Xbox in 2000, will its graphics still reign supreme.......!?

[Simon F]

Naomi2? Is that the SEGA arcade cabinet running VF4? If it is, you guys have to realize that it had buttloads of RAM. Something which is not too cost efficient for home consoles.....

The Naomi2 obviously included a reasonable amount of RAM because it was an arcade system. Arcade machines can usually pay for themselves in a month. They are meant to be better than home systems.

Compare the Naomi1 and DC. In terms of CPU and GC, they are identical, but the former has far more memory because they can afford to put more in. In terms of ports, however, what usually happened is that the DC version would use texture compression more agressively (eg 75+% of textures).
BTW, PC-Engine is correct in pointing out that N2 uses SDRAM.

AFAIK, N2 had 32mb main RAM, 2 x 32MB VRAM, 32MB data memory and 8MB of sound memory.

I think it only had 32Mb of VRAM, and I can't remember the other numbers.

PS2 sold for $300 and SONY was losing $150 per unit at launch which means about $450 to manufacture. By that time DC was already down

I'd love to comment but that would not be appropriate....

And it also remained me most/all N2 based games is released around mid 2001. And N2 technology was showcase with some early demo, in mid/late 2000, when everyone was complaining about the jaggieness of PS2. PS2 hardware was showcase in 1999.

N2 probably can't be mass produce or release in early 2000. PS2 was already late, they were planning for 1999 released.

The N2/Elan system was finished a long time before its public launch.

Look at it this way, even DC could not match N1 amount of RAM. N2 had 2 CPU and 2 GPU and 136mb of RAM.

It had a single CPU and I don't think (but I could be wrong) that it had that amount of RAM. BTW just because a system can address a certain amount of memory doesn't mean that it has to be fully populated.

Looking at that, PS2 isnt that bad in some areas. It could be so much better if the amount of RAM was doubled. But the bang/buck is not too efficient for home systems.

Ignoring RAM for the moment, you haven't considered silicon area. You'd be surprised how small that figure is compared to the monster that is PS2.

As far as using the Naomi2 hardware, check what the non lit or single light poly throughput is for the hardware. Considering the typical amount of HW lights used, I don't think picking a platform that couldn't keep up with a GeForce1 would be a good idea.

Funnily enough, there was a requirement for something that wouldn't suddenly slow to a crawl when a reasonable number of lights, and this includes lights of reasonable complexity, were enabled. The GF1 soon ran out of steam - its T&L unit is best described as having a lowercase L.

the effective fillrate is useless to discuss, this is a console not a PC. When you remove the need for bandwith to fill fillrate PVR's technology loses a lot. It is cheap, doesn't use many transistors and uses a lesser amount of bandwith for rasterization then an IMR. None of those are major factors to a console. Poly throughput most certainly is,

Again, why? I don't recall any law saying "a console game must not have any overdraw".


Anyway, I've had enough of this and I've just run out of lunchtime.

 

[Guest]

The Naomi2 obviously included a reasonable amount of RAM because it was an arcade system. Arcade machines can usually pay for themselves in a month. They are meant to be better than home systems.

Compare the Naomi1 and DC. In terms of CPU and GC, they are identical, but the former has far more memory because they can afford to put more in. In terms of ports, however, what usually happened is that the DC version would use texture compression more agressively (eg 75+% of textures).

Yeah, that was what i am trying to tell PC here.



N2 specs(repost in case anyone missed it)
CPU: Hitachi SH4 200MHz (360 MIPS / 1.4 GFLOPS)
GPU: 2 x PowerVR2 + Videologic "Elan" T&L chip
APU: ARM7 + Yamaha (64 2D Voices)
Main Memory: 32MB 100MHz 64-bit SDRAM with 0.8 GB/Sec
Video Memory: 2 x 32MB with 1.6 GB/Sec total (0.8 GB/Sec per PowerVR2)
Model Data Memory: 32MB with 0.8 GB/Sec
Sound Memory: 8MB
10 Million Polygons/Sec with 6 Hardware Lights
200 Megapixels/Sec (800 with 4x overdraw)
200 Megatexels/Sec (800 with 4x overdraw)

 

[marconelly!]

Using DX and PC native hardware developers were utilizing GF1 level techniques for XBox launch titles and they still managed to easily outdo what the PS2's second/third generation titles looked like.

Correct me if I'm wrong, but since when perpixel lighting (Halo, Wreckless), vertex shaders (Halo, Wreckless, DOA3) and dot3 bump mapping (Halo) are GF1 features? Do I have to remind again that even with the hardware it runs on, Halo was 30FPS with all the chugging along? What would it be like on a hardware that is almost 2x less powerful?

That is not to comment that Halo would *not* be ready for the launch on that date (they barely made it anyways!) - The only Xbox killer app from the launch would not be there along with probably other games that had to be rushed even for the real launch (DOA3) I think that would make console a complete failure considering how much more hyped PS2 was, and how insignificant difference in launch titles quality actually would be.

 

[BenSkywalker]

Chap-

Features front---i take it as bumpmapping, AA, texture filtering and such?
PS2 GS is undoubtly very basic, but one thing to remember though, GF1, as we know it, might have a cool list of features, is it worth it to do them during gameplay?

What do you mean is it worth it to do them in gameplay? Play Giants or Mafia @640x480, runs quite nicely on GF1 level hardware.

GF1 can run the game w/ full texture detail? At what framerate? How about your other graphics settings?

If you are having problems with the game running on a system with GF1 level hardware drop the resolution to 640x480 and make sure you have enough RAM. The highest quality shadows can be a bit nasty in portions although the texture quality need not be dropped.

Simon-

Funnily enough, there was a requirement for something that wouldn't suddenly slow to a crawl when a reasonable number of lights, and this includes lights of reasonable complexity, were enabled.

What are the last six console games you have played? I'd be very interested hearing about those that used four simultaneous lights, let alone six. Despite the couple of hundreds of dollars I spend a month on games I don't seem to have a single one in my library that uses six lights at once.

The GF1 soon ran out of steam - its T&L unit is best described as having a lowercase L.

Which games show this problem? It's like showing how weak current PVR hardware is against the three year old GeForce1 in MCAD applications, in a theoretical sense it is a valid point.

Again, why? I don't recall any law saying "a console game must not have any overdraw".

640x480x60(TV's limit)= 18.432Million pixels per second. We'll figure for four texture layers and we have 73.728MTexels. With 6X OD the GeForce1 would only have 37.632MTexel fill left over(while maintaing 60FPS). Fillrate is not terribly important on consoles, the GF or PVR based chips would both run into other bottlenecks before raw fill became relevant(binning/bandwith).

marconelly!-

Correct me if I'm wrong, but since when perpixel lighting (Halo, Wreckless), vertex shaders (Halo, Wreckless, DOA3) and dot3 bump mapping (Halo) are GF1 features?

Vertex shaders aren't a feature on the GF1, the rest of your list is included however. Are you thinking of the TNT perhaps?

That is not to comment that Halo would *not* be ready for the launch on that date (they barely made it anyways!) - The only Xbox killer app from the launch would not be there along with probably other games that had to be rushed even for the real launch (DOA3) I think that would make console a complete failure considering how much more hyped PS2 was, and how insignificant difference in launch titles quality actually would be.

I'm just talking about hardware. Look at PGR compared to GT3(and even then, GT3 wasn't a launch title).

 

[Simon F]

Simon-
What are the last six console games you have played? I'd be very interested hearing about those that used four simultaneous lights, let alone six. Despite the couple of hundreds of dollars I spend a month on games I don't seem to have a single one in my library that uses six lights at once.

If you can't do lots of lights cheaply you don't use lots of lights. N2 was for new games.

 

[marconelly!]

Vertex shaders aren't a feature on the GF1, the rest of your list is included however. Are you thinking of the TNT perhaps?

My memory doesn't go that far, but I found no mention of
per pixel lighting in this article or pixel shaders for that matter (and they are responsible for dot3, correct?)
http://www6.tomshardware.com/graphic/99q4/991011/index.html

and this press release about GF2 card says:
"GeForce2 GTS fully leverages powerful new features like complex per-pixel lighting from NVIDIA's newest GeForce2 GTSâ„¢ graphic processing unit" http://www.leadtek.com/geforce2pr.htm

Also Toms Hardware mentions it only in their feature about GF2:
http://www6.tomshardware.com/graphic/00q2/000427/geforce2-07.html

I believe you are right, though, and that's beyond the point as the proposed 2000 xbox hardware would have a GF2 or better card anyways.

I'm just talking about hardware. Look at PGR compared to GT3(and even then, GT3 wasn't a launch title).

True, PGR undeniably pushes more stuff around, but most reviewers found it actually worse looking than GT3 (texture art, etc). On a strictly hardware level, you have to take into account that PGR actually runs on *present* xbox hardware, not on something almost 2x weaker. If in it's present state it takes a techhead to see advantages PGR has, I'm pretty sure the lower level hardware would make things only worse, perhaps not RRV worse, though

My point is, MS did a good job delaying xbox, upping it's specs and giving developers more time. That was their only chance to make some impact on market, IMO, as the xbox would went completely unnoticed if it wasn't for the technical specs they could boast about and it's good launch lineup.

Fillrate is not terribly important on consoles, the GF or PVR based chips would both run into other bottlenecks before raw fill became relevant(binning/bandwith).

However, it is really important on PS2, as the games that utlize it properly, often make good use of it. I remember reading an interview about porting MGS2 to Xbox, where they said that due to high fillrate on PS2 they were able to do things they wanted, but had to recode everything and change the logic because 'xbox doesn't work that way'

 

[PC-Engine]

The Naomi2 obviously included a reasonable amount of RAM because it was an arcade system. Arcade machines can usually pay for themselves in a month. They are meant to be better than home systems.

Compare the Naomi1 and DC. In terms of CPU and GC, they are identical, but the former has far more memory because they can afford to put more in. In terms of ports, however, what usually happened is that the DC version would use texture compression more agressively (eg 75+% of textures).

Yeah, that was what i am trying to tell PC here.



N2 specs(repost in case anyone missed it)
CPU: Hitachi SH4 200MHz (360 MIPS / 1.4 GFLOPS)
GPU: 2 x PowerVR2 + Videologic "Elan" T&L chip
APU: ARM7 + Yamaha (64 2D Voices)
Main Memory: 32MB 100MHz 64-bit SDRAM with 0.8 GB/Sec
Video Memory: 2 x 32MB with 1.6 GB/Sec total (0.8 GB/Sec per PowerVR2)
Model Data Memory: 32MB with 0.8 GB/Sec
Sound Memory: 8MB
10 Million Polygons/Sec with 6 Hardware Lights
200 Megapixels/Sec (800 with 4x overdraw)
200 Megatexels/Sec (800 with 4x overdraw)

If you're being sarcastic then please ignore this reply, otherwise read on.

N2 uses cheap SDRAM. Reiterating what Simon F. has said, N2 based console also doesn't have to have the same amount of RAM as the arcade variant just as the DC doesn't have the same amount of RAM as N1.

Even still, the ports were identical. How did DC run perfect ports of N1 games? Easy, the arcade version of the game uses 25% compression while the console variant used 75% compression. That cuts down the RAM requirements in half compared to N1. In addition, again reiterating what Simon F. said, the arcade variant can afford to be more expensive so little consideration is put into making it dirt cheap like a console requires. That's why the N2 was so expensive.

Now looking at the above information, it's not difficult to see how a N2 based console can be made at a cost that's comparable to PS2 as it would only need half as much memory, cheap SDRAM to be exact. I'm pretty sure $250 would cover the costs of the additional RAM, second PVRDC, and Elan chip. These chips were not huge on silicon area either.

Like I've said before N2 in console form didn't make it to market to battle PS2 because it would cannibalize DC sales. Also SEGA wouldn't be able to eat the $150 cost to make it a mass market item. It's analogous to releasing Xbox II next year...suicidal

However the prospect of a N2 based console being backward compatible with the N1 based DC was very enticing. It would only cannibalize a portion of DC sales instead of total 8)

It would definitely fragment the market though

 

[Crazyace]

silly talk

Naomi 2 as a console is pretty silly - there are too many seperate components, and some horrible ( almost 3dfx like ) features ( 2 video chips will split the rasterising, but triangle lists and textures will be duplicated )

A much better alternative would be a 200MHz video chip combined with a 400MHz sh4 or sh5 - Although the T&L chip is nice, I actually like the basic vector matrix apply on the sh4 - It's much more flexible..


PS2 just follows the PSX design style - simple features with high performance transform, ( The best thing about the PSX GPU [Nvidea wasnt the first....] was the dithering of gourard shaded polys - that (alongside the transparencies) lifted it above the Saturn.. )

 

[Ozymandis]

This is a bit OT, but does anyone here know anything about the changes in the arcade board named "Hikaru"? It's an upgraded Naomi board, but I'm curious as to what changes were made. This is the board that runs Planet Harriers and Virtual On:Force I believe?

 

[marconelly!]

Ozy, this is the known information about Hikary board:

http://www.system16.com/sega/hrdw_hikaru.html

It's polygon count is estimated at 5 Million polys/sec (2.5 for Naomi) and rendering speed at 1000 M pixel/sec (500 for Naomi)

 

[PC-Engine]

A much better alternative would be a 200MHz video chip combined with a 400MHz sh4 or sh5 - Although the T&L chip is nice, I actually like the basic vector matrix apply on the sh4 - It's much more flexible..

If only the tech you mentioned existed at that time at a reasonable cost or existed at all

N2 tech used mostly off the shelf parts which was cheap. It would use existing system board designs further reducing costs. Using two PVRDC chips isn't really silly when process tech and high costs associated with higher clocks was the limited factor. Duplicated memory pools are pretty negligable when you're using cheap SDRAM and again only half of what N2 needed to be exact. The Elan chip and PVRDC only operated at 100MHz ie low heat low power low cost making it a good choice for a console. ATI was selling their MAXX series dual chip boards for awhile btw. Single higher clocked chip would be nice if it was available at a reasonable cost..

 

[Ozymandis]

Ozy, this is the known information about Hikary board:

http://www.system16.com/sega/hrdw_hikaru.html

It's polygon count is estimated at 5 Million polys/sec (2.5 for Naomi) and rendering speed at 1000 M pixel/sec (500 for Naomi)

Hmm. Yeah, I saw the specs there before. Maybe it's a dual PVR2 configuration, like Naomi2 but without the T&L unit?

 

[PiNkY]

My memory doesn't go that far, but I found no mention of
per pixel lighting in this article or pixel shaders for that matter (and they are responsible for dot3, correct?)

GeForce1 can do Dot3 so per pixel lightning(as in doom3) is possible (Geforce 2 did not add any more features over Geforce 1, but nvidia just started hyping their more flexible reg.comb. as NSR for GF2 launch. DX8 Pixel Shaders basically are more advanced register combiners, so yes dot3 ops are done in the pixel shader on dx8 hw.

 

[V3]

The N2/Elan system was finished a long time before its public launch.

You guys were stockpiling N2/Elan that early ? So I guess you guys are nearly finished with the Naomi3 that Sega recently announced, care to give us some hints ? Will it surpasses the 9700 and the seems elusive NV30.

Normally arcade ends, surpasses the consumer ends by quite a bit, not so anymore now days. Lets hope N3 impresses.

 

[Simon F]

So I guess you guys are nearly finished with the Naomi3 that Sega recently announced, care to give us some hints ?

I know absolutely 0 about Naomi3.

 

[bas1975]

"Will it surpasses the 9700 and the seems elusive NV30. "

Isn't it the Naomi3 that is supposed to be based off of XBOX hardware or am I completely out to lunch on that one? If it is based off of XBOX hardware, it won't have a chance of surpassing the 9700.

 

[PC-Engine]

"Will it surpasses the 9700 and the seems elusive NV30. "

Isn't it the Naomi3 that is supposed to be based off of XBOX hardware or am I completely out to lunch on that one? If it is based off of XBOX hardware, it won't have a chance of surpassing the 9700.

You're out to lunch

Chihiro is the Xbox based hardware. NOAMI 3 is still shrouded in secrecy 8)

My guess is PowerVR Series 4 with quad pipes with quad PS and embedded RAM. Quad GPU board confiuration. Elan 2 at 200 MHz with quad VS. 100 million polys with 16 lights. Supports anatomically correct physics based animation ie muscle movement under elastic skin etc.

 

[BenSkywalker]

Simon-

If you can't do lots of lights cheaply you don't use lots of lights. N2 was for new games.

Given the choice of ~20Million polys/sec with no or one lights or six million polys/sec with eight developers chose the former pretty much exclusively(I don't know of any exceptions). Most games today still rely on lightmaps for the majority of their 'lighting effects'. Given the current direction we are seeing from titles in the works the amount of lights, in terms of the direct lighting calculations, seems to be taking a back seat to shader effects.

Marconelly-

http://www.anandtech.com/video/showdoc.html?i=1056&p=6

In particular they are talking about CEM, but that also covers per pixel lighting. If you were around these boards back in that timeframe we had a considerable amount of discussion concerning the usefulness of some of the features on the GeForce in terms of ever seeing them show up in games. Then Giants hit and used about all of them and ran quite nicely on a GeForce1.

If in it's present state it takes a techhead to see advantages PGR has, I'm pretty sure the lower level hardware would make things only worse, perhaps not RRV worse, though

I can only assume people who can't see the edge in PGR are blind It is in another class compared to GT3. Superior texturing, model complexity, environmental effects and the biggest difference is GT3's horrendous aliasing.

However, it is really important on PS2, as the games that utlize it properly, often make good use of it. I remember reading an interview about porting MGS2 to Xbox, where they said that due to high fillrate on PS2 they were able to do things they wanted, but had to recode everything and change the logic because 'xbox doesn't work that way'

With the XBox you simply throw textures at it. I'm a bit confused as to how they could be chewing through the XB's fillrate. Running in 480p on the Box, where MGS2 was running 480i, you could have ten texture layers with 5X OD @60FPS. No way MGS2 is pulling anything close to that.

 

[marconelly!]

I can only assume people who can't see the edge in PGR are blind It is in another class compared to GT3. Superior texturing, model complexity, environmental effects and the biggest difference is GT3's horrendous aliasing.

On a GT3 side you have much better texture art, smoother updated enviro mapping, better executed static light maps and more dramatic looking lighting/highlights. I think it's mainly texture art that makes GT3 simply look more true to life and believable. As I've said, most reviewers that don't concern themseves with the 'things behind the scene', tend to agree. Btw, I see almost nothing of that 'horrendous antialiasing' on my TV with component cables. I have a small TV though, that probably helps a lot.

With the XBox you simply throw textures at it. I'm a bit confused as to how they could be chewing through the XB's fillrate

I don't know. Perhaps someone with real life game coding experience on PS2 can comment (Archie or someone?) MGS2 really has quite a lot of layered effects in some scenes, and perhaps theoretical performance simply doesn not work in real life.

 

[Simon F]

Simon-

If you can't do lots of lights cheaply you don't use lots of lights. N2 was for new games.

Given the choice of ~20Million polys/sec with no or one lights or six million polys/sec with eight developers chose the former pretty much exclusively(I don't know of any exceptions).

On what piece of hardware? If it was something like a GF, then, IIRC, to get that performance those lights have to be very simple (eg parallel) which, I agree would be rather dull and boring. As I said, the GF had very good transform capability (at the time) but the lighting power was sorely lacking. Even today Elan's lighting unit would be competitive.

Most games today still rely on lightmaps for the majority of their 'lighting effects'.

Which are generally rather static. That approach is certainly ok for the non-moving lights and/or for low tessellation regions.

Given the current direction we are seeing from titles in the works the amount of lights, in terms of the direct lighting calculations, seems to be taking a back seat to shader effects.

Do you mean shader effects as in per-pixel lighting? You still have to do the set up for the per-pixel calculations at the vertices which is not insignificant.