ps2 and ngc technical details

[Teasy]

AFAIK the game cube outputs the least amount of polygons when it's stressed.

I assume by stressed you must mean loads of effects (lighting, textures, shading ect). In which case why do you think that? Earlier you were talking about theoretical T&L performance (it being a well known fact that Flipper is slower theortically at pushing polys). Which really equates to poly pushing when not stressed. So you haven't mentioned why you think its the weakest when stressed.

I agree again. I didn't claim it could push twice as many polys as the gamecube, but from what I've seen PS2 can beat the gamecube in raw poly throughput. Although raw poly's don't make a complete game, but I was just stating that one point. That's all I was saying...

No you didn't say that directly. But you did ask me why I thought Nick's numbers were incorrect. Which I then took to mean that you agreed with his numbers. Those numbers suggested that PS2 could push twice as many polys as GC in game and XBox could push three times as many as GC in game. Which is what I originally disagreed with.

PS2 can beat GC in raw polygons, I agree absolutely, since that is a technical fact from its specs. But I was talking about acheivable output in games. Aren't you also talking about acheivable output in games when you say stressed (as in lots of effects)? Or when you said stressed did you just mean when someone tries to stress the T&L (as in tries to pump as many polys as possible without many effects)? In which case we're arguing about two different things.

 

[Guden Oden]

Not only that, they benchmarked it using game code for the most part, so I really don't see how it could be rubbish

Q,

You KNOW EA coders have never been considered top-notch EVER. When was the last time they released a heavily optimized, top-notch title done by their in-house coding teams, hm?

They're a company of jack-of-all-trades, good at porting games from one platform to the other but masters of none. That they "benchmarked" the 3 current consoles against each other and came up with some numbers doesn't mean at all those numbers are representative of anything, ESPECIALLY as they used their own half-assed, most likely ported game code.

That the cube is a weak T&Ler strikes me as somewhat surprising a conclusion. Fire up Metroid Prime and run to any of the larger rooms in the game. Stand in a corner and select the default game and start to hammer away at the fire button so that your shots sweep along the floor/a wall. Note they light up their surroundings. Note how the game still runs at 60fps with this many lights flying around.

Have you seen anything similar on the other consoles?

Yeah, we know the GC doesn't deliver top-notch peak figures, it is not an arab fullblood like the xbox; fast yet quirky and temperamental, it is an ardenner that can pull heavy loads at a steady pace. It's been posted here by devs that it's possible to get pretty decent performance out of the thing even when deliberately doing stupid things when coding it, thanks to its forgiving large-cached and low-latency design.

In perspective, looking back at that lying Mike Abrash presentation that tried to explain how the box was supposed to be able to sustain virtually maxed-out poly rates with bandwidth to spare, one can only laugh at how wrong it was.

 

[Teasy]

Phil

Yeah PS2 can push twice as many polys as GC, but GC could also push twice as many as PS2, it depends on the game/demo you look at What I'm talking about is the general acheivable poly performance in games. For instance what would you say if someone asked you what each console could acheive in games as far as polys are concerned. Would you agree with 12m, 22m, 30m (GC,PS2,XBox)?.. I definitely wouldn't.

 

[Phil]

Teasy:

PS2 can beat GC in raw polygons, I agree absolutely, since that is a technical fact from its specs. But I was talking about acheivable output in games. Aren't you also talking about acheivable output in games when you say stressed (as in lots of effects)? Or when you said stressed did you just mean when someone tries to stress the T&L (as in tries to pump as many polys as possible without many effects)?

How is beating GC in raw polygons any different than "in-game" achievable output in games?

Is a demo pushing close to 30 million polygons any less credible than a game doing half as much? When does a piece of software actually fullfill the requirements that it can be called a game? Perhaps you mean this because a demo itself is only semi-interactive and has no [ insert game elements, i.e. AI ]? The thing is, as soon as you actually start to compare games themselves, an actual comparasment becomes nearly impossible because there are so many factors to consider:

- how efficiantly is the engine using the hardware?
- how much performance is left?
- slowdown?
- amount of textures / geometry
- AI complexity
- physics complexity
- image quality
- pixel effects
- art-direction or raw performance?

With all this in mind, it becomes impossible to actually argue the strengths of one console over the other. Why? Because every console has it's strengths and comparing games begs the question what the game is pushing and that in itself needs to be analysed to be used in actual comparasments. Without the source itself, this becomes next to impossible since we can't imagine what the engine itself is actually doing to make things look the way they do on screen. We can only imagine. Speculate. Let our biasness flaw our judgement.

In the end, demo's even make comparasments easier because they are not as flexible, yet are in realtime and half of the factors above are non-issues because the demo probably isn't even doing that (for example: AI).

Games are very subjective. A PS2 game could potentially be done using raw-polygons with little to no texturing (REZ look?) or single texturing and under given circumstances could achieve a polygon throughoutput that would double that what would be possible on GCN. On the other hand, a different developer could make a game on GCN emphasizing on different strengths doing a game with more texture layers and get easily to a scenario in which the GCN will be pushing as many polygons or even more. Given this fact that there are so many possibilities, I ask you, is it worth it even comparing in-game achievable polygons numbers? In-game itself is just a term and could apply to a demo aswell (unless of course, you give certain requirements like AI, physics etc). The point is, demos are a much better representation on what a console can do or can't (of course not to the end-user) - or even better, theoretical specs, as they are what the console should be able to achieve asuming maximum efficiency. In the end though, even a demo is a victim of the sheer infinite amount of possibilities a developer can take to emphasize on effects to maximize his little application.

EDIT: Just saw your reply teasy, thought you might have missed it. I will still post this reply as I think it's still quite relevant and perhaps brings my point better across.

 

[Qroach]

I assume by stressed you must mean loads of effects (lighting, textures, shading ect).

No, stressed as in pushing high amounts of polygons.

No you didn't say that directly. But you did ask me why I thought Nick's numbers were incorrect. Which I then took to mean that you agreed with his numbers. Those numbers suggested that PS2 could push twice as many polys as GC in game and XBox could push three times as many as GC in game. Which is what I originally disagreed with.

Actually I asked you what "led to your conclusion", but that didnt' mean I agreed with his numbers posted. I also made note in the same post that I recalled seeing some sort of statement regarding fully featured poly's being around 12 million pps, however I don't remember how many textures and what fully featured really means. For isntance is the consoles are pushing two textures per poly I wouldn't be suprised to see PS2 get whooped by the gamecube.

The only thing I claimed as a fact is that flipper isn't as fast as PS2 when strictly pushing polys.

 

[Qroach]

Guden,

You KNOW EA coders have never been considered top-notch EVER. When was the last time they released a heavily optimized, top-notch title done by their in-house coding teams, hm?

Ok then, do you have a better way of answering the which is faster question? If you can't trust what a big developer like EA can do (and they do have quite the R&D department there) then who will you trust? Yes they are a multiplatform developer and could be considered a jack of all trades, but at the same time what developer goes crazy trying to find the perfect optimizations when they are developing on all 3 platforms?

That's sorta like saying (and I shudder) that 3dmark shouldn't be used as a hardware benchmark because it's not optimized for every video card out there.

In perspective, looking back at that lying Mike Abrash presentation that tried to explain how the box was supposed to be able to sustain virtually maxed-out poly rates with bandwidth to spare, one can only laugh at how wrong it was.

Um, you need to read mike abrash's articles again. He never claimed anyone would be able to sustain the peak throughput with xbox hardware... You also never saw more of his articles that actually proved how much you could push the xbox hardware in reality. he came pretty darn close to the performance numbers he through out there.

 

[Teasy]

Ok so nobody here actually disagree's on the GC/XBox/PS2 polygon pushing thing then? (appart from maybe me and Nick Laslett ). Hmm this is unusual...

 

[Ooh-videogames]

I haven't really seen confirmed raw numbers for GC. I've only read theories on it but the number was 90mpps, that was from IGN. Factor 5 numbers for Rogue Leader were 15mpps, but thats another story. Many devs have said the numbers Nintendo issued are conservative, and numbers acheiveable are greater. I could only imagine how many polygons RE4 is pushing. The only tech demos I remember from the unveiling of the GC were, Link and Ganon, Mario128, and the car demo.

 

[Lazy8s]

What EA's games "look" like has little to do with whether the company can competently test and measure each machine for a specific output. Accomplishment in visuals for programming comes mostly from its nature as a game - from putting the focus on a design's strengths (LOD management being important) and off its weaknesses (for example, using obscured viewfields but not drawing attention to that fact) - and wouldn't factor in much for a raw benchmark. It's more about how those computational resources are used than whether or not a developer has access to them, especially regarding companies with extensive dev resources like EA.

Any competent developer would achieve roughly the same results from a benchmark, but their skill at crafting those resources into something that looks pretty will vary.

 

[Phil]

Ok so nobody here actually disagree's on the GC/XBox/PS2 polygon pushing thing then? (appart from maybe me and Nick Laslett ). Hmm this is unusual...

Sorry Teasy... Perhaps if it helps, I was indirectly even agreeing with you on those numbers to a certain extend: those numbers are just peak numbers and at that quite empty as we don't know under which circumstances they were achieved. I was more disagreeing or questioning the stance of your argument on in-game performance or demo - basically pointing out that neither is a definite source for a comparasment. If anything really, theoretical performance figures is the best we could possibly use and at the same time the most misleading since we as "end-users" will never see those 'peaks' in actual games.

 

[Squeak]

Isn't it true that hardwired TnL will almost always be faster, than a more flexible hardware with the same transistor count?
GC arguably has the fastest core CPU of the three, so it should have spare resources for a bit of custom TnL.
Looking at games from all platforms, I would say that in a typical game situation they are very equal when it comes to geometry.
Sure, VU1 can process scary amounts of vertices if not much else is going on (like Haven Forest in Jak II), but in a game with fair use of CPU they look very equal to me.

 

[cthellis42]

Ok so nobody here actually disagree's on the GC/XBox/PS2 polygon pushing thing then? (appart from maybe me and Nick Laslett ). Hmm this is unusual...

I'm mainly waiting for Nick to expound on his statement, since arguing the point doesn't do much until we know what precise point he was bringing up. There's always a lot of interpretational sway.

 

[Fox5]

Panjev,

Qroach, that T&L unit in Flipper is also quite good at Lighting as the number of polygons per seconds drops by a much slower rate than it does on other consoles if you read those performance figures.

Well that's true, but that doesn't mean it still isn't the weakest of the three regardless of how fast performance drops... it still doesn't output as many polygons as PS2 or xbox.

The best comparrison of all three hardware platforms was provided by EA. If someone could find that article cmoparing all three, I think it would shed some light on the question.

I remember an article from EA canada(they had been benchmarking a soccer game or something on the 3 systems) and it had the gamecube as the most powerful. It was just a cinema scene, no ai, and ps2 was at I believe 20 million pps(not sure about texture layers), xbox was 25 with 4 textures layers, and gamecube was 25 with 8. However, I believe there was another branch of EA that had completely different numbers.(maybe those EA canada numbers are achievable if they devote the cpu to graphics?)

 

[Li Mu Bai]

Okay, let's all agree that the PS2 is the weakest feature set wise of the 3. No DOT3, EMBM, pixel shader, (TEV) or dedicated vertex shader. (NV2A) This makes a very noticeable difference when applied to games.

 

[Fox5]

Okay, let's all agree that the PS2 is the weakest feature set wise of the 3. No DOT3, EMBM, pixel shader, (TEV) or dedicated vertex shader. (NV2A) This makes a very noticeable difference when applied to games.

But doesn't the ps2 have the greatest raw power, but it wastes it the fastest?

 

[Farid]

It was just a cinema scene, no ai, and ps2 was at I believe 20 million pps(not sure about texture layers), xbox was 25 with 4 textures layers, and gamecube was 25 with 8.

8 Textures layers for the Gamecube and 25 Mpps...
If you could find a link to that article Fox5, that will be appreciated.
8 layers + 25Mpps sounds overkill for me.

BTW EA has done a lot of "crappy" port to NGC, the Need for Speed serie is the first example to come to my mind, NFS: Hot Pursuit 2 was HORRIBLE on NGC (bad graphics, slowdowns (a lot of them)), NFS: Underground had less slowdown but it graphics were inferior to the PS2 version.

OTOH, we've all heard the Nintendo statement that NGC was pushing 9/12 Millions full featured polygons, the question that always bugged me was what do they called "Full Featured polygons" ?
Did they ever explain theirselves about that?

Edit: (before posting )

Real-world polygon 6 million to 12 million polygons/second (Peak) (Assuming actual game conditions with complex models, fully textured, fully lit, etc.)

So full featured polygons means complex models (What do they mean by complex models? Complex model usually means "detailed model" wich is bound to the polycount, maybe they meant "animated model with a skeleton"...), + a light, + a texture, the etcetera is maybe for a simple AI.
So the Nintendo "Full featured polygons" are the classical "in-game like" figures, polys + light + textures.

 

[PC-Engine]

The ELAN chip is also fixed function and only capable of 10 million polys/sec raw. Now why was the lighting and poly numbers downgraded in the PS2 version of VF4 since it can do 20+ million polys/sec in-game while NAOMI 2 can only do 10 million in-game???

EE can do 66 million polys/sec with no lights.
ELAN can do 10 million polys/sec with no lights.
EE can do probably 5 million polys/sec (MAX) with 6 complex lights.
ELAN can do 10 million polys/sec with 6 complex lights.

Now some games would use more lights than others so (depending on the game) EE would push more/less polys than ELAN. Same situation with Flippper.

So in summary, PS2 can push more polys in a game like REZ or Battle Zone...

 

[Li Mu Bai]

Okay, let's all agree that the PS2 is the weakest feature set wise of the 3. No DOT3, EMBM, pixel shader, (TEV) or dedicated vertex shader. (NV2A) This makes a very noticeable difference when applied to games.

But doesn't the ps2 have the greatest raw power, but it wastes it the fastest?

Clarify raw? In terms of hypothetical raw polys? (which means nothing) It does have an advantage in fill-rate, but that would hardly qualify as dubbing it the system with the most "raw power." It simply cannot acheive effects that both the GC & Box can.

 

[Steve Dave Part Deux]

There are enough amazing looking GC games to convince me that ArtX spent their transistors in the right places.

 

[Fox5]

http://cube.ign.com/articles/088/088713p1.html

This must be the ea canada article, but apparently I was way off with the numbers....could have sworn I heard those numbers somewhere though, but Xbox and PS2 aren't even mentioned in this.

Gamecube development hardware running with eight texture effect layers + all other effects on: Approximately five million polygons per second
Gamecube development hardware running with four texture effect layers + all other effects on: Approximately 14 million polygons per second

Wording regarding the remaining benchmark tests was vague, but evidently the company also did experiments with the Gamecube development hardware running at least four hardware lights and other effects with impressive results of approximately 17 million polygons per second. Sources we spoke with said this is not only entirely possible, but highly conservative.

So all effects + 8 texture layers = 5 million polygons
All effects with only 4 texture layers = 14 million(wow, it takes a bit of a dive at 8..memory limited maybe?)
And 17 million under what probably could be in game conditions.
BTW, early ea games on gamecube were sometimes better than the ps2 version, sometimes not. Agent under fire was much better, and nightfire was better too.
Still pretty sure I remember an artificial graphical demo out of EA that did 25 million pps though....

http://gameztech.8m.com/consolewar1.htm
From here....

Also, developers have recently stated that the Gamecube can push more 20 million polygons per sec.

http://cube.ign.com/articles/094/094556p1.html
Here's a factor 5 article, just thought it was interesting how the only thing he noted gamecube had over xbox was memory bandwidth.(which means better textures) Memory access times too.

http://www.segatech.com/gamecube/overview/
Scroll down a bit and there's a blurb about factor 5 stating they could do 20 million polygons per second with all effects, with effects standing for texture layers and not actual effects.(umm...does gmaecube have a physical max for texture layers?) Maybe at its old clockspeed it could do 25?(but then it'd have a bottleneck in the cpu...)