ps2 and ngc technical details

[Simon F]

According to a previous poster, the best achieved on PS2 was typically 7.5MP/s. The 10MP/s on N2 was a measured performance rate (the peak figure is higher

The PA tends to only measured polys that generate valid pixels - I'm curious whether the N2 number is measuring pre T&L triangles or post.

Wouldn't that be because PS2 doesn't have backface culling HW and so it's cheaper to render everything?

I suspect that the Elan figure would be pre-T&L because I think Elan had a flag to do BF culling if requested. Don't quote me though, it's been a long time.

Both flipper and N2's elan ( unless I'm mistaken Simon ) probally have a lot processing power than VU1 on the PS2, but the programable nature can allow you to get a lot more bang for your buck.
There's a lot of room in terms of almost quick and dirty subdivision schemes to lesson the poor look of vertex compared against pixel lighting, and I'm sure that if AM2 had spent more time on the PS2 version of VF4 they may have increased the visual quality. ( It's normally true that as a programmer you can think of a least half a dozen major improvements to any project once it's actually out )

My assumption was that the N2 version was already highly tessellated and that most of the effects would have been done with per-vertex calculations. It always seemed to me that Sega liked vertex operations as opposed to texture-effects (as used, say, by "id").

 

[Phil]

While certainly a great game, only Sega knows why the PS2 version suffered and if there is left for improvement..

According to the grapevine (and this is a completely unofficial statement), Sega's own opinion was that N2 was far more powerful (in the practical sense) than the PS2.

Well, everyone's entitled to their opinion. I say opinion simply because "powerful" is quite ambiguous and boils down to developers personal preference. We all have visions and naturally, some hardware will be better suited than others to accomplish those. Not at all trying to discredit the N2 by the way - it's just that the two platforms are so different by nature, any comparasment is damned to fail on the fact that there's no way to messure two different things and it boils down to personal preference and bias in the end. Something, we're all a victim of.

 

[Simon F]

I'm just giving the _rumoured_ opinion of a developer with considerable experience.

 

[London Geezer]

I remember someone else saying something along those lines....



























Cough*Lorne Lanning*Cough...

 

[V3]

My assumption was that the N2 version was already highly tessellated and that most of the effects would have been done with per-vertex calculations.

VF4 characters might be high in polygon count, but it can use like at least 5-10 times higher amount of tessellation at times.

On the characters, the vertex lighting flawed due to lack of tesselation, was apparent in N2 VF4.

It always seemed to me that Sega liked vertex operations as opposed to texture-effects (as used, say, by "id").

As a consumer, I must say I prefer the typical Sega arcade looks, than than what "id" Doom3 looks. Doom3 looks, how can I put it, "boxy".

The thing that Sega need, is to have control over their tesselation, so they can get the correct tesselation level at every LOD. They have alot of camera work in their game, its really a shame to be destroyed by poor LOD.

Another thing is improving how the skin deformed around the joint, which is really really bad in N2 VF4. It looks painful at times. Perhaps Naomi3 can give them enough registers and instruction slots and speed to implement something like blendshape or other techniques that can solve it, if simple weighted skin can't solve the problem.

 

[overclocked]

This comes to my mind when talking "power"...
RebelSrike on GC is the best tech-demo-game IMO.
This debate is soo weird but it never ends...

Only conclusion i had first was a teen not afford to buy all systems and going rack about the thing he can´t get. Now when i see grown up´s being mean for something that should be fun, sure i love tech but it´s gone too far now.

Thats my opinion. 8)

 

[Lazy8s]

As evidenced by this thread, it's not easy to count lights just by looking at the screen. Even Dead or Alive 2 on DC used two infinite and two point lights together for its cinematics.

Polygon counts in VF4 were a lot higher in the arcade version - double the amount for the characters, and fully modeled backgrounds as opposed to bitmaps for some distant objects in the PS2 version - where its per-vertex lighting would see great results with the finer tesselation.

 

[PC-Engine]

archie has stated that N2 is on par with Chihiro aka Xbox arcade board.

 

[archie4oz]

That was *roughly* on par (mainly becuase of the large amount of memory), and that doesn't reflect the strengths and weaknesses of either platform. Also that's not *my* experience but just the impression my old university roomie (works at SmileBit) gave me...

 

[Fafalada]

Wouldn't that be because PS2 doesn't have backface culling HW and so it's cheaper to render everything?

It's cheaper if all you render is the most basic model - single textured polys with 3 lights or less.
Anything more complex can usually run faster (both rasterizing and VU shader itself) if you bf cull.
So if I was doing something as perverted(and generally not necessary) as 6 local lights per vertex, I'd definately work-in culling to speed things up.

Anyway, I personally find the comparison not very relevant unless you're porting from one hw to another. Using a programmable model you'd pretty much never want to do exact same processing as a fixed unit - that basically defeats the whole purpose of having programmable shaders.

 

[Simon F]

Wouldn't that be because PS2 doesn't have backface culling HW and so it's cheaper to render everything?

It's cheaper if all you render is the most basic model - single textured polys with 3 lights or less.
Anything more complex can usually run faster (both rasterizing and VU shader itself) if you bf cull.

Thanks Faf',
I was curious as to where the breakeven point would be.

 

[Fafalada]

Speaking of lights Simon, can you give a bit more detail on the local light model used in the 6light@10Mpp/s benchmark?
I'm also curious what types of lights did Elan support, Flipper has two for instance (infinite with specular, or local omni).

 

[Simon F]

Speaking of lights Simon, can you give a bit more detail on the local light model used in the 6light@10Mpp/s benchmark?
I'm also curious what types of lights did Elan support, Flipper has two for instance (infinite with specular, or local omni).

Describing them as "Lights" is actually selling them a bit short. They were general vertex shading/colouring modifiers and so could also mess around with the other shading parameters (eg fog/translucency). For example, you could have a beam of, err, "magic light" that made things slightly transparent if you wanted to, or you could set up clouds of fog. In a sense, a "light" was a big complex instruction.

I think it took several pages to describe all the parameters for the lights so I'll have to be very brief.... and besides... I can't remember all the details

All lights could compute diffuse and specular (for free). There were directional (i.e. parallel) and spot lights (both cost the same). 6 lights automatically came for free (as the processing got hidden by the time taken by something else). Spots had a position, principal lighting direction, distance attenuation (parameterisable) and angular (relative to principal lighting direction) attenuation. I seem to recall you could set them up to drive the dot product bump mapping (ie. map global light direction to local surface coordinates).

Is that detailed enough?

 

[Lazy8s]

Simon F:

They were general vertex shading/colouring modifiers and so could also mess around with the other shading parameters (eg fog/translucency).

Nice. That kind of range for functions makes a nice base for lots of effects.

 

[PC-Engine]

Simon F:

They were general vertex shading/colouring modifiers and so could also mess around with the other shading parameters (eg fog/translucency).

Nice. That kind of range for functions makes a nice base for lots of effects.

Yeah it's really a shame SEGA didn't develop more games on N2.

 

[London Geezer]

Simon F:

They were general vertex shading/colouring modifiers and so could also mess around with the other shading parameters (eg fog/translucency).

Nice. That kind of range for functions makes a nice base for lots of effects.

Yeah it's really a shame SEGA didn't develop more games on N2.

How many games actually run on N2 anyway? I can't think of any, other than VF4... Now that they have Chihiro, i guess they focused on that.

Can't wait for N3

 

[PC-Engine]

AFAIK they had LA Riders or some CELL shaded motorcycle game and Club Kart which is a go kart racing game.

 

[Fafalada]

Is that detailed enough?

Yes that was great, thanks
Anyway, it's a bit similar to GCs 'lights'(they are also used as general modifiers, including BM setup) but a fair bit more fully featured.

 

[Farid]

How many games actually run on N2 anyway? I can't think of any, other than VF4... Now that they have Chihiro, i guess they focused on that.

Can't wait for N3

I know that king of Route 66 use it, as PC-engine said there's a Cel-shaded moto racer game on it , Virtua Striker 3 too is on naomi 2... that's all i' remember.

Edit: System 16 is here for this kind of question.

Club Kart
Club Kart Prize
King Of Route 66
Jet Squadron
Quest Of D
Sega Driving Simulator
Soul Surfer Virtua Fighter 4
Virtua Fighter 4 Evolution
Virtua Fighter 4 Evolution Ver.B
Virtua Fighter 4 Final Tuned
Virtua Striker III
Wild Riders