• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

ExtremeTech on Xbox360 vs PS3

sorry if already posted


Xbox 360 vs. PS3—Who Will Win?

Right now, the raging message board wars about next-gen consoles seem centered on which is more powerful. Is having the most powerful console a recipe for certain success? I don't think so. Is coming out first enough to do it? Not if the Dreamcast has taught us anything.

There's an update on the blog of Xbox Live's own Major Nelson, with a Microsoft-prepared counter-argument that "proves" the Xbox 360 is in fact more powerful than the PS3. It's an interesting read, even if many of the arguments are flawed. It brings up a good point: Either system can be made to look more powerful than the other, depending on how you want to bend the specs.

In truth, the systems aren't entirely comparable. The 360 has a single, unified 512MB bank of GDDR3 memory. The PS3 has 256MB of GDDR3 connected to the graphics chip and a bank of 256MB XDR connected to the CPU. That's more total bandwidth in the PS3's favor, but what if there's some penalty for the graphics chip accessing the XDR? Then there's the issue of the Xbox 360's "smart EDRAM,"--10MB of embedded RAM that stores the back buffer, z-buffer, and stencil buffer. It performs a bunch of useful blending and z-compare functions right within the RAM so that the graphics chip doesn't have to. It's also got an insane 256 GB/sec of bandwidth to the GPU. The PS3's graphics chip is a traditional PC architecture, with separate and discrete vertex and pixel pipelines. The 360 uses a new unified shader architecture with 48 ALUs, each capable of performing multiple shader operations per cycle and fed by a scheduler intended to keep them fully utilized all the time. So when it comes to graphics, it's simply not possible to tell, simply by looking at the numbers, which one is more "powerful."

The CPUs are a similar story. The 360 uses a three-core PowerPC derivative with a shared L2 cache, where each core can process two simultaneous threads. It's got SIMD floating-point units similar to SSE3, and is certainly no number-crunching slouch. The PS3 uses the Cell, which has a single PowerPC core and eight "synergistic processing units" optimized for particular tasks. What these SPUs are good at and what they're not is a point of contention, even among programmers working on it, because it's no small task to get the Cell CPU firing on all cylinders, if you will. Is it more powerful than the 360's CPU? Well, it certainly has a higher peak "gigaflops" rating, but that doesn't really mean it's going to be better for games. Games use operations like load/store, and perform logic functions that need to access memory in a random, rather than streaming, fashion. That's not necessarily the Cell chip's forte. As with the GPUs, the raw numbers don't tell the whole story here. It's simply not possible to look at the specs and say "this one is more powerful for what game developers need to do."

Right now, there are no developers working on kits that include hardware resembling the true final production Xbox 360 or Playstation 3 systems. It's not until they start banging on that final silicon that we'll really know which is more powerful, and I suspect there will still be disagreement. At the end of the day, in a general sense, I think both systems are going to be able to get about the same quality stuff on the screen.

So if hardware horsepower won't determine the winner, what will? It will come down to the following criteria:

Manufacturability
Demand will outstrip supply at first. Which system can be more easily manufactured in large quantities, have production scale up more quickly, and ramp down the price more easily? I think Microsoft has an advantage here. It's using somewhat smaller chips, and less exotic components (like XDR RAM and Blu-ray optical storage). Still, Microsoft is building in a hard drive, so that's one more thing to worry about.

Development ease
Which platform makes it easier, faster, and less costly for publishers to bring out impressive next-generation titles? With neither system shipping final dev kits, this one is still up in the air. Nvidia's involvement with the PS3 is a big boost in this area and will make the graphics side of game programming much easier on PS3 developers than it was on the PS2.

Business deals
Who's greasing the right palms and working the deals necessary to secure gotta-have-it games exclusively for its system? Who's going to grab the next Grand Theft Auto, or the next GTA-like phenom? It's hard to pick a winner here. Both companies have very deep pockets and good relationships with lots of third-party publishers.

First-party titles
One way to get exclusives is to make them yourself. Will Microsoft Game Studios' stable of franchises beat out Sony Computer Entertainment's? You used to be able to hand this one to Sony, but Microsoft has built some well respected and popular franchises in the last few years, it's finally delivering games from premiere developer Rare, and has a little something called Halo.

Services and non-gaming features
Both systems are flaunting online play, online community features, and media features. Sony has a lot of catching up to do to compete with the universally praised Xbox Live, and Microsoft is able to show off some really impressive services and media features for the 360 already. That they can't demo any of this stuff for the Playstation 3 speaks volumes about how much further ahead Microsoft is on the development of this side of the system. Will the media capabilities of a particular console work with your devices and with the file types and formats of all your data?

Right now, it's too early to call a "winner." Each system has its strengths and weaknesses. Fortunately, you don't have to decide which one to buy just yet. Prices and exact release dates haven't even been announced yet. You can wait a few more months, and read more about the games as they move from development to final hardware, before you make a buying decision.


Is PC Gaming Doomed?

No, it's going to be just fine. There were plenty of great PC games on display at the show. Alan Wake is definitely intriguing, and gorgeous. Battlefield 2 should come out real soon, and it will change PC multiplayer play for the better. Call of Duty 2 looked just as good on the PC as it did on the Xbox 360, and you don't need a high-definition TV to truly appreciate it. Plus, it's the kind of shooter we'd all rather play with a mouse and keyboard. The Elder Scrolls: Oblivion may be the prettiest, most interactive and reactive role playing game ever. A group of guys from Blizzard left to make their own company, and they're bringing PC gamers a sort of first-person shooter version of Diablo that just reeks of fun with Hellgate: London. And what PC shooter fan isn't waiting for Serious Sam II?

It's really saying something when, for all the next-generation console talk, the "game of the show" buzz all centered on Will Wright's Spore, a PC-only title.

Perhaps the most exciting thing in PC gaming is something I don't even really know about. I ran into a senior exec from the Windows Graphics and Gaming Technologies group's Xbox 360 party. He told me that something really big is happening over there with regards to PC gaming. It's something that he's been pushing for a long time and is perhaps the main reason he joined Microsoft in the first place. It just got the full stamp of approval from Bill Gates and Steve Ballmer and other top-level executives, so he was really excited. He couldn't tell me what it is, but it sounds big. I would guess that it has something to do with Longhorn, but we probably won't hear anything for months.

All those next-gen consoles sure sound like they have some ripping hardware, and they do, but PCs aren't going to be left in the dust. Nvidia and ATI both had next-gen graphics cards working at the show, even though they weren't really unveiling specifics just yet. In fact, the Alan Wake demo at ATI's booth was running on its next-gen card, the R520. By the time the Xbox 360 comes out, we'll have these next-gen graphics cards. By the time the PlayStation 3 is 2 years old, hardly halfway through its life cycle, we'll have graphics cards 4 to 6 times as powerful, and CPUs 2 to 3 times, not to mention four times the RAM on standard PCs.

Let's not forget about new hardware innovations like the physics chip from AGEIA. Cards will hit the market late this year in the $250 to $300 price range, and will only become more affordable from there. One of those cards can deliver physics simulations far greater than any next-gen console.

As Microsoft drives toward Longhorn next year, I think we'll see a greater push toward PC games. Next year, when Longhorn is on the near horizon and PS3 and Xbox 360 launches are behind us, I bet the PC will be a particularly strong platform.

http://www.extremetech.com/article2/0,1558,1820466,00.asp

there are pages to this article before, and after, the stuff posted above, but this was the meat of it.
 
I think everything they say is pretty obvious however the 256GB/sec is not across the GPU it is across the logic functions of the Edram.
 
some common-sense there.

Still not sure about the CPU stuff though. Yes, games use general purpose code, but overall housekeeping should be a much smaller percentage of overall CPU time next-gen, especially if CELL is doing vector stuff.

One interesting thing he says about RSX having discrete Vector and Pixel shader pipes. Does that mean the '136 shader ops per cycle' includes vertex shading? How does that affect previous comparisons with Xenos?
 
It brings up a good point: Either system can be made to look more powerful than the other, depending on how you want to bend the specs.

So when it comes to graphics, it's simply not possible to tell, simply by looking at the numbers, which one is more "powerful."

It's simply not possible to look at the specs and say "this one is more powerful for what game developers need to do."

At the end of the day, in a general sense, I think both systems are going to be able to get about the same quality stuff on the screen.

He basically says four different times the systems are about even. This thread is doomed.
 
No, he guesses once that they are about even, and says three times that you can skew the argument in your favour based on manipulating the information available.

As you have just proven nicely :)
 
mrklaw said:
One interesting thing he says about RSX having discrete Vector and Pixel shader pipes. Does that mean the '136 shader ops per cycle' includes vertex shading? How does that affect previous comparisons with Xenos?

It doesn't affect previous comparisons. How would it? Xenos uses its unified ALUs for vertex and pixel shading, so it makes sense to compare to RSX, vertex and pixel shading combined.

A lot of previous comparisons have been flawed, imo, however.
 
mrklaw said:
some common-sense there.

Still not sure about the CPU stuff though. Yes, games use general purpose code, but overall housekeeping should be a much smaller percentage of overall CPU time next-gen, especially if CELL is doing vector stuff.

One interesting thing he says about RSX having discrete Vector and Pixel shader pipes. Does that mean the '136 shader ops per cycle' includes vertex shading?

Yes.

mrklaw said:
How does that affect previous comparisons with Xenos?

Link
 
It depends what you are comparing.

Previous 96 Vs 136 seem to favour the RSX.

But if you are looking at pixel shading only, what would the numbers be? Xenos could use all pipes for pixel shading so stay at 96. What would the RSX number be?

I guess what I'm asking is - do we know how many pixel shader ops the RSX can do yet?
 
mrklaw said:
But if you are looking at pixel shading only, what would the numbers be? Xenos could use all pipes for pixel shading so stay at 96. What would the RSX number be?

In such comparisons, Xenos would win, but it's pointless. Xenos will never be able to do all pixel shading or all vertex shading at once. This is where the seeds of that stupid "1.9Tflop" Xenos claim came from. It has to split its power between vertex and pixel shading, just as RSX does - the only difference is that in the latter's case, the split is made for you. The only fair comparison of power is of total power.

mrklaw said:
I guess what I'm asking is - do we know how many pixel shader ops the RSX can do yet?

No.
 
Indulge me for a moment :)

How unlikely is it that Xenos will do pixel shading only? Isn't it feasible that the XeCPU can do all vertex transformations, and leave the GPU to do pixels only? Not all games will have crazy ass physics and AI.

Its even more likely with CELL IMO. Those SPEs are designed for vector processing. A pixel shader only RSX would be the perfect partner IMO, and I don't know why it has vertex shaders at all, unless its a hangover from its PC roots?
 
mrklaw said:
How unlikely is it that Xenos will do pixel shading only? Isn't it feasible that the XeCPU can do all vertex transformations, and leave the GPU to do pixels only? Not all games will have crazy ass physics and AI.

Trust me, XeCPU needs all it has for game stuff. You could maybe set aside one core for vertex shading, but the throughput would be low. Maybe you'll have one or two games in the system's entire lifespan that take such an approach, but in general it's highly unlikely anyone will do this.

There's also the question of texturing logic on the GPU - I'm not at all sure, but not every ALU will have access to it at all times (there's fewer texture samplers than there are ALUs for example). That could make using the whole chip for pixel stuff problematic (?)
 
OK, I simply don't believe (or understand) this, so I'll just have to sit out and see what happens.

Basic game logic won't increase very much. Physics will, as will AI, but some games will stick to standard rag-doll/rigid body stuff, and simplish AI. Say a few times more complex than current games.

But these CPUs are crazy powerful compared to before. I just cannot bridge the gap, considering that previous generations ahve been all about the games (hey, pretty games sell, right?) - that companies like Sony would spend billions of dollars on something that makes people fall down stairs nicely.
 
$300 for a physics card? :lol

Oh, and PC gaming is never going to die, just like Nintendo will never go third party.
 
Marconelly said:
Wasn't it 0.9Tflop (at least that's what ATI and MS said) Remember 0.9 (GPU) + 0.1 (CPU) = 1TF?

There was a popular claim that because Xenos could use all its ALUs for either vertex or pixel shading, you could count them twice ;)
 
There was a popular claim that because Xenos could use all its ALUs for either vertex or pixel shading, you could count them twice ;)
Don't forget that because 4xAA is free, you can count that number 4 times, Xenos 32bit color runs 2x faster then 64bit, so another *2 for that, because of 256GB bandwith their efficiency is at least 8x higher...
So in summary, the total count is actually 64+TFlops!!!!!!! :D
 
I think that's a pretty good article and echoes my thoughts on the situation. I know you die hards are never going to let the "tech war" die... but I'm confident that 360 devs will be able to crank out games just as beautiful as PS3 devs.
 
Slo said:
$300 for a physics card? :lol

A dedicated physics card is going to be a hard sell I think. It's not like when the first 3d cards came out and you could look at a game and think "wow, it doesn't look like shit any more". It's a much less tangible benefit.
 
I think that's a pretty good article and echoes my thoughts on the situation. I know you die hards are never going to let the "tech war" die... but I'm confident that 360 devs will be able to crank out games just as beautiful as PS3 devs.
I don't really doubt that, but I'm curious just which one wil have more performance in general. For example, whether U3 engine, or Renderware engine games will perform any better or one or another (when I say perform, I think about framerates mostly). If the games look identical but run at more stable framerate or one or another (or even 30FPS vs. 60FPS) that's an advantage right there.

Also, it will be interesting seeing whether Xenos will be anle to compete in things RSX seems to be excelling in, like HDR support, or the shader complexity that gives effects such as subsurface scattering and volumetric lighting and effects, or conversely whether RSX will be able to hold it's own in things where Xenos seems to be excelling, like FSAA.

For HDR support for example, we know that Xenos defaults to a kinda cripled 32bit solution, but is it faster than RSX's 64bit or 128bit solution? Xenos maybe even can do 64bit frame blending for HDR, but how fast and efficient that is?
 
I agree with Marconelly and Jedi.

I love the tech wars, and don't really care who wins (I'll have all three machines anyway). I just love all the ins and outs of it.
 
A lot of this stuff is over my head but I think the author may be getting a little ahead of himself when he says CPU's will be more powerful in gaming in 2 years than Cell, or even Xbox 360 for that matter. Perhaps in theoretical performance, hey maybe that dual core FX-57 equivalent will be out by then (minus $1000 to your wallet). But will it actually be applied in games? My hunch is no. Whereas in console world devs are going to be forced into multi-threaded programming, the PC world is much more wide open. It'll be interesting to see how that divide works itself out.

Man, $250+ for a PPU? The bang for buck for consoles next gen is going to make PC's look downright quaint.
 
It's really saying something when, for all the next-generation console talk, the "game of the show" buzz all centered on Will Wright's Spore, a PC-only title


Seems like he misspelled Zelda: Twilight Princess.

Damn typos.
 
mrklaw said:
I agree with Marconelly and Jedi.

I love the tech wars, and don't really care who wins (I'll have all three machines anyway). I just love all the ins and outs of it.
Yes, the ins and outs is fun. The developer working on Heavenly Swords actually says that the Sony and MS teams are actually impressed with each others work, its the fanboys who really get into these pissing contests.

I will say power is important but the difference has to be clearly noticeable.
 
Top Bottom