Dreamcast-Talk.com

Gabbyjay wrote:

RoMagnusLarsson wrote:Sega Model 3 (1996): Processor 66Mhz / Audio RAM 1Mb / Vídeo 8Mb RAM 66Mhz.
Sega Dreamcast (1999): Processor 200Mhz / Audio RAM 2Mb / Video RAM 8Mb 100Mhz

1) Completely different CPU architectures (PowerPC and SH-4) with different instruction sets and different numbers of instructions-per-clock, so clockrate alone means jack shit.
2) Model 3 Step 1.5 and Step 2 had CPU-clockrate increased to 100 MHz and 166 MHz, respectively. Similar for the (V)RAM-clockrates.
3) (V-)RAM-bandwith is a product of bus-interface width and it's clockrate, so what's the use of stating the clockrate without the bus width? Model 3 has way more bandwidth available in total.
4) Model 3/Real3D-pro is a complex architecture where always multiple chips and busses work together! Picking one of them and disregarding the others does not work here, it's not the same division into CPU-GPU-RAM-VRAM as in later systems!
5) You don't seem to understand your own quotes.
The Real3D Pro-1000 is a system that takes up a hole PCB (one of several stacked PCBs in Model 3). You posted a quote stating this yourself: "The Pro-1000 occupied an entire motherboard-sized board. It was therefore not a true "GPU" because it consisted of multiple large custom chips." But now you just took a few of the components on it and forgot about the rest. It does not work this way.
6) You completely ignore that Model 3 was an MROM-based architecture, which was one of its main advantages.
Thus, it had VASTLY more main-, video- and sound-memory - and bandwidth - available than the DC. This is especially important since we all know that Dreamcast's capabilities in displaying polygons are mostly limited by available memory.

I did not say that, I uploaded Supermodel's developer responses to my questions. And what you quoted up there was DC/M3 specs, which are right

Ro Magnus Larsson wrote:I did not say that, I uploaded Supermodel's developer responses to my questions.

I know, I have commented on that.

And what you quoted up there was DC/M3 specs, which are right

Only they are by far not complete, as I have explained, and by just picking some random specs or comparing two completely different architectures, the reader will probably get a false impression. So it was necessary to point out what's missing, which changes the impression completely, especially when you talk about ports (or even emulation of M3 on DC o.O).
Otherwise, it would be like looking at a roman galley and only think about the sails, but forgetting that there are two dozen men rowing underneath.

This turned out to be quite an informative thread! Thanks goes to those who took the time to respond with a long-form answer.

Gabbyjay wrote:

Ro Magnus Larsson wrote:I did not say that, I uploaded Supermodel's developer responses to my questions.

I know, I have commented on that.

And what you quoted up there was DC/M3 specs, which are right

Only they are by far not complete, as I have explained, and by just picking some random specs or comparing two completely different architectures, the reader will probably get a false impression. So it was necessary to point out what's missing, which changes the impression completely, especially when you talk about ports (or even emulation of M3 on DC o.O).
Otherwise, it would be like looking at a roman galley and only think about the sails, but forgetting that there are two dozen men rowing underneath.

While thats i still say something funny here. Finally got around to extracting sega bass fishing 1 for dc, which is a model 3 port. Its pretty close to perfect. When compared to the models of sega marine fishing it seems marine fishing actually was somewhat upgraded from bass fishing graphically with higher quality models complexity wise. So definitely it seems to be able to match( maybe even slightly surpass)the model3 at least model detail wise.

I still believe it really is effort based. It probably just wasnt worth it for them to put in so much work for these short arcade conversions as opposed to chop here and there to get it up running.

Sega marine fishing



Sega bass fishing

cloofoofoo wrote:While thats i still say something funny here. Finally got around to extracting sega bass fishing 1 for dc, which is a model 3 port. Its pretty close to perfect. When compared to the models of sega marine fishing it seems marine fishing actually was somewhat upgraded from bass fishing graphically with higher quality models complexity wise. So definitely it seems to be able to match( maybe even slightly surpass)the model3 at least model detail wise.

I still believe it really is effort based. It probably just wasnt worth it for them to put in so much work for these short arcade conversions as opposed to chop here and there to get it up running.

When comparing polygon-counts, please also consider quadrilateral polygons on Model 3 vs triangle Polygons on Dreamcast. Just saying.

BTW, graphics in games are ALWAYS (!) driven by effort as well as hardware.
With enough effort, imagine what you could get out of Model 3! The amount of memory MB is in the hundreds, you get stuff like transparencies, z-buffering and anti-aliasing with no performance penalty, you have a HUGE amount of bandwidth, you got almost zero loading times and so on.

But the arcade is different from home systems since you usually can just upgrade the hardware. Only months after Model 3 came Model 3 Step 1.5, shortly followed by Step 2 after that. And shortly thereafter, Sega abandoned the too expensive and too complicated Model 3 and introduced Naomi.
With so few games released in the arcade, there's just no need (and no time!) for several years worth of optimizing as on a home system - you can just put new technology in. And that's what they did.
On the other hand, there are home systems where a lot more effort is required, since the games usually have a huge target market and several years of lifespan. Plenty of time to optimize your libraries.

I'll give one example:
Compare Ridge Racer 1 and Ridge Racer Type 4 on the PS1 for example. Ridge Racer 4 is arguably even better looking then Ridge Racer 1 in the arcade - even tho the latter ran on the mighty System 22 board!
Does that mean the PS1 is more powerful than System 22?
Now way in hell!
System 22 kicks PS1's ass so hard that it would land somewhere in another galaxy.
But as you said, effort is VERY important here, as is the general evolution of the 3D-knowledge-base in general, as I tried to explain.

Dreamcast also had the advantage of way easier programming.
Still, there are no Model 3-ports on the DC that are better graphically then in the arcade as far as I know. Only a bit worse.
Please note that I'm not saying that 1:1 conversions (minus the anti aliasing, of course, see Bass Fishing) wouldn't be possible with enough effort (we shall never know...). That is a different story.
I'm just trying to say Model 3 is a bit more powerful hardware, at least Step 2, since this was important for this topic of direct ports or even emulation.
That does not make the Dreamcast-conversions any worse, on the contrary, one should congratulate People like Genki to get the games at home at almost arcade quality.

The last thing I said is even more impressive since we did not get this level of almost arcade-quality at all in the time before Dreamcast. Dreamcast was really the first system to truly get arcade-quality at home.

(The only exception would be the Neo Geo, where you always had 1:1 at home, since the hardware specs were identical. But it came at high cost for each game, so it's a different story.)

Gabbyjay wrote:

cloofoofoo wrote:While thats i still say something funny here. Finally got around to extracting sega bass fishing 1 for dc, which is a model 3 port. Its pretty close to perfect. When compared to the models of sega marine fishing it seems marine fishing actually was somewhat upgraded from bass fishing graphically with higher quality models complexity wise. So definitely it seems to be able to match( maybe even slightly surpass)the model3 at least model detail wise.

I still believe it really is effort based. It probably just wasnt worth it for them to put in so much work for these short arcade conversions as opposed to chop here and there to get it up running.

When comparing polygon-counts, please also consider quadrilateral polygons on Model 3 vs triangle Polygons on Dreamcast. Just saying.

BTW, graphics in games are ALWAYS (!) driven by effort as well as hardware.
With enough effort, imagine what you could get out of Model 3! The amount of memory MB is in the hundreds, you get stuff like transparencies, z-buffering and anti-aliasing with no performance penalty, you have a HUGE amount of bandwidth, you got almost zero loading times and so on.

But the arcade is different from home systems since you usually can just upgrade the hardware. Only months after Model 3 came Model 3 Step 1.5, shortly followed by Step 2 after that. And shortly thereafter, Sega abandoned the too expensive and too complicated Model 3 and introduced Naomi.
With so few games released in the arcade, there's just no need (and no time!) for several years worth of optimizing as on a home system - you can just put new technology in. And that's what they did.
On the other hand, there are home systems where a lot more effort is required, since the games usually have a huge target market and several years of lifespan. Plenty of time to optimize your libraries.

I'll give one example:
Compare Ridge Racer 1 and Ridge Racer Type 4 on the PS1 for example. Ridge Racer 4 is arguably even better looking then Ridge Racer 1 in the arcade - even tho the latter ran on the mighty System 22 board!
Does that mean the PS1 is more powerful than System 22?
Now way in hell!
System 22 kicks PS1's ass so hard that it would land somewhere in another galaxy.
But as you said, effort is VERY important here, as is the general evolution of the 3D-knowledge-base in general, as I tried to explain.

Dreamcast also had the advantage of way easier programming.
Still, there are no Model 3-ports on the DC that are better graphically then in the arcade as far as I know. Only a bit worse.
Please note that I'm not saying that 1:1 conversions (minus the anti aliasing, of course, see Bass Fishing) wouldn't be possible with enough effort (we shall never know...). That is a different story.
I'm just trying to say Model 3 is a bit more powerful hardware, at least Step 2, since this was important for this topic of direct ports or even emulation.
That does not make the Dreamcast-conversions any worse, on the contrary, one should congratulate People like Genki to get the games at home at almost arcade quality.

The last thing I said is even more impressive since we did not get this level of almost arcade-quality at all in the time before Dreamcast. Dreamcast was really the first system to truly get arcade-quality at home.

(The only exception would be the Neo Geo, where you always had 1:1 at home, since the hardware specs were identical. But it came at high cost for each game, so it's a different story.)

The dreamcast if you didnt know wasnt actually all that easy to program for, sure the sdk libraries of ninja/kamui gave you some short cuts to get things running. The dreamcast didnt live long enough to get that optimized state otherwise we wouldn't have people like TapamN or moop that actually made sh4 cpu libraries that were many times faster than the official sdk in polygon transformation and lighting , instructions parallelism , which is ridiculous if you think about it.

Nothing is free, those transparencies will always eat fill rate. For example the author of model 3 emulator mentioned very little games actually used the hardware lod system and he couldnt see why unless there likely a performance penalty because in the end the games still uses cpu based lod systems.

Not sure why youre mentioning triangles vs quads the comparison up there if you read the numbers are both triangle vs triangle. And i wasnt able to get sega bass fishing 2 but pretty sure that one even trashes bass fishing wise even further.

I always bring up doa2 because it uses naomilib like vf3dc and fv2dc. Its running over 2x the detail of vf3 and 3x detail of fv2. Not sure how you not conclude 1:1 are very likely if the budget and effort was actually put into these.

I do agree with you on genki, they got virtua fighter 3 finished in a couple months with unfinished tools and iam sure it was mentioned with in 90% of polygon count intact. And virtua striker 2 was done really well, clever use of technical tricks got it looking really nice.

Sega rally 2 was very interesting, its pretty much devoid of lighting on model 3 except for the night stage but on dreamcast they added lighting to the track( which unfortunately makes it dark) and shadows that darken the car when in them and a red point light that lights up the track and car when the muffler goes off. So that was actually upgraded while detail was downgraded.

Fv2 is probably the worst model 3 effort, stages have severely chopped, missing sections and replaced with 2d billboards. Characters detail were massively reduced worst than vf3 . The light that turns on when you hit opponents is also gone.

cloofoofoo wrote:The dreamcast if you didnt know wasnt actually all that easy to program for

The whole point of Dreamcast's architecture was to facilitate easy programming.
They FINALLY settled on triangles, they used known and widespread hardware, the standard CPU-GPU RAM-VRAM-configuration and a GPU that came directly from the PC-sector.
Compare that to the complicated Saturn or even more so the complicated Model 3, which had none of these advantages (or - to a lesser degree - the PS2, which came after it).
Sega lost a lot of developer support due to the complicated hardware of the Saturn, that is widely known, and they learned from it. Launching another console system was risky and they had to make sure the got all the support they needed.

The dreamcast didnt live long enough to get that optimized state

That is correct, compared to systems like the PS1 the Dreamcast did not have as long a lifespan.
Still, we're talking of about 2.5 years from the japanese launch here and games were released even after that.
On model 3, we're talking about only 3 months from Step 1 to Step 1.5 and 6 months to Step 2. No time to optimize anything and also, no need to.

Not sure why youre mentioning triangles vs quads the comparison up there if you read the numbers are both triangle vs triangle.

Yes, I know. You were not comparing Model 3 to Dreamcast, but two Dreamcast-games.
It was only a reminder that should you compare M3 and DC, just consider this when doing the counting.

For example the author of model 3 emulator mentioned very little games actually used the hardware lod system and he couldnt see why unless there likely a performance penalty because in the end the games still uses cpu based lod systems.

That is only speculation. He mentioned other possible reasons like devs not accustomed to the hardware-LOD as well. Some games use it, others don't and some use it to do transparencies.
I think it's very impressive that Real3D had this feature, or things like the micro-texturing!

I do agree with you on genki, they got virtua fighter 3 finished in a couple months with unfinished tools

It's amazing how protective people are when it comes to the Dreamcast. Yes VF3 on the DC was troubled and there are many problems when you release a game at launch.
But so was Model 3's release, which was also highly troubled, with the final specs only arriving a few months before the release. This went so far that they even had to release a game on Model-2 instead since the hardware arrived so late.
So no smooth ride here either, and I find it to be very selective if people only look at the difficulties for the Dreamcast but ignoring them on other platforms.

I always bring up doa2 because it uses naomilib like vf3dc and fv2dc. Its running over 2x the detail of vf3 and 3x detail of fv2. Not sure how you not conclude 1:1 are very likely if the budget and effort was actually put into these.

Dead or Alive 2 is a PRIME example of what I'm trying to explain here.
The original game ran on Naomi, which is without doubt more powerful than the Dreamcast.
It has twice the RAM, twice the VRAM (both of which are THE limiting factors in Dreamcast's polygon and texture-capabilities), more Sound-RAM, higher bandwidth due to the higher clocked memory-bus and also has almost zero loading times, even when the GD-drive is used. The rest is identical, so we know Naomi is clearly the more powerful system.
Now if you compare DoA2 (and many other games!) on both systems, they almost look identical!
(The only differences I could EVER spot are ever so slight differences in the mountains in the background and some textures are slightly less sharp.)
Considering the VAST memory advantage Naomi has, you should think they had to make cuts all around, the games SHOULD look a lot worse on Dreamcast, right? But they don't.
Does that mean Dreamcast is as powerful as Naomi? Nope. It means the games were either not very well optimized on Naomi (texture compression, for example) or they were very well optimized on DC. Or both.

Sega rally 2 was very interesting, its pretty much devoid of lighting on model 3 except for the night stage but on dreamcast they added lighting to the track( which unfortunately makes it dark) and shadows that darken the car when in them and a red point light that lights up the track and car when the muffler goes off. So that was actually upgraded while detail was downgraded.

I remember Riviera having decent lighting on the original arcade hardware. Not sure about the other tracks as they are in daylight anyway, but maybe they added some lighting. They did similar with RR1 on the PS1 as well.
In total, SR2 on Dreamcast was really no match for the arcade.
Arcade had higher draw distance, more detailed textures, higher polygon count on cars and landscape, anti-aliasing and most of all, a pretty stable framerate which Dreamcast did not.
It is also worth mentioning that if you activate the 60fps-mode on the Dreamcast, a LOT more details are cut just to have the same framerate as the arcade (and it still doesn't quite reach it).