-
Hey Guest. Check out your NeoGAF Wrapped 2025 results here!
ARS Technica 360 Article - Part 2 now up
- Thread starter 3rdman
- Start date
Suikoguy
I whinny my fervor lowly, for his length is not as great as those of the Hylian war stallions
Great, so the games will look good, but developers will really have to try hard to get them to play well?
Btw, its certainly interesting that Microsoft made the killzone demo :lol
Also, you gotta love comparing tech sheets :lol
Btw, its certainly interesting that Microsoft made the killzone demo :lol
Also, you gotta love comparing tech sheets :lol
MetalAlien
Banned
This is the only part any of us should care about with hardware that's on this level.
OG_Original Gamer
Member
I thought the SPE's were capable of handling physics, A.I, and game control is something I had not thought about.
WHOAguitarninja
Member
Umm...what? Isn't this supposed to be about X360?
Pseudo judo
Member
The more information that comes out about the X360 the more it becomes obvious that the console will be able to stand side by side with the PS3 without any real concern about being over shadowed both graphicaly and in heavy CPU related calculations.
MS have set themselves up very nicely nextgen interms of power, online, hardware design (controller mainly), 20 gig hdd standard, lower price point and first to market.
What ever money Sony's got left in it's coffers best go to the marketing department.
MS have set themselves up very nicely nextgen interms of power, online, hardware design (controller mainly), 20 gig hdd standard, lower price point and first to market.
What ever money Sony's got left in it's coffers best go to the marketing department.
Xenon
Member
I'm sure they'll be able to work around this issue. But, I knew there was going to be some shortcoming to these new systems.
MetalAlien
Banned
Pseudo judo said:The more information that comes out about the X360 the more it becomes obvious that the console will be able to stand side by side with the PS3 without any real concern about being over shadowed both graphicaly and in heavy CPU related calculations.
MS have set themselves up very nicely nextgen interms of power, online, hardware design (controller mainly), 20 gig hdd standard, lower price point and first to market.
What ever money Sony's got left in it's coffers best go to the marketing department.
Well yea, they are coming out less than 6 months apart... Sony isn't magic!
Pseudo judo said:The more information that comes out about the X360 the more it becomes obvious that the console will be able to stand side by side with the PS3 without any real concern about being over shadowed both graphicaly and in heavy CPU related calculations.
MS have set themselves up very nicely nextgen interms of power, online, hardware design (controller mainly), 20 gig hdd standard, lower price point and first to market.
What ever money Sony's got left in it's coffers best go to the marketing department.
*gasp*
The two systems are competitive, but different? Run, man. They crucify people for saying things like that around here.
Pedigree Chum
Banned
GhaleonEB said:
With all the hype that Cell and RSX are getting on this baord you'd think they're from two differnet gens. Not bloody likely, MS has created an awesome system with X360, it isn't going to be like DC->PS2 or PS2->Xbox.
Not having a clear winner in hardware performance is not a good thing. Think how drawn out and vicious the ps2 vs. xbox debates were this gen, despite the fact that xbox is clearly superior. In the past year or two they've quieted down considerably, but I wouldn't expect that to happen next gen..
Pseudo judo
Member
I'm fairly new here so I need to know if you're attempting to make a joke or if you're serious.Rhindle said:
Pedigree Chum
Banned
Rhindle said:
40? It can do double that easily! :lol
Tenacious-V
Thinks his PR is better than yours.
Well it's just a case of reality is all. There were so many hardcore Sony fans who fell for the hype that PS3 and CELL was some god machine. Going from the greatest thing since sliced bread, to comparable to a product out 6 months earlier, is a little shocking to say the least.
I knew it all along, you just have to be a realist. Don't fall for PR.
I knew it all along, you just have to be a realist. Don't fall for PR.
WHOAguitarninja
Member
sangreal said:
I suppose. I interpreted to mean that the Killzone demo was part of the MS pre-launch hype. Poor wording, I suppose.
thorns said:Interesting article. It seems both are going to be a pain to program game code for.
You were supposed to put a title like : CELL game code performance worse than X360.
Also I thought that each core only had 1 VMX unit...the article specifically states two per core. So its got 6?
Ok question to techies:
So basically it's saying that with non-graphics code (AI, physics, etc) 360 will have the edge over PS3. But with graphics code PS3 will have the edge.
My question is this:
What if a 3rd party makes a game for 360 with insane AI, physics, etc that will have to take a (small) hit on PS3. How would this work? Lower frames per second? Lower the graphics quality?
So basically it's saying that with non-graphics code (AI, physics, etc) 360 will have the edge over PS3. But with graphics code PS3 will have the edge.
My question is this:
What if a 3rd party makes a game for 360 with insane AI, physics, etc that will have to take a (small) hit on PS3. How would this work? Lower frames per second? Lower the graphics quality?
Agent Icebeezy
Welcome beautful toddler, Madison Elizabeth, to the horde!
I'm a techie, but wow, I'm impressed to say the least. Microsoft has done a good thing with this X360
Tenacious-V
Thinks his PR is better than yours.
Razoric said:
My guess is Graphics per say wouldn't take a hit, but physics (explosions etc would be dumbed down a bit) and AI would. More dudes walking into walls..... :lol :lol
]Tenacious-V said:
Is this a joke? Seriously, you singled out Physics which is a task that's inheriently concurrent and inheriently suited towards an architecture like CELL. The whole premise of the Physics Processing Unit/PPU is to have a dedicated massively SIMD co-processor (read: CELL-esque) dedicated towards parallel constraint solving and integration; and the PPU is likely NOT going to be comparable to CELL in terms of preformance.
AI is the same thing, it's yet another problem with many solutions in the academic community based around connectionist concepts that haven't been used by game developers to this point because not only have they not had the preformance to run them, but they've had no need to since they're running their games on single-cored, single threaded x86. What game has the most advanced AI? The Sims and it's small use of FuSM? I mean, give me a break.
Mark Rein said:
aaaa0 said:
Now this is most definitly a joke.
aaaaa0 said:
Ars seems to be suggesting all game logic, physics and AI code would have to be squeezed onto a PPE - that's simply not the case. The SPEs should excel at physics, for one. You may need to do a little dance to get the SPEs to sing, but once they do..
The SPEs are for far more than just graphics/procedural content.
It's interesting to note the comments about "branch-intensive" code, though, no? So no OOOE, poor branch prediction (or constrained by cache) - and there've been similar rumblings for a while out of a couple of devs on B3D re. X360's performance with certain code - kinda deals a heavier blow to Mr. Nelson's assumptions than the corresponding comments on Cell help them.
It's also interesting to note that Ars's comments are confined to discussion of branch-intensive code. With the SPEs it's not even a consideration since as mentioned, there is no branch prediction logic. So devs are forced to use loop unrolling and so forth - something which the SPEs nicely accomodate with their healthy number of registers. You effectively have to "do a little dance" to get the SPEs to work at all with that kind of code, whereas on X360 you can not make much effort to accomodate the architecture and just get mediocre results.
I'd also take issue with the comments on cache amount in Cell - the PPE has 512Kb of cache, but it's not like that's necessarily shared with any of the SPEs. The SPEs don't have cache, but they have 256Kb of their own local memory. The PPE may well have the entire cache to itself if you want. There's over 2MB of on-chip memory between cache and local sram in Cell versus 1MB on X360, so I don't think that'll be as much of an issue for Cell, again as long as you work with it instead of throwing your code at it and just hoping it'll "go".
Tenacious-V
Thinks his PR is better than yours.
Vince said:
Jesus guy, calm down.. Don't take it like I was inherently bashing CELL or PS3. I know you worship it but geez, don't bite my head off. I was being light hearted. Notice the "my guess" and the lol heads at the end? Don't be so quick to attack and defend.
MightyHedgehog
Member
Doesn't it seem weird that these write-ups seem to be nothing more than easily digestible (for the layman...me) explanations and speculation on hardware he obviously doesn't have full access to? I mean, I appreciate the effort put into them, but it's a bit odd that he's doing this now instead of waiting for full disclosure of the actual hardware that will be in the system.
Cerebral Palsy
Banned
X360 > PS3
OH NOS!
OH NOS!
Tenacious-V
Thinks his PR is better than yours.
GhaleonEB said:
No kidding.....
MightyHedgehog said:
There's a lot more out there on the 360 than the PS3, so that's why they did it. The comments relating to the PS3 were limited, but focused on areas where the most information is available. If history is any judge, a full comparison will come out down the road with a more robust analysis.
I think the best analysis anyone can read that I've seen is available HERE and HERE .
Remember, gofreak and Panajev are students, and Vince writes medical software. From the way aaaa0 writes, it's pretty clear he's been developing games for a long time, and is probably a lead engineer.
edit: fixed the link
Remember, gofreak and Panajev are students, and Vince writes medical software. From the way aaaa0 writes, it's pretty clear he's been developing games for a long time, and is probably a lead engineer.
edit: fixed the link
How depressing, it seems that intel chips ie. PCs will kill the 360 and ps3 when it comes to running most gameplay code, but gee this poor gameplay sure will look pretty. 
No thanks. It seems like its going to take a programmer alot of work to merely get to the level that we're currently at gaming wise.
No thanks. It seems like its going to take a programmer alot of work to merely get to the level that we're currently at gaming wise.bad article from Ars.
they basically take all their assumptions from PC-centric solutions, apply them to X360 and bam! - shit for games.
The simple fact is that both Sony and MS have gone for non OOoE chips, non DP peformance in favour of massive parallel streaming processors.
Now I'm no expert, but I'm guessing that was for a good reason, otherwise they'd have just stuck a couple of Athlon 64s in there.
The simple fact is that Ars' assertion that the XeCPU cores are no good for game code, physics or AI is bollocks. Yes, in theory game code ported straight from a PC might not be optimal, and there might be some workarounds needed for very general code. But General housekeeping code is a small part of a game. Much more important are AI, Physics, Vertex processing. All of these are perfectly suitable for parallelising and vectorising (i.e. squirt through a massive SIMD pipe). Thats what the SPEs are for, and thats what the XeCPU is concentrating on.
J Allard even said that they took some of the non game critical stuff out of the XeCPU, such as Double Precision, to concentrate on what matters for games.
they basically take all their assumptions from PC-centric solutions, apply them to X360 and bam! - shit for games.
The simple fact is that both Sony and MS have gone for non OOoE chips, non DP peformance in favour of massive parallel streaming processors.
Now I'm no expert, but I'm guessing that was for a good reason, otherwise they'd have just stuck a couple of Athlon 64s in there.
The simple fact is that Ars' assertion that the XeCPU cores are no good for game code, physics or AI is bollocks. Yes, in theory game code ported straight from a PC might not be optimal, and there might be some workarounds needed for very general code. But General housekeeping code is a small part of a game. Much more important are AI, Physics, Vertex processing. All of these are perfectly suitable for parallelising and vectorising (i.e. squirt through a massive SIMD pipe). Thats what the SPEs are for, and thats what the XeCPU is concentrating on.
J Allard even said that they took some of the non game critical stuff out of the XeCPU, such as Double Precision, to concentrate on what matters for games.
rastex said:
Possibly. But also possibly a PC coder.
Its already been commented on that although you don't have branch prediction, you can lay down branch hints earlier in the instruction stream and prepare for them, especially where these branches are predictable.
And no matter how good any individual is at commenting on architecture, we don't have a full architecture flow for either 360 or PS3, and no block diagram for PS3 at all. Plus, these guys would not spend 100's of millions of dollars on a new console, with a completely different CPU layout to PCs, unless there was a clear benefit from a gaming point of view.
I hope you're right. I've been waiting to find out the nasty weakness of these machines, you know the stuff that the big guys don't talk about in their pie in the sky press conferences. Last gen it was the ps2 and its AA problem and the xbox and its bandwidth issues. This gen it looks like non-graphical code will be a problem. I'm sure there will be work arounds but they hardly workout very well or without compromises.mrklaw said:
aaaaa0 addresses this:mrklaw said:
On an SPE, you would have to make sure you schedule a DMA to upload the function you're about to call into SPE LS. Then you have to wait for that upload to finish. That will destroy performance. So you need to structure your code into overlays that you can swap in and out of SPE LS. Then you need to always make sure the functions you use together, get swapped in and out together, so you avoid doing as many DMAs as possible. If you want to get fancy, you need know in advance which functions you might want to call, so you can pre-queue a DMA so the function you want to call is in LS before you need it. Oh that means you need to reserve a chunk of LS for this, hm, that means you can fit less other stuff into LS, which means you might need to DMA more often for other stuff. Hmm.
So it turns into a big balancing act if you want to run the kind of code a normal CPU runs, and introduces a whole load of new stuff you have to worry about that you don't on a normal CPU.
So this is the way I understand it. Since people like to bring up physics a lot, here's a scenario that I think highlights what aaaaa0 is saying. If not, I hope he corrects me. So if you have a method that runs through all your colliding objects to resolve your collisions, so
Code:
for each collidedPair
if(STICKY) processOptimizedStickyCollision()
else if(PERFECT ELASTIC) processOptimizedElasticCollision()
else processGeneralCollision()
end forso let's look at X360 first. The first time the loop is run, none of the methods are in cache, and without loss of generality, say the first method is called. Depending on the sophistication of the caching mechanism, either the entire routine will be loaded into cache (GOOD) or line by line will be loaded into cache (bad on the first run). On the next iteration of the loop the same call could be made, in which case the method will already be in the cache so it'll run very quickly, or another method will run and be placed into cache. After a while all 3 methods will be placed into cache, so no matter what branch is taken all 3 methods will execute very quickly with absolutely no extra thought taken by the programmer. The other advantage is since the cache is shared, then you can have the loop running on one thread, and on the first iteration the method will be dispatched to a 2nd thread, then on the next iteration while the 1st method is still being calculated, if there's a free thread somewhere well you can now use it AND make use of cache coherency. Again, all without anything special required from the programmer.
Now for the Cell approach. I'll make a few assumptions: You can specify which SPE you wish to run code on, and you complete control over local memory for an SPE. On any iteration of the loop you don't know which method will be called. If you wait until the method is determined, and the method has not been loaded up on a specified SPE, then you have to load it up which will stall the SPE, let's say SPE01. So instead of stalling, you could preload your methods, but since you don't know which method will be called, you preload ALL of them. Now do you put all 3 methods on 1 SPE? 3 different SPEs? Have 2 on one, 2 on another? Any combination is possible, and the best, or at least a very good one has to be determined. The downside of preloading your method is that you're now taking away from the available Local Memory of the particular SPE, and THAT must be considered when you wish to preload other methods. So to get performance comparable to what will happen automatically on the XeCPU, the programmer has to take many more things into consideration. And the more things a programmer has to take into consideration the much higher chance of bugs being introduced. And if you've ever wrote a concurrent program before, then you'll know how annoying concurrency bugs are to track and fix. And this is one very limited and trivial example, but I think demonstrates the idea, hopefully...
While true that we don't have all information, you can't automatically assume that the missing pieces of information will be the optimal solution for game development. Every console has issues and mistakes in their design that makes things more difficult for developing games and these next consoles will be no different.
Well one thing to note is that this kind of situations is what code overlays were made for - except that back in the day when those were popular they would have to schedule code uploads around ass-slow floppy disc reading rather then very fast DMA uploads.rastex said:
But either way a concept that's been proven to work without having the programmer manually micromanage every function upload.
Moreover the fact that we're dealing with pretty fast DMA means implementing real caching schemes is also feasible.
Of course, in this particular example the overhead of function calling is negligible to the bulk of real processing that happens before and after this little switch statement, so it's kinda moot point how you'd handle it
understood, but consoles are a very closed system, and developers have complete control over what goes where.
With 7SPEs at your disposal, you'd configure them based on your game. So if its Physics heavy, you'd maybe dedicate one or even two SPEs to the tasks as necessary.
With 256K of local store per SPE, I'd think you'd be able to have several methods in one SPE.
With 7SPEs at your disposal, you'd configure them based on your game. So if its Physics heavy, you'd maybe dedicate one or even two SPEs to the tasks as necessary.
With 256K of local store per SPE, I'd think you'd be able to have several methods in one SPE.
rastex said:
I have not yet had a chance to respond to aaa0's post in that thread, but suffice to say I disagree with some of what he writes - he seems to be lifting wholesale from the points made in MS's spin without further comment (for example "SPEs can't touch main memory" and to access main memory you have to schedule a DMA - just saying that without qualification sounds ludicrous. Every CPU can only operate directly on the data in its registers, let alone local memory or cache - beyond that, the memory heirarchies are different, but the penalty to touch main memory on either is sufficiently severe on either chip. You just avoid that in different ways on both). I also don't think you know me or Pana or aaaa0 or Vince enough to try and prejudice the argument in this manner. I'm no longer a student, btw.
To me, it seems to be increasingly suggested that X360's CPU has tried to make many of the same tradeoffs as in Cell, but without nearly as much gain. It could be an uncomfortable middleground. If Cell, to use Ars's words, is just "a bit" less suitable for branch-heavy, load/store heavy processing or "general processing", that's a surprising turn of events and makes that tradeoff for vastly increased performance with other, very relevant workloads seem all the more worthwhile.
Razoric said:
I love this - either respond to my post and argue where you think I'm right or wrong, or shut up. Attacks on character are a last resort for the desperate. I've seen debates like this degenerate into useless tittle tattle on other boards because of posts like this.
SteveMeister
Hang out with Steve.
pj325is said:
This is why Xbox Live is so important to Microsoft. Beyond exclusive titles, Xbox Live is the feature that really makes the Xbox 360 stand out. From everything I've heard or read so far, Sony has *NO* cohesive online plan -- they'll probably just let developers who have online games do whatever they want, which means that consistency is out the door.
So unless Sony has something Live-like up their sleeve, they only have two ways to "beat" Microsoft -- exclusive titles and the PlayStation brand. Wonder if it'll be enough?