• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Virtua Fighter 2 Coming To PS2

dark10x

Digital Foundry pixel pusher
DaCocoBrova said:
No, it was the hardware.

This has been discussed to death, so rather than argue, prove it. :p

Regardless of that fact, though, the point remains; current consoles can produce significantly better visuals than Model 3 at 60 fps.
 

DaCocoBrova

Finally bought a new PSP, but then pushed the demon onto someone else. Jesus.
I just touched base w/ sixtoe over at System16.

We will know soon enough. I'm not making this up though.
 

Lazy8s

The ghost of Dreamcast past
Shinobi:
If Saturn VF2 was pushing anymore then 60,000 polys per second, it was a miracle.
It was probably pushing more than that comfortably. 1,000 quads per frame? Saturn games were measured to be higher, especially those based off the later SGL distributed to developers that AM2 based off their VF2 work.
For PS2 VF4 to be similiar, it'd have to be pushing 3 million polys max. And I doubt the number was that small.
3 million is actually quite characteristic of PS2 performance, Sony's Performance Analyzer tests finding most games ranging from 2 to 5 mpps and the upper limit at 7.5 mpps. Namco revealed that Tekken Tag Tournament for PS2 averaged around 3 mpps, for reference.
Specs be damned though, I just use my eyes. Does PS2 VF4 look different to the arcade version? No doubt. But the differences between the arcade and Saturn versions of VF2 are so much more pronounced it isn't even funny. The fact that there isn't a single 3D element in Saturn VF2's backgrounds where everything in the foreground for the arcade version was 3D, coupled with the total lack of lighting, makes it look flat as a pancake in comparison. Same goes for Sega Rally as well.
Right, visual design, smart modeling, and diminishing returns go a long way.

I'm the first to say how last gen's consoles couldn't touch Model 2 graphics quality by a long shot despite misleading spec comparisons, but people tend to underestimate how much more T&L complexity the Naomi 2 sustains in real-world situations over PS2.

dark10x:
Perhaps many people would frown on the changes, but I think it would have been quite interesting to see AM2 take a different approach to rendering the stages.
That was a good rethinking of the visual design. And of course that goes both ways for any game when considering the possibilities on two different hardwares. Imagine PS2 MGS2 with metal surfaces that really looked like metal and highly stylized lighting to go along with its stylized art direction. If you go back and change the visual concept and design of a game, it really starts to become a different game - more like something that's ideal to the other hardware and similar to one of its showcase titles.

Marconelly:
God, not this Model 3 idolizing crap again. Please stop.
No, what he said has merit. There are areas where each hardware does have advantages over the other. Model 3 does some especially high quality effects with IQ and texturing, like AA actually used in games and never with a sacrifice to proscan.
Best looking games on Model 3 (like Super GT) look like pieces of poo when put against stuff like Burnout 3 on PS2.
The PS2 does a lot of stuff better, no argument. The biggest factor here, though, is that the craft of visual design has come a long way from the first Model 3 game like Super GT, released back in 1996, to a PS2 game that hasn't even released yet in 2004.
 

Shompola

Banned
DaCocoBrova said:
No, it was the hardware.

You mean there is some super sofisticated LOD Engine behind every Model 3 game that makes it sure that the framerate is 60fps no matter what the programmers and artists do?
There is no magic behind Model 3 buddy.
 

Lazy8s

The ghost of Dreamcast past
Edit on my last message on figures. Accidentally inverted.

Shompola:
You mean there is some super sofisticated LOD Engine behind every Model 3 game that makes it sure that the framerate is 60fps no matter what the programmers and artists do?
There is no magic behind Model 3 buddy.
A lot does come down to solid programming, true, yet the games were actually reported to employ some kind of framerate locking LOD system (which is something that can be done.)
 

MAZYORA

Member
DaCocoBrova said:
Wow. I really need to find this game in the arcade, because I'm very impressed w/ the PS2 version. I almost like not being able to compare the two though. One less thing to think about.

Yeah dude, the Naomi 2 version is so awesome graphically. It has by far the best lighting i've ever seen in a game.
 

dark10x

Digital Foundry pixel pusher
MAZYORA said:
Yeah dude, the Naomi 2 version is so awesome graphically. It has by far the best lighting i've ever seen in a game.

When it was released, I might agree...but it has been topped by a number of games.
 

Panajev2001a

GAF's Pleasant Genius
3 million is actually quite characteristic of PS2 performance, Sony's Performance Analyzer tests finding most games ranging from 2 to 5 mpps and the upper limit at 7.5 mpps. Namco revealed that Tekken Tag Tournament for PS2 averaged around 3 mpps, for reference.

Numbers were averages and not peaks and included only actually drawn triangles ( number of triangles T&L by the VUs is actually higher ).

Going by characters alone TTT would have hit almost 3 Mpps.
 

doncale

Banned
the Real3D/Pro-1000 GPU has the feature built into hardware where it maintains 30 fps or 60 fps (both are options). no matter what graphics features are turned on, even if all of them are on, resulting in 750,000 rectangle polygons per second sustained performance. not theoretical performance but realworld performance.

since Model 3 has two of these GPUs (and no seperate T&L like Naomi2 since T&L is built into the Real3D GPUs) the sustained performance of Model 3 is over 1 million rectangle polygons/sec or 2 million triangle polygons/sec

every single Model 3 game runs at 60fps. not because Sega programmers are so good (they are good) but because the hardware had the feature of locking everything down at 60fps.
 

dark10x

Digital Foundry pixel pusher
doncale said:
the Real3D/Pro-1000 GPU has the feature built into hardware where it maintains 30 fps or 60 fps (both are options). no matter what graphics features are turned on, even if all of them are on, resulting in 750,000 rectangle polygons per second

since Model 3 has two of these GPUs (and no seperate T&L like Naomi2 since T&L is built into the Real3D GPUs) the sustained performance of Model 3 is over 1 million rectangle polygons/sec or 2 million triangle polygons/sec

every single Model 3 game runs at 60fps. not because Sega programmers are so good (they are good) but because the hardware had the feature of locking everything down at 60fps.

I still believe that it has more to do with programming...
 

Lazy8s

The ghost of Dreamcast past
Panajev2001a:
Numbers were averages and not peaks and included only actually drawn triangles ( number of triangles T&L by the VUs is actually higher ).
Right, and quoting anything else (peaks, which by definition aren't characteristic, and redrawn triangles, which don't add to geometry) wouldn't have been representative.
 

drohne

hyperbolically metafictive
i suspect the polygons per second numbers tossed around on boards like these are grossly inflated. you can get far cry to display a polys per second counter, and that tends to hover around 3-5 million even in huge, detailed outdoor vistas. i suspect most console games push far fewer polys than far cry, and i doubt there are any games that push 4-6 times as many polys.
 

btrboyev

Member
I agree that ps2 is more powerful than model 3, but I think you sony supporters are missing the point completely. People are saying the ps2 probably couldn't do a perfect arcade port of some of the higher end model 3 games without some kind of modification..and for the most part I think thats true..the hardware works completely different and the ps2 image quality just wouldn't compare.

The reason model 3 racers don't look as good as current gen racers is due to lighting..if you took lighting out of games like burnout and GT they would look like crap. If there was dynamic lighting on say daytona 2 or Super GT they would no doubt look as good as any current game IMO.
 

dark10x

Digital Foundry pixel pusher
I agree that ps2 is more powerful than model 3, but I think you sony supporters are missing the point completely. People are saying the ps2 probably couldn't do a perfect arcade port of some of the higher end model 3 games without some kind of modification..and for the most part I think thats true..the hardware works completely different and the ps2 image quality just wouldn't compare.

Obviously, the game would have to be designed differently, but I feel that the PS2 is more than capable of displaying a flawless version of the game. Porting the game could be tough, due to the differences...but actually re-creating the game for PS2 hardware could produce extremely accurate (and likely superior) results.

Model 3 doesn't exactly have great image quality, though. Naomi was a HUGE step-up from Model 3 in terms of actual image quality. Burnout 3 runs in progressive scan mode on PS2, so it isn't as if the image quality couldn't compare...
 

TKM

Member
Right, and quoting anything else (peaks, which by definition aren't characteristic, and redrawn triangles, which don't add to geometry) wouldn't have been representative.

Then the same would apply to Model 3 and Naomi 2 which quote transform figures. It would be misleading to compare 3M PS2 to 1M Model 3 and 10M for N2.

donscale,

every single Model 3 game runs at 60fps. not because Sega programmers are so good (they are good) but because the hardware had the feature of locking everything down at 60fps.

Hmm, I wonder how that would work though. If it runs out of frame time, does the hardware just give up?
 

Shinobi

Member
dark10x said:
Yes indeed. I really do think people have simply forgotten what those Model 3 games look like.

How can that be, when I played the damn game on Tuesday?

Here's the deal with Model 2 and Model 3. They both represented the biggest technological and visual leaps in gaming by a country mile, and so completely and utterly destroyed everything that was out at the time they pretty much seared themselves into the conciousness of any Sega fan that played those games. Model 2 wasn't really superceded by any hardware until Model 3, and DC wasn't much more then a slightly enhanced Model 3 when it came out what, three years later.

There's also the artistic look to these games that add a ton to their visual profiles. Just a month ago I was looking at VF2 at the arcade and gazed in appreciation at how far ahead of it's time that game was. I can look at most of the Model 2 or 3 games without wincing. I can't really do that for most of the Voodoo-based stuff that came out back then. Not sure I can do that for Ridge Racer (Daytona cleaned it's clock visually), or any of the System 22 games (Rave Racer might've had a shot, never saw it though). And I sure as fuck can't do that with anything on the last generation of consoles (good riddance to those turds).

I've got a PS2 with GT3 and VF4 Evolution. I've got a XBox with Rallisport 2 and DOA3. No doubt these games blow away the Model 2 and 3 stuff technically, and any suggestion that the current trio of systems couldn't do perfect Model 2 or 3 translations is pure foolishness. But artistically the Model stuff was (for the most part) so well designed, that they have a hard time looking dated or nasty, unlike almost any pre-DC 3D game that came out. And they're all 60fps games, and even now they still look smooth as butter. That goes a long way in creating that suspension of belief when viewing a game.

Really, you give Super GT some real time lighting and shadows, and it isn't looking much worse then many of the racers out there. There's no discernable pop up. The backgrounds get the job done. The car models are recognizable both afar and up close, featuring drivers inside the cars and 3D rims. To say the game looks like a piece of poo is just nonsensical in my view. But, my view is all it is. Then again I never mistook House of the Dead for a Model 3 game, which was a pretty common mistake when that came out.




Lazy8s said:
Shinobi:

It was probably pushing more than that comfortably. 1,000 quads per frame? Saturn games were measured to be higher, especially those based off the later SGL distributed to developers that AM2 based off their VF2 work.

So WTF did all the polys go? The characters had like half the polys of the arcade version. No individual fingers were in sight. They were generally a lot leaner. And the arcade version obviously had better looking shadows as well. And again, not one poly present in the background (with the floor being the only possible exception). I don't



Lazy8s said:
3 million is actually quite characteristic of PS2 performance, Sony's Performance Analyzer tests finding most games ranging from 2 to 5 mpps and the upper limit at 7.5 mpps. Namco revealed that Tekken Tag Tournament for PS2 averaged around 3 mpps, for reference.

Sounds pretty low to me...unless DC games were stuck in the 1.5 million range (if that). Hard to compare poly counts anyway though, considering that each hardware does 'em differently.
 

doncale

Banned
I still believe that it has more to do with programming...

believe what you like. but in the case of Model 3, its hardware.

otherwise, you would have had a 60fps Shenmue, Shenmue II on Dreamcast and Xbox, instead of around 30fps.

in most cases, where the hardware does not have a 60fps lockdown feature, then it IS to do with the programming if a game is 60fps or any given framerate.


Shinobi: btw, Dreamcast (Nov 1998) came out roughly 2 years after the first Model 3 game, VF3 (fall 1996) came out.

Model 3 was basicly finished in late 1995, Dreamcast was finished in late 1997 or early 1998.

but anyway, I pretty much agree with most of your post. Model 2 games in the mid 1990s and Model 3 games in the late 1990s absolutely destroyed everything on the consumer market with the exception of a few (and i mean a FEW) Dreamcast games in 1998-1999.

I agree that Xbox and Gamecube could easily handle Model 3 games perfectly, but I dont know about PS2. its not a matter of power, because PS2 totally out performs Model 3 in terms of geometry & lighting, but NOT in texture mapping or image quality. so exact ports of Model 3 games on PS2 might be impossible. texture mapping and image quality (FSAA, trilinear filtering, mip mapping) is a huge part of what made Model 3 games so good. not the polygon count. and framerate. PS2 just does not have the same quality of graphic feature implementation that Model 3 had. but Xbox and Cube probably does. disclaimer: i own a PS2 :)
 

jarrod

Banned
Shinobi said:
Model 2 wasn't really superceded by any hardware until Model 3, and DC wasn't much more then a slightly enhanced Model 3 when it came out what, three years later.
It was 2 years, and DC/Naomi was actually quite a leap over Model 3. Better effects, lighting and 3 times the geometry.
 

doncale

Banned
Dreamcast/NAOMI actually has more like 50% more geometry than Model 3. not 3 times :)

Model 3 - 1 million rectangle polys or 2 million triangle polys
DC/NAOMI: 3 million triangle polys.

I forget the exact quote, but a high up Sega official said it takes 2 Dreamcast polygons to equal 1 Model 3 polygon.

on the other hand, if we take developer word for fact that Dreamcast can do more like 5-6 million polygons, then we get closer to 3x the geometry of Model 3 :)
 

Lazy8s

The ghost of Dreamcast past
TKM:
Then the same would apply to Model 3 and Naomi 2 which quote transform figures.
The comparison being discussed didn't use those transform specs to compare to the PS2's characteristic performance. It was simply brought up that AM2 had to remodel the characters for the VF4 conversion with about half the polygon budget (an example: from about 14k to 7k on Jerky, which interestingly puts it in the range of DOA2 on DC with its 2+ on-screen simultaneous characters) and had to replace some distant, modeled background objects with bitmaps.
It would be misleading to compare 3M PS2 to 1M Model 3 and 10M for N2.
Indeed.
 

Panajev2001a

GAF's Pleasant Genius
Doncale:

PlayStation 2 has much more fill-rate than Model 3, doesn't it ?

Spending fill-rate you can do tri-linear ( you are still left with 600 MPixels/s ) and you can give upload more textures to the GS ( you do not have to fill all the textures in the VRAM to render the current frame, you can keep swapping them in ).

You can do Vertical AA using the CRTC ( 640x448 to 640x224 and then it displays it: a flicker-free filter ) or you can downsample from a higher resolution.

As far as rendering features go, I am not aware that the Model 3 has particularly powerful and flexible color combiners ( or support for a DOT3 blending mode ).
 

Shompola

Banned
"believe what you like. but in the case of Model 3, its hardware."

Please explain how it works instead of saying it is. Are you talking about a simple vsync frame lock, LOD or what?
 

doncale

Banned
honestly I do not know how it actually is done. but the fact that Model 3 can lock down at 60fps was said by at least one Real3D engineer. I will try to find the quote.


Panajev, ok I will give PS2 the advantage of fillrate to reproduce Model 3 image quality by sheer brute force.

btw, Model 3 actually transforms 60 million polygons, raw. pretty close to PS2's 66 million.


edit: found something

http://www.textfiles.com/computers/DOCUMENTATION/r3dspec.txt

UPDATE RATE:

Basic system performance is specified as 30 Hz, but 60 Hz update
is also available.
TRANSPORT DELAY:

Transport delay for the R3D/PRO-1000 is measured from the time a
control input is received from the Application Host until the end
of the display of the first video field affected by that input. The
normal transport delay is 112 milliseconds for 30 Hz operation, or
56 ms for 60 Hz operation.

that might not answer all the questions. I'm looking for more info.

edit 2: more to backup the claim of sustained 60fps on Model 3 (Real3D Pro 1000)

http://66.218.71.225/search/cache?p...3d.html&w=real3d+pro+60+hz&d=89AEB2E5F1&icp=1

April 7, 1997

Menlo Park, California — Accom, Inc. (NASDAQ: ACMM) announced several revolutionary new virtual set product offerings at the National Association of Broadcasters convention which opened this morning in Las Vegas, NV. Accom’s new offerings include ELSET™ LIVE-NT, which operates on a standard PC running Windows NT and incorporates the REAL 3D® PRO-1000 Professional Series graphics engine. ELSET LIVE-NT enables video program producers to complete virtual set based programs in real time. Utilizing the REAL 3D PRO-1000 ensures reliable 60 Hz broadcast quality results.

Lance Kelson, VP of Virtual Studios for Accom stated, "Accom has led the transfer of virtual sets from interesting technological marvel to practical production tool. We now offer customers a series of ELSET virtual set tools to complete their productions using the solution which is most appropriate for their video content creation project."

ELSET LIVE-NT is one of the most revolutionary virtual set systems ever conceived. ELSET LIVE-NT is the live, on-air companion to ELSET POST-NT, which is an off-line, non-real time virtual set production system that operates on a PC running Windows NT. Broadcast-quality real time virtual sets are made possible through the use of a highly optimized interfaced between ELSET LIVE-NT and Real 3D’s PRO-1000 Professional Series graphics engine. Real 3D has developed the PRO-1000 Professional Series as a family of low-cost graphics engines designed to provide reliable real time performance at full 60Hz field rates required for broadcast video standards. The PRO-1000 Professional Series generates the highest realism possible for this class of device, due to its superior real time polygon generation capability, high capacity texture memory, true anti-aliasing and potential for multi-channel operation. Accom and Real 3D have worked closely to ensure that ELSET LIVE-NT and PRO-1000 Professional Series graphics engines work seamlessly together.

John Lenyo, Director of Marketing for Real 3D noted, "The PRO-1000 is a proven performer in mission critical configurations. Working with Accom is exciting for our team because we can bring this proven technology to an entirely new market. We look forward to continuing to enhance the performance of virtual set productions by working with Accom and its customers to advance cost-effective on-air virtual set options."

From the beginning, Accom’s philosophy has been one of open systems. "An open system means open to several computer platforms, peripheral equipment and other software tools," said Kelson. "The intent of our strategy is that ELSET virtual set software is extremely flexible and as a result is the is easiest to use and generates the best looking sets. The PRO-1000 Professional Series gets us to a new price/performance level previously unattainable for live virtual set productions. Customers really benefit since they can easily move their productions to the platform that delivers the right performance level required for their production."

To reflect the expanded product offerings, Accom’s initial virtual set product ELSET has been renamed ELSET LIVE . At the NAB opening this morning, Accom introduced ELSET POST, ELSET POST-NT, and ELSET LIVE-NT . Each of these solutions is based on the award-winning ELSET virtual set system in use around the world by a number of major video program producers. First shipments to customers of the new products are expected during the second calendar quarter of this year.

The REAL 3D PRO-1000 Series is a family of low-cost, high performance graphic engines designed to provide high realism, high throughput and sustained real-time response. Based on real-time technology first developed by Lockheed Martin for high performance simulators, the PRO-1000 Series targets commercial applications such as engineering visualization, architectural walkthroughs, location based entertainment, simulation based systems, and now virtual sets. PRO-1000 systems excel at rendering the high scene density and high image quality demanded by real time virtual set production. All models deliver 1,000,000 fully occulted, anti-aliased, textured, shaded, fogged, faded, translucent, 3 or 4 vertex polygons per second, and up to 200 million fully textured pixels per second.

Accom designs, manufactures, sells, and supports a complete line of digital video production, recording, and editing tools, as well as the ELSET virtual set system, for the professional worldwide television production, post production, broadcasting, and computer video marketplaces.

REAL 3D is a company within the Lockheed Martin Commercial Systems Group. The unit was formed as a separate entity to commercialize real-time graphics technology developed by Lockheed Martin for high performance simulators. The REAL 3D product line includes arcade graphics boards developed for Sega Enterprises, standalone PRO-1000 Graphics Engines targeting the commercial training and engineering visualization market, and the R3D/100 chipset for accelerating OpenGL applications in Windows NT environments for high performance PC 3D graphics applications. Real 3D has also entered into agreements with Intel Corporation to co-develop real-time 3D graphics chip technology for consumer level PC products, and with Chips and Technologies, Inc. to co-develop 2D/3D graphics technology for portable and desktop graphics solutions.

Lockheed Martin is a highly diversified $27 billion advanced technology company with core businesses in defense, commercial, civil government energy, and international markets.

Real3D Pro 1000 was derived from Lockheed Martin's older simulator technology. most simulator technology from SGI, E&S and Lockheed Martin / Martin Marietta was designed to run applications at sustained 60 fps rates because military and industry demanded it. This also benefits arcade games as well :)
 

DaCocoBrova

Finally bought a new PSP, but then pushed the demon onto someone else. Jesus.
dark10x - Told you so

3emetongue.jpg
 

Lazy8s

The ghost of Dreamcast past
You'd have to do extra calculation to fix up the PS2's mip-mapping, too.

The AA run by Model 3 was a nice edge multi-layered AA approach. Nice, also, that it didn't drop from its 1,000,000 quads even when the trilinear filtering, AA, and its supported lights were turned on.
 

doncale

Banned
yeah, the 1,000,000 rectangle polygons per second (i per saying that than quads :) was with lighting, texture mapping, g-shading, mip-mapping, perspective correction, trilinear filtering, alpha blending, z-buffering, multi-layered edge anti-aliasing, etc.
 

Panajev2001a

GAF's Pleasant Genius
Lazy8s said:
You'd have to do extra calculation to fix up the PS2's mip-mapping, too.

The AA run by Model 3 was a nice edge multi-layered AA approach. Nice, also, that it didn't drop from its 1,000,000 quads even when the trilinear filtering, AA, and its supported lights were turned on.

Extra LOD calculations yes, you do them on VU1: it can be done and mip-mapping actually speeds up the GS ;).


Oh... so we are talking edge-AA ? Well, take the time to depth sort polygons and the GS can do that too :).
 

Panajev2001a

GAF's Pleasant Genius
doncale said:
yeah, the 1,000,000 rectangle polygons per second (i per saying that than quads :) was with lighting, texture mapping, g-shading, mip-mapping, perspective correction, trilinear filtering, alpha blending, z-buffering, multi-layered edge anti-aliasing, etc.

Rectangles as in Quads or Quad-strips ?

However, using triangle strips you do a quad with a max of 2 vertices in average ( exclude the very first triangle and each successive triangle is realized with only one additional vertex ).
 

Panajev2001a

GAF's Pleasant Genius
doncale said:
honestly I do not know how it actually is done. but the fact that Model 3 can lock down at 60fps was said by at least one Real3D engineer. I will try to find the quote.


Panajev, ok I will give PS2 the advantage of fillrate to reproduce Model 3 image quality by sheer brute force.

btw, Model 3 actually transforms 60 million polygons, raw. pretty close to PS2's 66 million.

Actually the EE can calculate RAW (simple transform with perspective projection) the following:

~60 MVertices/s with VU1 (5 cycles for the Transform loop taking advantage of the second FDIV on the EFU)

~42.85 MVertices/s with VU0 (7 cycles for the Transform loop: you are limited by the single FDIV latency here).

You can merge display lists from both VUs (a bit of a pain in the ass: I am only using VU1 so far and DMA already had its toll on me, but I am learning more day by day :)).

The GIF-to-GS bus ( 64 bits wide and clocked at ~150 MHz ) obviously cannot carry all those vertices and the 66 MVertices/s figure you see is a more practical number.

VU1 is able to push about 18 MVertices/s with 4 infinite/directional lights, perspective corrected textures (w is used for STQ parameters calculations) and fog.
 

Lord Error

Insane For Sony
Right, and quoting anything else (peaks, which by definition aren't characteristic, and redrawn triangles, which don't add to geometry) wouldn't have been representative.
Well, unless we can find the measured average poly throughput for games on arcade machines such as Naomi 2, anything pretty much goes. It's hard to verify even the peak numbers for those arcades, much less measured averages.

i suspect the polygons per second numbers tossed around on boards like these are grossly inflated. you can get far cry to display a polys per second counter, and that tends to hover around 3-5 million even in huge, detailed outdoor vistas. i suspect most console games push far fewer polys than far cry, and i doubt there are any games that push 4-6 times as many polys.
Far Cry uses very efficient LOD system. If you disable it, you will notice that the counter gets up to 25M polys/sec but the huge detailed outdoor scenes look almost the same as with LOD that cuts them to 5M polys. Not all games use LOD systems, though, some resort to brute force.

The reason model 3 racers don't look as good as current gen racers is due to lighting..if you took lighting out of games like burnout and GT they would look like crap. If there was dynamic lighting on say daytona 2 or Super GT they would no doubt look as good as any current game IMO.
The reason Model 3 games don't look as good as today's games is pretty much everything. Polycounts, textures (yes, they look very weak compared to even better PS2 games), lighting and the complete lack of any advanced special effect, such as complex particles, post processing effects or good environment mapping. The hardware was good for it's time, but give it a well deserved rest.

btw, Model 3 actually transforms 60 million polygons, raw. pretty close to PS2's 66 million.
Yeah, and my grandma transforms 120 bajilion polygons while she's asleep. Honest, she told me so!
 

dark10x

Digital Foundry pixel pusher
dark10x - Told you so

Actually, I still don't believe it. The wording of that article doesn't actually suggest physical hardware at work locking the framerate. The hardware is powerful enough to provide reliable performance with full 60 Hz frames...

So, by your logic, any developer could toss together shit code and expect 60 fps? I call bullshit. The hardware aided the developers in achieving that goal, but it was the code that pulled it off.
 

Lazy8s

The ghost of Dreamcast past
Marconelly:
Well, unless we can find the measured average poly throughput for games on arcade machines such as Naomi 2, anything pretty much goes.
Not for this comparison. This one was a direct relation: number of polygons used to model the characters in the arcade version versus the home version.
The reason Model 3 games don't look as good as today's games is pretty much everything. Polycounts, textures (yes, they look very weak compared to even better PS2 games), lighting and the complete lack of any advanced special effect, such as complex particles, post processing effects or good environment mapping. The hardware was good for it's time, but give it a well deserved rest.
On the other hand, Model 3 games filter textures with trilinear interpolation, use more sophisticated mip-mapping, output in proscan, and have nice quality edge anti-alaising - all of which are not common for PS2 games.
 

Lord Error

Insane For Sony
On the other hand, Model 3 games filter textures with trilinear interpolation, use more sophisticated mip-mapping, output in proscan, and have nice quality edge anti-alaising - all of which are not common for PS2 games.
Not common, but certainly doable (and have been done). As I've said, if you just compare some of the best offerings of both machines (like Super GT vs Burnout 3), it becomes painfully obvious that some people have massive delusions about that arcade hardware.

Not for this comparison. This one was a direct relation: number of polygons used to model the characters in the arcade version versus the home version.
Yeah, I guess that article said that VF4 character models had two times the polygons of the PS2 conversion? But what was the ratio of Model 2 and Saturn VF2 polycounts?

If it's really true that Saturn version had 5x less polygons, that would actualy mean that the PS2 VF4 would have to be running at less than 1Mp/s (average) to have it's polycount downgraded as much.
 

jett

D-Member
I don't have anything to add other than that Virtua Fighter 2 looks like absoulte shit nowadays, at least VF1's graphics have some sort of endearing charm to it.

Also, the arcade version of VF4/E don't look as good as some people would like to make you believe. Anyway, check out the intro movies in VF4evo for the PS2, they come straight from the arcade version. Not that much of a difference, huh? Those disgusting textures in Sarah's stage were ported over pretty faithfully, I'd say. :p
 

Lazy8s

The ghost of Dreamcast past
Marconelly:
Not common, but certainly doable (and have been done).
They're all standard in the games for Model 3, but they're only used rarely in PS2 games, and then only at separate instances. So, Model 3 graphics do sport that technical advantage rather distinctly.
 

Shinobi

Member
jarrod said:
It was 2 years, and DC/Naomi was actually quite a leap over Model 3. Better effects, lighting and 3 times the geometry.

I don't know...DC/Naomi never showed me it was a huge leap in the polygon or texture areas. I mean Daytona doesn't look much better then say, Super GT, aside from the lighting. Though DOA2 certainly looks better then the arcade VF3. And I guess Crazy Taxi looked better then Harley Davidson. Still, for a two year gap it really wasn't a huge difference. Then again I am comparing a console to a state of the art arcade board.

BTW, I miss Real 3D's Model boards big time...would be wonderful to see what their Model 4 and 5 boards could've produced, based on the Model 2 and 3 costs.
 

ourumov

Member
I don't think there is a single DC game that reachs Daytona USA 2 level of graphics as a whole package...
But to say tha Model 3 is still the king is a big stupidity...
 

doncale

Banned
I agree with Lazy8s comments, and most of Shinobi's.


PlayStation2 has vastly more geometry and lighting than Model 3 for sure. but Model 3 is clearly the winner when it comes to image quality because its implementation of various features, mip mapping, trilinear filtering, anti-aliasing are better than PS2's and every Model 3 game uses them combined together, at all times.

If Model 3 kept its image quality and we bumped the geometry/lighting performance up to PS2 levels, Model 3 would wipe the floor with PS2.

the creme of the crop of PS2 games look better than Model 3 games, but the vast majority of PS2 games look worse overall than Model 3 games, IMO. to me, if not many people, it proves that polygon count doesnt mean a whole lot. btw, I happen to think that most Dreamcast games overall, also do not look as good as most Model 3 games. every single Model 3 conversion to Dreamcast looks worse than the Model 3 original. either by a little (virtual on 2/voot, sega bass fishing, virtua striker 2) or alot (VF3tb, Sega Rally 2)

I much prefer the visuals of Model 3 Harley Davidson over NAOMI/DC Crazy Taxi. I would take the heavy texture mapping, trilinear filtering, edge anti-aliasing and lower polygon count of HD over the modest texture mapping, bilinear filtering, no anti-aliasing and higher polygon count of CT, any day. I feel that Model 3 HD models Los Angeles better than DC/NAOMI CT models San Fransisco.


I own all 4 consoles so I dont have much bias. I feel that Xbox and Gamecube are the only current remaining consoles that could easily reproduce Model 3 games close to exact.
their graphics processors (NV2A and Flipper) are closer to Model 3's Real3D Pro 1000s in features and quality, while being much more powerful. until I see a PS2 port of a Model 3 game, I remain skeptical. lets see how well PS2 handles Model 2 VF2 first.
 

NotMSRP

Member
So who here really knows about hardware architecture, computer graphics, linear algebra, and the likes in great detail?
 

Crazyace

Member
I thin a lot of arcade conversions such as VF4 on PS2 suffer from memory limits at least as much as any other technical issues - so it's quite possible that a poly reduction was a space saving measure, and the lighting mismatches grew out of that.
 

Lazy8s

The ghost of Dreamcast past
It's a reasonable expectation that RAM space was a large factor in the downgrading of the VF4 conversion and also of the VF2 conversion.
 
Top Bottom