Discussion in 'Sega Dreamcast Development and Research' started by Dreamcast, Feb 8, 2014.
It's waiting for Vimeo to process it
Looks interesting. Hopefully you will make a video of the finished version. I think it could have warranted a full game. Maybe a streaming/open world behind the player view type of Armada like exploring shooting game. Considering it's not even using full range of the DC capabilities it looks real nice.
I'll release a patch .Good luck
On the subject of early Dreamcast tech demos, some of you may remember that the UK Official Sega Saturn Magazine described being shown an internal video tape of various pieces from while the Katana preliminary title was still in use, which included two different "slices" of SCUD Race... Well, it seems as if someone made a direct feed copy of that very same tape:
While watching that found the Naomi 2 tech demos linked to it . They look Impressive , though you can probably reach similar results on the DC just cut down a bit.
I'll never understand how Yu Suzuki concluded through months of extensive technical research that recreating the essence of Virtua Fighter 3 was within the Saturn's capabilities, yet he didn't feel the same was possible for VF4 on the Dreamcast, which is supposedly why a conversion was only ever announced for the PlayStation 2. Sure, AM2 would have needed to lower polygon counts and reduce the overall number of individual light sources on Dreamcast when compared with the NAOMI 2 original, but I suspect there's more to this story... Perhaps the real reason why PS2 owners received VF4 exclusively is because Sega knew the Dreamcast was on the way out, and that it made the most financial sense for their flagship fighter to go where it had the greatest chance of success? Either way, I'd be interested to see whether a commercial NAOMI 2 game could reach the performance levels seen in that video once you factor in such elements as AI that can be interacted with or collision detection, since there's nothing shown you couldn't do by scripting certain objects and effects to appear at a given point in time. For this alone, I'm normally reluctant to pay too much attention to the kind of tech demo that's all about throwing around hundreds of objects, preferring the more realistic depictions of something like the Katana SCUD Race footage, since I can imagine that very much being within the Dreamcast's limitations. (Of course, something would need to have been done regarding that painfully obvious pop-in, and the frame rate may have taken a hit once cars were implemented - maybe that's why this project was never taken further?)
So whatever happened to the future city updated/patch thing?
Virtua fighter 4 was heavily downgraded for the ps2 though. Characters are half the poly county, textures are horrid and the resolution is something around 512x224.
Characters went from 14k+ polys to half that around 7k polys,
Face has lost lot of detail, and the ears are just mostly a flat plane on ps2 instead of full modeled, also it is not visible in this shot but characters don't have eye balls is a static mesh reducid in polycount too.
For reference Dead or Alive 2 characters for the Dreamcast are around 7k- 9k polys each, but they waste too many polygons on things like the eyes (almost 2k wft) and little details like modeling the nails lol. and of course the bobs
Helena DOA 2 Dreamcast
Jan lee Doa 2 Dreamcast
To put it into perspective the tekken series on ps2 put characters polycounts from 4k to 6k..
I do have an old revision of the katana SDK, R4, it was available for download here once, it has scud files, I think more than the R9 one which only have the Ferrari F40.
It was dated february 98 IIRC, but I can't even run the katana R9 demos as a CDI on the emulator, might have forgotten how to do it.
They would have to be cut down quite a bit to run on the DC (or the original Naomi, for that matter) - the Naomi2 has two complete PowerVR subsystems running in parallel and a hardware transformation and lighting processor added because the (still single) SH4 wasn't capable of keeping both graphics pipes busy on it's own.
2x PVRs was getting same polygon/vertex lists, and had exact same texture sets in video RAM, but just different tiles enabled when rendering (usually like chequered flag). there is zero overhead when pushing data to 2x PVRs, because it is done in "broadcast"-way - single DMA or CPU write goes to both TileAccelerators or video RAMs.
since beginning, the main bottleneck of Dreamcast and NAOMI was SH4 CPU, basically it was able to OR T&L quite many vertices to fully load PVR2 GPU but with quite poor lighting/effects, OR T&L relatively small number of vertices with good lighting and effects. but you can't get both.
ELAN coprocessor was made to solve this problem, it is much faster in T&L than SH4, iirc like 10x or more.
So are you trying to suggest that the paired PVR setup has exactly the same rendering throughput as a single one? Obviously, that can't be true, because if it was then the whole arrangement would be utterly senseless.
Since I think we can agree that the setup in the Naomi2 can, in fact, render more than the single PVR in the Naomi1, then the comment about increased SH4 load if it was still being used for T&L holds. The fact that the same data is fed to both chips is irrelevant - what IS relevant is that there is more of it.
of course 2x PVR2 have as 2x fillrate.
but at practice, Naomi2 games frames have 60K-100K vertices = 3M-6M per second.
and this is "raw input" vertices numbers, some part of them will be clipped by ELAN after Transformation phase, if they are no visible (ie behind the camera, too far etc), so real vertices output to PVR2 GPUs will be even less number, like 50-60K/frame = 3-3.5M/sec which is about doable even by single PVR2.
but, SH4 CPU will be not able to Transform, Clip, and apply spot and directional light sources (usually about 6) to the same number of vertices.
moreover, SH4 will not be able to T&L even typical for cool DC/Naomi games 20K-30K vertices per frame with 6-8 light sources. iirc SH4 can do ~25K-30K per frame @60Hz with max 2 direction lights, like in Dead or Alive 2 for example, where is a hell lot of crazy optimizations.
OK, I'm still confused here - what exactly were you saying "no" to?
I think we both agree on the following:
1) In the DC and original Naomi the SH4 was often the limiting factor for graphics performance, especially in scenes with non-trivial lighting
2) The dual PVR setup in Naomi2 has a larger rendering throughput than the single PVR in Naomi1/DC
3) If the dual PVR setup had been implemented without hardware T&L, then the SH4 would have been even more of a bottleneck
Which is basically what I was saying in the first place.
OK, I'm not denying that adding hardware T&L to a system with a single PVR would have improved performance - just that doubling up the rendering capacity of the system meant it went from "nice to have" to "the system would be horribly unbalanced without it".
Too bad Sega didnt release some home console based on the naomi 2... dreamcast 2 which would be backwards compatible.
@TriMesh agree with 1) 2) 3) except for "non-trivial lighting", I'd replace it with "good lighting".
Dreamcast capable of quite nice visuals for the year of birth (1998), and next year, but graphics tech progressed very fast these years, and in 2000 DC was not really good enough. PVR2 GPU was good enough, the main problem was in weak SH4 CPU and lack of Transform&Lighting&Clipping coprocessor.
I'm mainly started this conversation because of these statements
I'm totally disagree.
even if reduce vertices number to half of typical in N2 games/demo vertices count - SH4 CPU will be nowhere able to do the same calculations fast enough for 60fps.
right. and I'm pretty sure N2 was initially planned for home use. backward compatibility feature is very not typical for arcade world.
OK, now I'm completely lost....
So you think Naomi2 games COULD be run on the original Naomi (or the DC) without cutting them down drastically?
And that a single SH4 could keep the dual PVR setup on the Naomi2 busy WITHOUT hardware T&L?
@TriMesh ugh, probably I said it totally wrong.
I wanted to say - SH4 will not be able to calculate nor half nor even 1/3 of typical Naomi2 vertex number, with the same lighting "quality".
Accidentally typed in the quote.
Edit:That makes me curious. As the demul emu author maybe you could check/shed light on this. In another forum someone claimed to have extracted extracted wacky races and that it was pushing 60k to 78k triangles @ 30 fps. That's definitely Naomi 2 like geometry at half the frame rate right? Do you think your verytcounts can deny or affirm that?
yes, I can confirm that. in places this game spikes to ~60K-70K vertices per frame @30fps
speaking in general, it is not that hard to do the math - lets take max number - 70000 * 30fps = 2100000/sec
and then SH4 CPU clock 200`000`000 / 2100000 = 95 SH4's clocks per vertex max (and there is also other tasks in the game which should be handled too).
it seems enough to vertex coords transform and apply one simple "point light" source, and also maybe some ambient.
Separate names with a comma.