Xcpu appeared to be "dead in the water". (my words
What's being talked about here sounds pretty much just like what anand was saying in the pulled GP article of both Cell and Xenon. Given time I would imagine both CPUs will have solutions to these very early issues.
Completely agree, but given the situation we find ourselves, some of MS's claims of advantage over Cell are looking even less convincing to me, and I'm wondering what else they've got. (See above also, I edited in a response to the latter part of your previous post).
scooby_dooby
Legend
How is it less convincing?
With the PS3 you only have a single PPE, in-order and not very strong.
The 360 has 3 of these cores, and has much more GP power up front.
In other words, dev's will be forced to rewrite code to run on the SPE's because the GP power of the PPE is insufficient.
Now, considering they already have to rewrite everything to be in-order, how much time do we expect them to have to rewrite AGAIN to optimize GP code to run using an SPE. My point is I don't think it will happen very often, and when it does it will be out of necessity because the PPE did not provide sufficient power.
So I don't think it's right to say they are roughly equal in GP power because the SPE's "could" be used, because the fact is the X360's GP power is much much easier to extract and will get utilized with less effort than extracting that power from the SPE's.
Without the SPE's the PS3 does not have very much processing power compared to the 360. The SPE's can't be discounted completely for GP code, but at the same time they dont make up for the fact the PS3 has much less GP power in it's main core.
btw - i've read the document in question. It was PM'ed to me a few weeks back. I don't see how you can come to the conclusion that the GP power advantage claimed by MS were false based on that. Whatever problems the XeCPU has, the PPE on the cell will have the same issues, but with 1/3 of the power.
And by and large according to that doc almost everything was faster on the Beta kits, with a few things marked as slower (don't have the doc on me, and the link is now dead)
With the PS3 you only have a single PPE, in-order and not very strong.
The 360 has 3 of these cores, and has much more GP power up front.
In other words, dev's will be forced to rewrite code to run on the SPE's because the GP power of the PPE is insufficient.
Now, considering they already have to rewrite everything to be in-order, how much time do we expect them to have to rewrite AGAIN to optimize GP code to run using an SPE. My point is I don't think it will happen very often, and when it does it will be out of necessity because the PPE did not provide sufficient power.
So I don't think it's right to say they are roughly equal in GP power because the SPE's "could" be used, because the fact is the X360's GP power is much much easier to extract and will get utilized with less effort than extracting that power from the SPE's.
Without the SPE's the PS3 does not have very much processing power compared to the 360. The SPE's can't be discounted completely for GP code, but at the same time they dont make up for the fact the PS3 has much less GP power in it's main core.
btw - i've read the document in question. It was PM'ed to me a few weeks back. I don't see how you can come to the conclusion that the GP power advantage claimed by MS were false based on that. Whatever problems the XeCPU has, the PPE on the cell will have the same issues, but with 1/3 of the power.
And by and large according to that doc almost everything was faster on the Beta kits, with a few things marked as slower (don't have the doc on me, and the link is now dead)
Similar threads
- Sticky