Picture of Sega's 3Dfx-based 'BlackBelt' console

[Megadrive1988]

3Dfx based Sega BlackBelt
http://www.users.on.net/~iterations/images/Blackbelt.jpg

assuming this is real, this is the console designed by SoA, SegaSoft and 3Dfx that LOST to the Dural/Katana/Dreamcast.

circa 1997.

found it in this assembler thread

http://www.assemblergames.com/phpbb2/viewtopic.php?t=170


cool find huh 8) :shock:

[PC-Engine]

Hey that's pretty neat.

[zurich]

Fugly as sin.. and what's with the cartidge slot in the back?

[Sonic]

Memory cartridge/upgrade

Just a guess.

:wink:

[Guden Oden]

It looks just bad, and the recessed joypad ports are a disaster from an ergonomic standpoint. Assume this thing stands on a knee-high bench or something, you'd have to lean way down to find where to stick the connector for your pad... Idiotic. It doesn't even hide the "unsightly" ports when one is sitting down in front of the TV. Simply bad design, lucky us it never got made.

Oh, and it's friggin HUGE too by the looks of it, nowhere near the slender thing that is the DC. Must all American-designed consoles follow the "bigger is better" school of thought? :roll:


:lol:

[Megadrive1988]

it reminds me of one of the M2 casings.

http://ryangenno.tripod.com/images/m2_1217b.jpg

http://ryangenno.tripod.com/images/m2_1217b.jpg

[hey69]

it reminds me of the playstation (the grey box)

[oddfellow]

Nice find 8)

[Fox5]

Looks like this dreamcast had a reset button, only two controller ports, and possibly a cart drive on top? Play old genesis games on it maybe? Or plug in a 2nd graphics card in a sli sort of way and have a dreamcast sli? Maybe that was the hardware upgrade I heard about for the dreamcast....

BTW, why is the M2's pad a rip off of the n64's?

[Silanda]

BTW, why is the M2's pad a rip off of the n64's?

Panasonic liked the N64 design and ripped it off, at least that's what the story was back in the day.

[Megadrive1988]

Looks like this dreamcast had a reset button, only two controller ports, and possibly a cart drive on top? Play old genesis games on it maybe? Or plug in a 2nd graphics card in a sli sort of way and have a dreamcast sli? Maybe that was the hardware upgrade I heard about for the dreamcast....


well it looks like BlackBelt did have a cartridge slot ala Saturn. probably for RAM carts, memory carts. probably not for Genesis compatibility, but that would've been very nice.

your thoughts about this slot being used for an SLI-style upgrade path is thought provoking. it was stated in Next Generation magazine that the Saturn2 might be an upgrade for Saturn, using the existing Saturn as a power supply, CD drive and I/O device. so I assume the Saturn2 upgrade would've used the Saturn cartridge slot. the BlackBelt might have had similar capability. upgrade to SLI Banshee2/Voodoo3 (the likely 3D chip in BlackBelt) or perhaps Voodoo VSA (Voodoo4/5).

[Fox5]

Hmm, if it was voodoo3, I don't see why sega didn't go for it. Voodoo 3s were very cheap even when they were new(mostly because 3dfx was cutting their profit margins thin, but it worked out well for the consumers, I think voodoo3 was the best selling retail card at the time, and may still hold the record, though perhaps geforce 2mx or geforce4mx may have taken that over) and probably would have outperformed what was in the dc 9 out of 10 times. If it was banshee or voodoo2 then the choice isn't so clear cut, and would probably break even in situations where one is faster than the dc's chip, or lose out, plus I think voodoo2 and banshee were kind of expensive when they were new, and voodoo3 wouldn't have been available for a 1998 Japan launch.

[Megadrive1988]

the CLX / PowerVR2DC (specific variant of PowerVR2 for Dreamcast) was much superior to Banshee or Voodoo2, and held its own against the Voodoo3 which was more powerful on paper. It's a toss up between the Voodoo3 and PowerVR2DC / CLX. in some areas the Voodoo3 is faster, in other areas the PowerVR2DC is better.

but there's little doubt the PowerVR2DC is superior in every way to the Voodoo2 and especially the Banshee.


can't wait until Simon F joins this thread 8)

[Fox5]

the CLX / PowerVR2DC (specific variant of PowerVR2 for Dreamcast) was much superior to Banshee or Voodoo2, and held its own against the Voodoo3 which was more powerful on paper. It's a toss up between the Voodoo3 and PowerVR2DC / CLX. in some areas the Voodoo3 is faster, in other areas the PowerVR2DC is better.

but there's little doubt the PowerVR2DC is superior in every way to the Voodoo2 and especially the Banshee.


can't wait until Simon F joins this thread 8)

My main counter argument to this area is that the voodoo3 easily crushed the neon250 in basically everything(especially if we add glide in), and that no dreamcast port of a pc game(and I believe any dreamcast game ported to the pc) ran as well on the dreamcast as a pc with a voodoo3. Now maybe sonic adventure 2 and shenmue were beyond what a voodoo3 could do, but they were the exception, not the rule, and shenmue was pushing the dc harder than it could handle as well, once the framerate dropped below 30 fps, we could say a voodoo3 could do it and also have low framerates.

[Megadrive1988]

I won't agrue with that really. in some cases, I know Voodoo3 PC outgunned the Dreamcast (in Sega Rally 2 for instance which was 60fps or closer to it on a Voodoo3 PC than on DC.

the Neon250 though, is a weaker chip than PowerVR2DC. Neon250 is clocked 25 Mhz faster but does not have as many pixel elements as the DC PowerVR, IIRC. there are probably at least several other major differences between Neon250 and PowerVR2DC that I am unqualified to talk about (Simon where are you!)

no question that Voodoo3 outpaces Neon250 in most areas. although even Neon250 had 32-Bit color, I believe.

[Fox5]

And what difference did 32bit color make, especially in those days? Even in modern games I have trouble telling the difference between 32bit and 16 bit color, and back then I couldn't tell the difference at all, especially since most games weren't made for 32bit color. Of course, since the dreamcast was a console it had an opportunity to change that, but wouldn't dc be memory limited to actually do it? I don't think tbr hardware suffers the performance hit, but it still requires double the memory, doesn't it? Anyhow, I think glide would have made more of a difference, from what was in the link to the forum one guy said it translated better to the actual hardware of the voodoo, and I think it may have had lower cpu utilization, which would matter on a console with a slow cpu. Plus glide was capable of effects that couldn't be done in direct3d, or at least not as fast. After the death of glide and the take over of opengl and direct3d, I felt 3d graphics took a bit of a step back, and I've only felt glide has been fully replaced with directx9.(but every once in a while I check out the emu scene, and it's still amazing to see that glide is still the best api for emulation, not only is it the fastest but there tend to be tons of effects that can only be replicated with it and what I believe are called shadow buffers, or with extremely buggy dx9 pixel shading) Of course, powervr had its own api, but I'm not sure how it compared, but I seem to remember pc benchmarks using it were nothing special, other than that they allowed full usage of its features and didn't have visual errors.

[Megadrive1988]

no question that as far as PCs, the Voodoo3 is better than the Neon250. especially where PowerVR's advantages were less likely to be used. Glided was far superior to DirectX back then, too. but the PowerVR2DC vs Voodoo3 in a console is a tough choice. I might've even been swayed to the Voodoo3 in Dreamcast if it could've allowed for better framerates in early DC games.

[Guden Oden]

Fox5,

You're smoking some seriously heavy rose-colored glasses in the last couple posts you've made, I haven't heard such BS in a long good while... :D

What EFFECTS have you seen Glide make that no other cards could do until DX9??? They don't exist!

The Voodoo3 was seriously stone-age tech compared to PVR2DC/Neon250 (these were the exact same chips tech-wise, except Neon250 was faster and had AGP interface). V3 had faster on-paper fillrate, but lack of overdraw elimination and general inefficiency of old hardware pretty much eliminated that advantage, especially when transparencies are involved.

V3 lacks not only the fairly pointless 32-bit color on textures and framebuffers, it also lacks the on-chip 32-bit rendering and one-step downsampling to 16-bit of PowerVR's. It doesn't have modifier volumes either, or antialiasing, or anisotropic filtering. Heck, it didn't even have true trilinear as far I know. On the other hand, it requires much faster (and more expensive) RAM chips on a memory bus twice the width to offer competitive performance, and it runs much hotter too. See what happens to an immediate-mode renderer once you halve its memory bus width; performance dives down into the toilet.

V3 may have been the most sold retail card at the time but that's a worthless thing to brag about; 3dfx was the *only* manufacturer of Voodoo cards at that time, lack of competition made them win that badge. If you were to look at which chip made most retail sales, it'd been the TNT2 instead, which was available from a dozen + manufacturers, hence driving down the sales volumes of each individual manufacturer.

So when you say you believe a V3 would smoke PVRDC 9 out of 10 times, I believe you should reconsider... ;)

[Fafalada]

and that no dreamcast port of a pc game(and I believe any dreamcast game ported to the pc) ran as well on the dreamcast as a pc with a voodoo3.
Run those PC games on a comparable CPU if you're arguing graphic chips (say, something in the range of a 400-500mhz Celeron) and then let's see how well they run.

And what difference did 32bit color make, especially in those days?
A lot of difference on consoles - because it's actually been used from day one. If you want more tangible stuff, texture compression and modifier volumes both made a difference.
And even featuresets aside, I would argue that a V3 chipset would need pairing with a stronger CPU then SH4 was, to work comparably well. As well as more expensive memory.

As for glide, it was just a thinner software layer then other APIs are. In other words, it lets you get more intimate with hardware then most APIs do - but that's pretty much what you get on most consoles by default.

Heck, it didn't even have true trilinear as far I know.
IIRC no PC card at the time did (TNTs used the dither hack too). Then again PVRDC trilinear was as good as useless also - and the games demonstrated that by pretty much never using it :P

[zurich]

Fox5,

So I take it you were a 3dfx fan back in the day? I feel like I just stepped through a timewarp to 1998 :lol:

[Fox5]

Fox5,

You're smoking some seriously heavy rose-colored glasses in the last couple posts you've made, I haven't heard such BS in a long good while... :D

What EFFECTS have you seen Glide make that no other cards could do until DX9??? They don't exist!

The Voodoo3 was seriously stone-age tech compared to PVR2DC/Neon250 (these were the exact same chips tech-wise, except Neon250 was faster and had AGP interface). V3 had faster on-paper fillrate, but lack of overdraw elimination and general inefficiency of old hardware pretty much eliminated that advantage, especially when transparencies are involved.

V3 lacks not only the fairly pointless 32-bit color on textures and framebuffers, it also lacks the on-chip 32-bit rendering and one-step downsampling to 16-bit of PowerVR's. It doesn't have modifier volumes either, or antialiasing, or anisotropic filtering. Heck, it didn't even have true trilinear as far I know. On the other hand, it requires much faster (and more expensive) RAM chips on a memory bus twice the width to offer competitive performance, and it runs much hotter too. See what happens to an immediate-mode renderer once you halve its memory bus width; performance dives down into the toilet.

V3 may have been the most sold retail card at the time but that's a worthless thing to brag about; 3dfx was the *only* manufacturer of Voodoo cards at that time, lack of competition made them win that badge. If you were to look at which chip made most retail sales, it'd been the TNT2 instead, which was available from a dozen + manufacturers, hence driving down the sales volumes of each individual manufacturer.

So when you say you believe a V3 would smoke PVRDC 9 out of 10 times, I believe you should reconsider... ;)

Sure, it's easy to talk about overdraw elimination when you say "only" an overdraw of 2x..but on the other hand that means every pixel on the screen is blocked by one other pixel, so I don't think 2x is as common as is said. And what old hardware? Voodoo3 was released 1999, pvr2dc hardware was prototyped in 1997, so voodoo3 was at least a year newer, and as for the cores both were based on, I think the voodoo core was only about 6 months older.

Voodoo3 had 22bit color or something, it easily matched nvidia's 32 bit color in most games.

V3 had trilinear and anisotropic, but I don't think it could do them at the same time as multitexturing. It had edge AA as well...sure pvr2dc had super sampling, but come on, how many games could actually take that performance/memory hit? Voodoo3 technically could have done super sampling, voodoo2 did.(not sure if the software that did it was compatible with voodoo3 or not, but performance died while doing it)

I don't know how the tnt2 sold at retail, but the voodoo5 during its first 5 months only sold slightly worse in retail than the combined sales of every geforce2 card maker, and slightly better than the combined top 2.

As for effects glide can do that couldn't be done till dx9, I just look at emulation, the pixel shader like effects of the n64 I've only seen done using glide and buggy dx9. The glide version usually uses the framebuffer to do it, because for whatever reason voodoo cards were extremely fast at accessing that, and the voodoo 4 and 5s could load the entire thing into a shadow buffer or something, and voodoo3 and I think banshee had partial support for shadow buffers.(don't know what a shadow buffer is, but that's how the plugin's site describes it)

As for v3 smoking a pvrdc....well show me something to prove it, not just looking at something on a dc and saying "V3 couldn't do that." Of course, my idea of performance on a voodoo3 may be a bit skewed, during winter I had mine overclocked to over 200mhz on both core and ram(don't think you could adjust them seperately) and in the high 190s during the summer.(probably why it died a few months after I sold it to someone once I got a geforce 3) Still, voodoo3 crushes the neon250 in basically everything, at best it can only hope to equal the voodoo3 under rare circumstances(I think quake3 with the neon250's quake 3 minigl driver was the only time I've seen a benchmark where it happened, and forget space and flight sims), and show me a port from the dc that came out while the dc still was alive(to make sure it doesn't have overly high hardware specs) that a voodoo3 couldn't do better. I played the typing of the dead demo on my pc, seemed to run just as fast if not faster than the dc version. Not sure how house of the dead 2 and crazy taxi would handle, I think they came out later and could be bloatware.

http://graphics.tomshardware.com/graphic/19990929/neon_250-05.html
Check this, on a p3 550mhz(while we're on this pro-dc ride, we can argue that a 200 mhz sh-4 with optimizations performs as well or better than a p3 550mhz, right? at least in floating point operations), neon 250 can sort of almost keep up at low res in shogo, crushed at high res. Same for expendable, but not as much. In quake 3 it can keep up fairly well, though it we go by what tom says, despite the tbr, it has memory bandwidth issues and that's why performance dies at high reses.
Performance is way behind for descent 3, and if I recall correctly, dc had quite a few games with wide open areas and little overdraw, so in those cases voodoo3 would have been a much better choice.

[Fox5]

Fox5,

So I take it you were a 3dfx fan back in the day? I feel like I just stepped through a timewarp to 1998 :lol:

Then? I'm still one now, if 3dfx was still making cards that were competitive I'd buy them. And a voodoo5 card is still the number 1 card for emulation.
Difference between a voodoo3 and a tnt2 today...niether one can run modern games, but at least a voodoo3 can still run emulators and the latest glide plugins. Glide support alone makes a voodoo3 card worth as much today as a geforce or geforce 2mx, and FSAA makes a voodoo5 card worth slightly more. IMO of course, but none of these cards can run very much today.

[zurich]

Fox5,

So I take it you were a 3dfx fan back in the day? I feel like I just stepped through a timewarp to 1998 :lol:

Then? I'm still one now, if 3dfx was still making cards that were competitive I'd buy them. And a voodoo5 card is still the number 1 card for emulation.
Difference between a voodoo3 and a tnt2 today...niether one can run modern games, but at least a voodoo3 can still run emulators and the latest glide plugins. Glide support alone makes a voodoo3 card worth as much today as a geforce or geforce 2mx, and FSAA makes a voodoo5 card worth slightly more. IMO of course, but none of these cards can run very much today.

Okay now I'm DEFINITELY back in 1998.

[ShinHoshi]

V3 was an amazingly fast card...I still remember how a friend used it to run the first N64 emulators (with a P150) and it was much more faster than my GF 2MX(coupled with a P3 667). (using Glide of Course).

[PC-Engine]

The exterior design isn't as simple and clean as the production DC but it isn't ugly either. PS2 looks even uglier when laid horizontally.

And what difference did 32bit color make, especially in those days?

Well when I played TR on the PCX1 and switched from 16-bit color to 24-bit color, I could see a difference.

it reminds me of the playstation (the grey box)

...or a SNES or a NES or a Famicom or a SuperFamicom...

[Clashman]

Maybe I'm wrong on this, but I was under the impression that many if not most DC games used 16 bit color anyway, so wouldn't that really be a moot point? Also, given the amount of texture memory available, I don't think too many games would have cursed 3dfx's other main curse, (256x256 texture size limit), but maybe I'm missing something here.

[PC-Engine]

V3 lacks not only the fairly pointless 32-bit color on textures and framebuffers, it also lacks the on-chip 32-bit rendering and one-step downsampling to 16-bit of PowerVR's.

[Sonic]

Dreamcast has VQ (vector quantization) texture compression and it usually had a space saving of 5:1 but often reach 8:1 in a few cases. This allowed for much higher resolution than a non compressed texture. VQ is what Sonic Adventure and Sonic Adventure 2 use as well as Shenmue as well as just about 90% of Dreamcast games.

[Simon F]

can't wait until Simon F joins this thread 8)
It's all been said before (http://www.beyond3d.com/forum/viewtopic.php?p=54619#54619)

[Rodéric]

It looks really bad, the Dreamcast design is way neater.
I really doubt this is the alternate design to the dreamcast, missing pad slots and looks generally bad for something design in 1997+...

heck even the XBox looks better than that ^^
(Everything IMO of course)

[Guden Oden]

Sure, it's easy to talk about overdraw elimination when you say "only" an overdraw of 2x..but on the other hand that means every pixel on the screen is blocked by one other pixel, so I don't think 2x is as common as is said.

Of course it varies from title to title, but 2x is probably a pretty good average across genres. Typically a game with a horizontal view would score higher than a game with a vertical view. I shudder to think of what the overdraw is in a game like Jak2 for example inside the city...

And what old hardware? Voodoo3 was released 1999, pvr2dc hardware was prototyped in 1997, so voodoo3 was at least a year newer

"Newer"? The V3 architecture goes right back to the voodoo graphics chipset (hence all the weird restrictions like 256-pixel textures, 16-bit pixels only etc).

Voodoo3 had 22bit color or something, it easily matched nvidia's 32 bit color in most games.

That's not true.

V3 had trilinear and anisotropic

No it did not. Even the VSA100 couldn't do proper trilinear if it was multitexturing, and it had never even heard of aniso.

It had edge AA as well

...Which was sorting-order dependent and a fuxxing bitch to use and pretty much worthless. Nobody likes edge AA.

Voodoo3 technically could have done super sampling, voodoo2 did.

V2 never had supersampling in any driver version, and no software support whatsoever.

I don't know how the tnt2 sold at retail, but the voodoo5 during its first 5 months

That's pretty much its LAST 5 months too, as it turned out. :lol:

only sold slightly worse in retail than the combined sales of every geforce2 card maker, and slightly better than the combined top 2.

Guess that is retail figures for the US only (where 3dfx had its greatest retail presence). Still, retail is such a small part of the total market that bragging about selling slightly better than the top 2 from the other camp is sort of like bragging about winning a race in the 100 meters dash over a bunch of geriatrics. You're still getting owned overall.

As for effects glide can do that couldn't be done till dx9, I just look at emulation, the pixel shader like effects of the n64 I've only seen done using glide and buggy dx9.

WHAT pixel shader-like effect? The N64 has no games with any such effects.

and voodoo3 and I think banshee had partial support for shadow buffers.

Shadow buffers came in DX7, first chip that supported it was the original ATi Radeon (R100). The voodoo series totally lacks any such stuff, and it has nothing to do with pixel shading.

As for v3 smoking a pvrdc....well show me something to prove it, not just looking at something on a dc and saying "V3 couldn't do that."

Difficult, since there are no consoles with a 200MHz SH4 CPU and a V3 graphics chip! It's clear though that the feature-list of the PVR2DC is much longer than V3's, and the effective raw performance is not that dissimilar when factoring in overdraw and transparencies and such.

You have to realize that the 183Mpix/s fillrate figure is only a best-case scenario, in actuality it is much lower, if you run the 3dmark fillrate tester you will get a MUCH lower score. Just think of how much performance is lost every time you have to do a read-modify-write operation against framebuffer DRAM on a V3 (happens every time you draw a transparent pixel). On a PowerVR chip you write against on-die SRAM with little or no penalty associated to it except the extra clock cycle it requires to draw another transparent polygon layer.

Still, voodoo3 crushes the neon250 in basically everything

You have to take into account the notoriously rotten drivers for the Neon. PC graphics is a completely different beast compared to consoles, DX in particular was not suited at all to the Neon's way of handling things so that it ran slow should not be surprising. And, when it wasn't slow, it tended to be buggy instead. Or both. The amount of compatibility option checkboxes in the driver panel for the Neon that one could (and often HAD) to tweak with was nothing but staggering! If PowerVR had had a popular custom API like 3dfx did with glide, I'm sure the situation would have been rather different, but that never was the case. Take a look at Kyro instead which is geared fully towards PC APIs (which the Neon was not) and had GOOD drivers, and then compare that with its contemporaries to see what a TBDR can do with vastly inferior resources.

[Megadrive1988]

It is pretty clear Sega made the 2nd best choice (Real3D would've been the best IMHO). between PowerVR and 3Dfx though, it was the best choice they could make. the 3Dfx chip in BlackBelt would have been a Voodoo3 at most. possibly something less than Voodoo3. If so, then the PowerVR2DC has even more of an advantage since nothing short of a Voodoo3 can even begin to stand up to the PowerVR2DC. Sega made its decision based on politics however. Japanese politics. even if BlackBelt beat out Dural performance wise, Sega was already in bed with NEC who would be fabbing Videologic's design. Sega had already made its choice before the BlackBelt-Dural showdown.

[Simon F]

It is pretty clear Sega made the 2nd best choice (Real3D would've been the best IMHO).
Please explain the logic underlying that conclusion.

[archie4oz]

Sega made its decision based on politics however. Japanese politics. even if BlackBelt beat out Dural performance wise, Sega was already in bed with NEC who would be fabbing Videologic's design.

Does anybody get tired of hearing the political BS angle? I swear the 3dfx/PC fanbase sure gets bent outa shape when their untouchable idol gets beat by a nifty little tiler... Meh, just sour grapes...

Please explain the logic underlying that conclusion.

Yeah, this I gotta hear too...

[Sonic]

SEGA chose the best GPU at the time from a price/performance/feature ratio. I would only call it politics if once considers taking actual performance and features of a GPU into account. 3dfx would have been the worst choice by far since they failed to deliver what SEGA wanted with their hybrid chip. The PVR2DC is a very capable chip that won out mainly because it is better all around. The Dreamcast sure did some amazing things with its 100 Megapixel fill rate. Overdraw in a real world scenario is usually greater than 2x as that is a very conservative number.

Could any 3dfx chip at the time to bump mapping, anisotropic filtering, or even 32 bit color?

Real3D wasn't much of an option since they didn't offer anything to the table either. SEGA didn't like the performance of their PC GPU's and ot wpi;d have been too costly anyway. It was fine for Model 3, but the Real3D chips in Model 3 are way faster than the ones used in PC cards. That's also not to mention the manufacturing process for those chips were huge!

[Megadrive1988]

I tend not to believe that Sega selected PowerVR Dural for political reasons, but I said it anyway because that is what was said so often.
I personally believe they chose PowerVR because it offered by far the best price/performance ratio. beyond Model 3 class performance for a tiny fraction of the cost.

I am certain that if Lockheed had come up with a console friendly chip, say a hypothetical "R3D-500", with geometry engine, 2-4 pipes, not based on i740 but more along the lines of R3D-Pro 1000, on .25 micron process
with 3-5 million pps performance with all features on, Sega would've jumped on it. especially with Yu Suzuki's enthusiasm for Real3D. but that was not to be, and Sega went to PowerVR2, the best option they had.

[Fox5]

"Newer"? The V3 architecture goes right back to the voodoo graphics chipset (hence all the weird restrictions like 256-pixel textures, 16-bit pixels only etc).


I was referring to the actual voodoo3 chip was finished after pvr2dc. And how old was the voodoo graphics chipset? 1994, 1995, or 1996?

Quote:
Voodoo3 had 22bit color or something, it easily matched nvidia's 32 bit color in most games.


That's not true.

It was at the time. Or can you tell the difference between halflife running in 16bit on a voodoo 3 and halflife running in 32 bit on a tnt2?

No it did not. Even the VSA100 couldn't do proper trilinear if it was multitexturing, and it had never even heard of aniso.


Not in official 3dfx drivers as far as I know, but 3rd party drivers did add the options.

...Which was sorting-order dependent and a fuxxing bitch to use and pretty much worthless. Nobody likes edge AA.

With some 3rd party drivers I was able to force it in all games. It sometimes worked.

V2 never had supersampling in any driver version, and no software support whatsoever.


Wasn't in the drivers. I'm fairly certain it was something 3dfx released around the time the tnt was coming out, to allow the voodoo 2 to do super sampling just to claim it was possible....however it had horrible performance and the readme for it even said it was common for the fps to go under 5 frames per second. At least I'm pretty sure I saw the program, I can't find it now, and I never tried it out myself.

WHAT pixel shader-like effect? The N64 has no games with any such effects.

The shininess on some objects in conker's bad fur day(such as the logo at the start), and the shininess on skarmory in pokemon stadium 2. I believe skarmory even sort of reflects the colors of objects/lights arounnd it. Also various other effects.



This feature also allowed me to emulate many things which are hardly possible to emulate with usual frame buffer emulation, e.g. dynamic shadows. HWFBE is fully supported only by Voodoo4/5 and partially by Voodoo3 and Banshee. Read readme.txt for details. Many thanks to KoolSmoky who gave me Voodoo5 card and hints how to implement this feature.

Looks like I was wrong about shadow buffers, it's hardware frame buffer used for shadows.

You have to realize that the 183Mpix/s fillrate figure is only a best-case scenario, in actuality it is much lower, if you run the 3dmark fillrate tester you will get a MUCH lower score. Just think of how much performance is lost every time you have to do a read-modify-write operation against framebuffer DRAM on a V3 (happens every time you draw a transparent pixel). On a PowerVR chip you write against on-die SRAM with little or no penalty associated to it except the extra clock cycle it requires to draw another transparent polygon layer.

Just looked up mym old 3dmark scores, 195.9 megatexels fill rate for single texturing, and 353 for multi texturing. Not sure what my mhz was at the time, but I'd say that's very very close to 100%. Assuming I had it at 200 mhz, it would be 200mega texels, and even the best voodoo3s didn't clock much over 200 mhz. Multi texturing rate is a little low though. So multi texturing was a weak point for the voodoo3, doesn't look like single texturing was, how about the pvr2dc?
Looked up some results for the neon 250, one was 189.8 single texturing, and 834.7 multitexturing(probably overclocked, but so the neon250 does kick ass in multi texturing...any examples of that on dc? obviously not a lot on pc, or the neon 250 would have done much better), and the other has 130.5 and 704.7, probably running at stock speeds.

Could any 3dfx chip at the time to bump mapping, anisotropic filtering, or even 32 bit color?

It supported some form of bump mapping, but same that any computer can use.(like the most basic 3d mark bump mapping test)
AF I believe it could, but I'm not positive.
32 bit internal color too, doesn't mean 3dfx couldn't have produced a version with 32 bit external color if they desired it. One understanding I had was that the voodoo3 was what was being designed for dreamcast, but after they lost the sega contract they basically stopped working on adding all the features sega wanted and just released it. However, with all the features sega wanted, it would significantly more expensive than the voodoo3 was.(2-3x the price) Not sure how it would have compared to the neon250 in price though.(actually, the story I heard didn't mention sega at all, just that 3dfx had planned a lot more for the voodoo3 and cut it for cost reasons, but it's a nice story for a 3dfx zealot, right?)

[Fox5]

I tend not to believe that Sega selected PowerVR Dural for political reasons, but I said it anyway because that is what was said so often.
I personally believe they chose PowerVR because it offered by far the best price/performance ratio. beyond Model 3 class performance for a tiny fraction of the cost.

Not so sure about that. Was vf3 the only model 3 port to dreamcast? It had something like half the graphical quality of the model 3 version, and I'm not sure if any dreamcast games ever exceeded virtua fighter 3 arcade in graphics.(haven't had enough experience with vf3 arcade to decide....that harley davidson game was ugly though)

[Fox5]

Oh, just a note, with my voodoo3 card, in the high polygon count(1 light) it got 3.3 million triangles(out of a theoretically rougly 8 million) and 1 million in the 8 lights test. That was on a 1.4 gig athlon, but back on a cpu of the day(with probably a slightly less overclock, but not much since the mtexel fillrate is still 191.9) the count was 2.1 and .7.
On 1.3 gig athlons the neon 250 did 2.4 and 1.2, out of I suppose a rougly 8 million possible polygon count. Strangely enough though, in the test that had the low fillrate, it did 4 million and 1.9 million. Oh I see, that one was done at lower resolution. Both systems appears to be using overcloked cpus as well, as their fsb was 104, while my 1.4ghz is listed as only 100 fsb.

Well, looks like the neon250/powervr2 dc fairs better under more complex situations, but the voodoo3 is faster under simpler situations.

[archie4oz]

Was vf3 the only model 3 port to dreamcast? It had something like half the graphical quality of the model 3 version, and I'm not sure if any dreamcast games ever exceeded virtua fighter 3 arcade in graphics.

Wow, these are some seriously rose colored glasses! VF3tb was a total pile of crap as a port, and Soul Calibur was far more graphically impressive (and that was an up-port from the Model 12). The only real nice thing about the model 3 boards (preferably the later variants) was having a dedicated geometry engine... In any case the DC pretty much owns the Model 3...

BTW, what's the setup performance for a Voodoo2/3? I know the CLX setup performance is pretty spiffy (well in excess of what the SH4 can feed it)...

[Megadrive1988]

Not so sure about that. Was vf3 the only model 3 port to dreamcast? It had something like half the graphical quality of the model 3 version, and I'm not sure if any dreamcast games ever exceeded virtua fighter 3 arcade in graphics.(haven't had enough experience with vf3 arcade to decide....that harley davidson game was ugly though)

VF3tb was the first Model 3 conversion to DC, but certainly not the only one. basicly, it looks roughly half as good as the arcade because, a.) it was coded in about 6 months. b.) it was pre-first generation software, being coded on Katana Step 4 dev kit with around half the performance of the final PowerVR2DC silicon (i think) c.) it was coded by Genki, and not AM2.

other Model 3 conversions to DC:

Sega Rally 2
Virtua Striker 2 '99 (possibly the NAOMI version)
Sega Bass Fishing
Virtual On 2 (VOOT)
Fighting Vipers 2


the port of VOOT to Dreamcast, which looks 90-95% the same as the arcade, shows the Dreamcast was capable of handling Model 3 ports if enough time and effort was spent on it. I think Sega Bass Fishing looks roughly the same as the arcade also. VF3 and Sega Rally 2 were first or pre-first generation software that was very much rushed. Dreamcast is capable of surpassing Model 3 noticablly ,in some areas, when the right effort is put in. yes, there are areas where DC does not match Model 3 (the use of anti-aliasing, steady 60fps framerates with all effects on)

[Fox5]

Was vf3 the only model 3 port to dreamcast? It had something like half the graphical quality of the model 3 version, and I'm not sure if any dreamcast games ever exceeded virtua fighter 3 arcade in graphics.

Wow, these are some seriously rose colored glasses! VF3tb was a total pile of crap as a port, and Soul Calibur was far more graphically impressive (and that was an up-port from the Model 12). The only real nice thing about the model 3 boards (preferably the later variants) was having a dedicated geometry engine... In any case the DC pretty much owns the Model 3...

BTW, what's the setup performance for a Voodoo2/3? I know the CLX setup performance is pretty spiffy (well in excess of what the SH4 can feed it)...

I never had any experience with vf3tb on dreamcast, all I know is it didn't look as good as vf3 in the arcades.

What's setup performance mean? I'm not familiar with the term and couldn't find any information on it. If you mean the polygon rate, I've heard clx could do around 5-8, and the voodoo3 could do 6 with the 2000, and 8 with the 3500.

BTW, it may not be quite fair to compare model 3 with all effects on to the dc, sincec wouldn't the dc have more raw power if model 3 gets effects for free? And I didn't know voot was a model 3 port, I thought more like model 2....I like the game, but it's not very good looking at all. Simple textures and effects, and polygon amount isn't that high. Also, 2 player goes into ultra small screen mode I think.(may have had an option to stretch out the image though)

[Megadrive1988]

I think he's referring to triangle setup performance. something that Voodoo2/Banshee/Voodoo3 and PowerVR2 did, that Voodoo1 and PowerVR PCX1/PCX2 did not do (3Dfx said Voodoo1 did 2/3 of the triangle setup)


about VOOT, it was indeed a Model 3 game. Model 3 Step 2 actually. and you could link 2 DCs together so each player has a full screen just like the arcade. it's probably the most impressive Model 3 to DC conversion/port. And I explained why VF3tb didn't look as good as the arcade. still DC VF3 was closer to the arcade Model 3 VF3 than VF2 on Saturn compared to the Model 2 VF2. in otherwords, DC Model 3 conversions were closer to the arcade than Saturn Model 2 conversions.

[Guden Oden]

I was referring to the actual voodoo3 chip was finished after pvr2dc.

Yes, I know that, but the basic design of the chip is still going back to the very first voodoo. Just like the geforce 4 have similarities stemming way back to the tnt.

And how old was the voodoo graphics chipset?

As it was released in 96 I believe it had to have been completely finished & debugged by then.

That's not true.
It was at the time.

No, V3 never had 22-bit color, it was a 16-bit device through and through in 3D mode.

Or can you tell the difference between halflife running in 16bit on a voodoo 3 and halflife running in 32 bit on a tnt2?

Was too long since I played anything on a V3, I couldn't say with any certainty, but that still doesn't mean the V3 have 22-bit color, because it does not.

Not in official 3dfx drivers as far as I know, but 3rd party drivers did add the options.

Look... The chip itself can't do proper trilinear, much less aniso, so it doesn't matter what checkboxes the driver presents.

With some 3rd party drivers I was able to force it in all games. It sometimes worked.

Like I said, it was a bitch and nobody liked it. My Rage128 can also do edge AA, but software support prevents it from working properly. It's not a "real" kind of AA.

Wasn't in the drivers. I'm fairly certain it was something 3dfx released around the time the tnt was coming out

You must be thinking of bumpmapping instead. 3dfx wanted people to think their voodoos could do it, but what they offered wasn't proper dot3 bm but a multitexturing trick instead (emboss). The V2 (or 3 for that matter) never offered SSAA.

WHAT pixel shader-like effect?The shininess on some objects in conker's bad fur day(such as the logo at the start)

That's environment mapping. Pretty much ANY 3D hardware can do that, all that's needed for that effect is a texture that wraps seamlessly in all directions and then changing the texture coordinates on a frame-by-frame basis.

Also various other effects.

Well, there's still nothing magic extra-special the voodoo chips alone can do through glide that no other chip can do except buggily through DX9... That's just prepostrous. Glide is nothing but a slimmed-down OpenGL API, and the voodoo series is a very simple and quite conventional rasterizer without any special tricks whatsoever. Even the TNT, which is a POS by today's standard, is considerably more feature-rich.

Looks like I was wrong about shadow buffers, it's hardware frame buffer used for shadows.

"Hardware frame buffer." :lol: Look, by nature, framebuffers are hardware (ie: typically: memory chips), and what games is it you think use "hardware frame buffers" for shadows? Typically, shadows are made either with either a stencil buffer (which only vsa100 supports) or with shadow buffers, which no voodoo chip supports.

Just looked up mym old 3dmark scores, 195.9 megatexels fill rate

TEXEL, yeah. Not pixel...

32 bit internal color too

Internal, wtf? It uses internally what it uses externally, so that would be 16-bit.

(actually, the story I heard didn't mention sega at all, just that 3dfx had planned a lot more for the voodoo3 and cut it for cost reasons, but it's a nice story for a 3dfx zealot, right?)

:lol: Yeah, haha. One's gotta have faith! :lol: Anyway, V3 wasn't really meant to be in the first place, it was a product of bad management.

[Megadrive1988]

[wild speculation mode]

If 3Dfx had Rampage ready at the time it was supposed to be ready, sometime in 1999, I wonder if Sega would've changed their minds. surely Rampage would've been superior to any PowerVR2 solution with the possible exception of NAOMI 2. Rampage was reportedly VERY good. especially with the on-board geometry processing from the SAGE chip. Imagine a 1 SAGE, 2 Rampage rasterizer configuration. probably would've allowed BlackBelt to compete better with PS2 on the technical front. BlackBelt gets released in Japan and U.S. in 2000, instead of 1998/1999. 10-20 mpps with features. would be comparable to Xbox and GCN today. 1 SAGE geometry engine and 2 Rampage rasterizers may even have been comparable, perhaps even better than, NAOMI 2's ELAN geometry engine and 2 PowerVR2DC rasterizers. Although, there is little doubt that a 1 SAGE + 2x Rampage solution would've been too costly for a console. of and of course, even IF Rampage was on time, in 1999, that still would've been probably too late for Sega's timeline since they started making the real BlackBelt and Dural probably in 1996.

[/wild speculation mode off]

[Silanda]

[wild speculation mode]

If 3Dfx had Rampage ready at the time it was supposed to be ready, sometime in 1999, I wonder if Sega would've changed their minds. surely Rampage would've been superior to any PowerVR2 solution with the possible exception of NAOMI 2. Rampage was reportedly VERY good. especially with the on-board geometry processing from the SAGE chip. Imagine a 1 SAGE, 2 Rampage rasterizer configuration. probably would've allowed BlackBelt to compete better with PS2 on the technical front. BlackBelt gets released in Japan and U.S. in 2000, instead of 1998/1999. 10-20 mpps with features. would be comparable to Xbox and GCN today. 1 SAGE geometry engine and 2 Rampage rasterizers may even have been comparable, perhaps even better than, NAOMI 2's ELAN geometry engine and 2 PowerVR2DC rasterizers. Although, there is little doubt that a 1 SAGE + 2x Rampage solution would've been too costly for a console.

[/wild speculation mode off]

I doubt they would have gone for this. It would have eliminated much of Sega's headstart over the PS2. Even with superior hardware they would have not been able to combat sony's hype, consumers still didn't trust Sega from their previous misadventures (Mega CD, 32x, to an extent Saturn). Also, sega were not very strong financially and a 2 rampage 1 sage combo would have been expensive if sega were trying to minimise losses. Wasn't that the suggested arrangement for the high end PC card?

[Fox5]

If v3 was a product of bad management, what was voodoo5?

http://www.anandtech.com/showdoc.html?i=923&p=9
" many users are upset with 3dfx for only supporting 16-bit rendering. 3dfx's response has been that their internal calculations are processed with 24-bit accuracy and then dithered to 16-bit, providing something similar to a hypothetical 22-bit color output, however the market is generally not buying that response at all.
"
Thought it was 32 bit internal, guess it was just 24 bit internal. Matrox did the same thing, and so do all powervr chips, iirc correctly, except they also provided the option for external.

"Just looked up mym old 3dmark scores, 195.9 megatexels fill rate


TEXEL, yeah. Not pixel... "

When you're doing single texturing, isn't it the same as the pixel rate? I mean, isn't that the purpose of them having both a single texturing and multitexturing test, to see how high its pixel rate is, and to see how high its multitexturing test can be when provided with as many layers as the hardware supports? Besides, those neon250 cards were agp cards, I had a pci voodoo, at a bit of a disadvantage eh?(half the potential bus speed the voodoo card could support I think, but the neon250 used agp4x)

BTW, how come for both psx and n64, the glide plugins are generally more compatible, full featured, and faster? Just happened that the plugin authors are more fluent in glide?

Voodoo cards also support 8bit palletized textures, which I remember was a big deal to me and my friends back in the day. Not sure how many games actually used that, but it sucked for them to get their new geforce and have to play their game in software rendering.

Also, from the banshee feature list....
Per-pixel MIP mapping and alpha blending
Most accurate LOD calculations
Internal rendering calculations @ 32bpp
Programmable expotential and fog-table
Floating point Z-buffer to eminate z-aliasing
Dynamic environment mapping
Bumpmapping
Full speed Trilinear Mip mapping quality
Full scene "order independent" edge anti-aliasing
3D rendering up to 2k by 2k surface

So internal 32 bit, bumpmapping, trilinear, and I doubt they reduced the voodoo3's capabilities.

16 Bit Floating Point Z-buffer
Games are designed for 64k Z-values
Content Developers: 24/32-bit Z-values
32-bit Z will run at half the speed of a 16-bit Z
32-bit Z requires double the memory storage
Games require high frame rates, not long depth range
32-bit Z buffer is primarily useful for CAD applications
16-bit floating point Z is more precise than some 24-bit fixed point Z buffers
Banshee's 16-bit W-buffer implimentation has ~22-bits of integer accuracy

So does that mean it can still run 32bit games, but at half speed and twice memory storage? Also, it seems they use the 16bit floating point capabilities to emulate 22-24bit integer accuracy?

"For one, the Voodoo 5, like a number of Voodoo products before it, includes a post filter that modifies images after they're rendered, but before they're output to the display. You may have heard 3dfx advocates talking about "22-bit color" output from the Voodoo 3 a while back. In 16-bit color mode, the Voodoo post filter averages pixel colors together. This filter smoothes things out, but sometimes causes images to look over-filtered, smeary, or dirty; or it introduces line-by-line horizontal banding. In its "high quality" mode, though, the 3dfx post filter arguably improves 16-bit image quality. "

So 3dfx's higher quality 16bit does exist.

"Why not? Probably because the VSA-100 chip won't do trilinear filtering and single-pass multitexturing at the same time. Because single-pass multitexturing improves performance, I suspect 3dfx decided to cheat things a little by not enabling trilinear filtering in their OpenGL drivers. Doing so would hurt benchmark performance versus some competing cards, where adding trilinear is "free" thanks to their rendering pipeline configurations. "

So the choice is between single pass trilinear or single pass multi texturing...doesn't say it can't do multitexturing and trilinear though.
Also found one interesting thing while looking around, the radeon 8500 couldn't do anisotropic and trilinear at the same time. Judging by the rest of the voodoo's comprimises(well, voodoo3-5, voodoo1 and 2, believe it or not, did lack features the voodoo 3 had), it would probably be done in a similar way. Couldn't find any articles on the voodoo having anisotropic though.(though I did find one person complaining that they can't enable anisotropic on their voodoo3 because the performance hit is too great) Oh wait, n/m, look what I found....

http://www.guru3d.com/voodoo3.htm
The Specs

Dual 32-bit texture rendering architecture
Single pass multi-texturing
Full hardware setup of triangle parameters
Support for multi-triangle strips and fans
Single Pass, Single-cycle bump mapping
Single Pass, Single-cycle tri-linnear mip-mapping
Alpha Blending on source and destination pixels
Sub-pixel and sub texel correction to 0.4x0.4 resolution
Per-pixel atmospheric fog with programmable fog zones
Full-scene polygon-based edge anti-aliasing
Floating point Z buffer
True per-pixel, LOD MIP mapping with biasing and clamping
Texture composting for multi-texture special effects
8-tap anisotropic filtering
Support for 14 texture map formats
8-bit palletized textures with bilinear filtering
Texture compression through narrow-channel (NCC) YAB format

It even have my beloved 8-bit palletized texture support! Oh, 8-tap aniso too.

Further down in the page, it says the voodoo3 supports a 4:1 compression ratio....using palletized textures and NCC. I'm guessing there's a reason this couldn't be used in every game like the dc's graphics chip's texture compression?

"Q: What is the highest color depth available for the 3D output? Is it still 16-bit color or has it been updated to 32-bit color? A: The Voodoo3 uses internal rendering calculations @ 32bpp with a proprietary compression algorithm to store results at 16bpp. This gives us the best of both worlds: high color depth and low memory bandwidth! "

"Q. Does Voodoo3 support anisotropic filtering?

A. When we launched at Comdex ‘98, this was primarily a software feature which we decided not to productize after having reviewed the resulting image quality that it provided.

Q. Does Voodoo3 support full scene anti-aliasing?

A. This is another primarily software feature that is under review.
"

Not sure what these mean, but the fsaa one goes along with what I remember from the super sampling program, it used the cpu to perform the supersampling.

"3Dfx Voodoo is one of the most successfull trademarks of the computer hardware industry, beaten only by the Apple Macintosh and IBM PC brand-names. Is the Voodoo 3 a product worthy it's name? We say it is. "

I can only laugh at this.... comparing the voodoo to two other failures, why not windows if they had such a high opinion of it? Oh well, I used to compare 3dfx to sega, and look how that turned out, sega nearly died too.

[Fox5]

[wild speculation mode]

If 3Dfx had Rampage ready at the time it was supposed to be ready, sometime in 1999, I wonder if Sega would've changed their minds. surely Rampage would've been superior to any PowerVR2 solution with the possible exception of NAOMI 2. Rampage was reportedly VERY good. especially with the on-board geometry processing from the SAGE chip. Imagine a 1 SAGE, 2 Rampage rasterizer configuration. probably would've allowed BlackBelt to compete better with PS2 on the technical front. BlackBelt gets released in Japan and U.S. in 2000, instead of 1998/1999. 10-20 mpps with features. would be comparable to Xbox and GCN today. 1 SAGE geometry engine and 2 Rampage rasterizers may even have been comparable, perhaps even better than, NAOMI 2's ELAN geometry engine and 2 PowerVR2DC rasterizers. Although, there is little doubt that a 1 SAGE + 2x Rampage solution would've been too costly for a console. of and of course, even IF Rampage was on time, in 1999, that still would've been probably too late for Sega's timeline since they started making the real BlackBelt and Dural probably in 1996.

[/wild speculation mode off]

You're assuming it had the same power it did when it almost released in 2001. Even then, a single chip rampage(which was the low end configuration, and 2 chips was either medium or high end) would be outperformed by gamecube(mostly since gamecube gets a lot of effects for free), but a dual chip rampage would have even outperformed an xbox with twice the memory bandwidth. Lets say a single rampage would have cost about as much as a voodoo3 sli(if such a thing existed), sega didn't go for a voodoo3 sli, so I'd think that was more than they were willing to pay.(though it would have offered incredible performance, which the sh-4 in the system probably couldn't utilize)

[Silanda]

BTW, how come for both psx and n64, the glide plugins are generally more compatible, full featured, and faster? Just happened that the plugin authors are more fluent in glide?

Perhaps because there is only a rather limited pool of hardware using glide? It cuts out a lot of variables if you are only writing for one manufacturers cards (most of which are ultimately based on the same design), rather than every card under the sun. Besides is this even true? What exactly is missing from d3d and ogl renderers for epsxe for example?

[Fox5]

BTW, how come for both psx and n64, the glide plugins are generally more compatible, full featured, and faster? Just happened that the plugin authors are more fluent in glide?

Perhaps because there is only a rather limited pool of hardware using glide? It cuts out a lot of variables if you are only writing for one manufacturers cards (most of which are ultimately based on the same design), rather than every card under the sun. Besides is this even true? What exactly is missing from d3d and ogl renderers for epsxe for example?

Don't think there's any features missing for psx, but the glide rendered are a lot faster. I don't know why either, there are things I could run full featured with glide on my voodoo3 that were unplayable with a d3d or opengl plugin on my geforce 3(though back then the drivers for the card really sucked, for whatever reason performance like doubled with later drivers), and still aren't as fast(though sometimes it's only when certain effects are used) on my radeon 9700 pro. It's not just that the glide plugins are 10 or 20% faster, they're faster than more powerful hardware using another plugin by quite a noticable difference. Though I did try a glide wrapper once on my radeon 9700 pro, not sure if it was faster than my voodoo3 with native glide(since it was running at a full 60 fps, which was good enough for me), but it was faster than d3d or opengl plugins, and tended to support more features, but not always, often visual effects would have errors or just not show up at all....and sometimes they were even effects that could be done with the other plugins.(perhaps voodoo cards had a faster way of doing the effects that newer cards don't support, and thus the d3d and ogl plugins are slower because they have to do more work to emulate the same effect? well, that could very well be true, since the d3d and ogl plugins I was using required ps2.0 capable hardware for those effects, and I haven't found any pre dx9 plugins that could do those effects)

BTW, one reason I claim a 3dfx powered console would have been more powerful than dreamcast...well it would have come out a year later. I think of dreamcast as coming out at the time it did in america, late 1999 when the geforce was just about to come out(and 3dfx could have produced a more powerful card at that time, many voodoo 3 cards would overclock over 200 mhz, and if 3dfx had just shrunk the die process the voodoo3 was on and focused getting the mhz up instead of thinking they could get voodoo5 out by christmas, they could have had a card about geforce power, but minus t and l), but the dreamcast came out in 1998 in Japan. I think of a 3dfx powered console as launching in in 9/9/99 worldwide, and it's rare that any company in the computer business can maintain a whole year's advantage on another company, especially a leader in the field at the time. At best, the chip in the dreamcast would be approximately equal to a v3.

However, I few things I found online:
The Blackbelt console was focused on ease of programming, and not power.(I guess sort of like gamecube?) It also allowed for quick conversions from PC to the console.(and not in the way windows ce did for dreamcast) Guess like an anti saturn?

Katana was more powerful most of the time than the prototype BlackBelts, but much more difficult and complicated to program for.

The number 1 reason BlackBelt was dropped was because it was more expensive. The number 2 is that it wasn't even close to the PS2 or GameCube in proposed power. No mention of how the final BlackBelt specs compared to Katana, but if the BlackBelt was more expensive or launched a year later than Dreamcast initially did, perhaps sega felt it was too much of a risk when they were already economically struggling, Katana's earlier release/lower price would matter more than any extra power, especially when it was far more powerful than anything else available at the time.

Katana gained an equally easy to use programming environment to the BlackBelt through the use of Windows CE. However, didn't Windows CE on dreamcast suck? Unless they managed to do quake 3 on it, then I don't recall any impressive games from it, even worms world party had framerate problems and they claimed they were unable to do the flame effects on it or something.(maybe it was just for balance reasons, or too hard to have while using a controller) Oh, and another reason why those comparisions between pcs with 500mhz processors running quake 3 and quake 3 on dreamcast aren't so invalid, dreamcast doesn't have to run a bloated OS, unless quake 3 on it was a windows ce port.

[DeanoC]

I've programmed Voodoo 1 and 2 at the glide level and I've programmed Holly (PVR2DC) and I also know Voodoo3 at the Direct3D level. So I probably can comment better than almost anybody here.

More importantly I've ran the same game on Direct3D on both a Voodoo3 and Dreamcast. Dreamcast wins hands down.

Why?
True 32 bit rendering, DC supports a accumulation buffer that enabled multi-pass effects at full bit resolution.
VQ compression, better quality and smaller than palettes
Modifier volumes, shadow were so easy
Clipping, the Dreamcast had 4 plane geometric clipping
High precision z-buffer without wasting valuable VRAM
Sort-indepedent transparency

That feature alone puts Dreamcast graphics chip up there with the all time greats.

Any idea that Holly (via Kamui or Direct3D) was harder to program than glide is rubbish, Kamui and glide were actually very similar, except under Kamui most of the calls were actual C Macros and placed data directly into the command stream.

The very things that made PowerVR a pain on the PC were what made it good on a console.

SEGA made the right choice with graphics chip for the Dreamcast, there was NO better out there at the time.

[Megadrive1988]

I've never heard the name Holly for PowerVR2DC.

ya learn something new everyday!

[ERP]

I agree with Deano on this one.

I didn't do much more than some initial engine work on DC, but considering the time frame it was a very impressive piece of kit, certainly more than a match for the suggested 3DFX part (which wasn't even a Voodoo 3).

There were actually at least 3 different libraries available for dreamcast , Win CE, Kamui which was a well written low level library written by Sega (although they later removed what I felt to be the most useful rendering mode from it) and something called Darkness, which was at the level of here are the register addresses and some really incomplete documentation go for it.

At the Kamui level writing DC code was comparable to writing Glide code for 3dfx on a PC.

[archie4oz]

I've never heard the name Holly for PowerVR2DC.

Really?

Anyways it's the name of the whole package/chip not just the PVR core, TA, and all the various Bus IFs...

Win CE

Dragon SDK, yech...

There were actually at least 3 different libraries available for dreamcast , Win CE, Kamui which was a well written low level library written by Sega (although they later removed what I felt to be the most useful rendering mode from it) and something called Darkness, which was at the level of here are the register addresses and some really incomplete documentation go for it.


Are you referring to Kamui or Kamui 2 (removed rendering mode)? I'd also toss in Ninja if you needed a more high level interface or were coming from a PC backround (and didn't want to go the Dragon route).

[Megadrive1988]

Really?

Anyways it's the name of the whole package/chip not just the PVR core, TA, and all the various Bus IFs...

neat. well maybe I've simply forgotten. but I didn't remember the name Holly when i saw it mentioned here.

[ERP]

I've never heard the name Holly for PowerVR2DC.

Really?

Anyways it's the name of the whole package/chip not just the PVR core, TA, and all the various Bus IFs...

Win CE

Dragon SDK, yech...

There were actually at least 3 different libraries available for dreamcast , Win CE, Kamui which was a well written low level library written by Sega (although they later removed what I felt to be the most useful rendering mode from it) and something called Darkness, which was at the level of here are the register addresses and some really incomplete documentation go for it.


Are you referring to Kamui or Kamui 2 (removed rendering mode)? I'd also toss in Ninja if you needed a more high level interface or were coming from a PC backround (and didn't want to go the Dragon route).

Not sure when it got removed we did DC development very early and very late in the platforms lifetime.

They remooved the model that allowed direct submit of tri's to the TA. In favor of intermediate memory buffers, with DMA submission to the chip.

The problem I had with them removing it was that Dreamcasts cache didn't support read under write, so having to write back to memory meant that you couldn't effectively prefetch vertex data, which vastly increased transform times for simple verts. This would have been avoidable if you could have DMA'd to the scratch pad memory, but that didn't work either.

When I complained about the removal of the rendering mode, Sega provided me with a copy of Darkness which was basically a bunch of register equates, some bad sample code and some docs which noted a couple of hardware bugs that it was necessary to work around. I used it to do some initial graphics work on a product that was canned when sega pulled out of DC development.

I like playing around at the register level, I like to understand how something is working :), that and I'm a masachist when it comes to hardware. One the other hand I also like not having to if I'm just trying to test something.

[bbot]

With all the talk about PVR, how about "what if MS chose a variant of PVR, instead of Geforce3 for the Xbox"?

[Guden Oden]

I like playing around at the register level, I like to understand how something is working :), that and I'm a masachist when it comes to hardware. One the other hand I also like not having to if I'm just trying to test something.

Hehe, that's cool. :) I guess this makes you unlike Carmack, who is a high-level algorithm whore instead. :D He'll gladly rewrite half his engine if it gets better in the process, but he'll only consider doing it in C++. :lol:

[Simon F]

Clipping, the Dreamcast had 4 plane geometric clipping
It couldn't actually do that* but it could do post projection XY screen region clipping and that, along with "near**" plane clipping (which was the SH4's responsibility) is probably all you typically need.

*though I suppose you could use modifier volumes to do some geometry clipping by making the "outside" region invisible - actually, that'd be quite powerful although it would burn fill rate.

**since depth vals were floating point, I guess "near" could be very close.

[Deepak]

This thread is extremely informative and interesting. Why not have similar threads about PS2/GC/XB (now that this gen is near its end :D ). Megadrive what do you think?

[Megadrive1988]

With all the talk about PVR, how about "what if MS chose a variant of PVR, instead of Geforce3 for the Xbox"?

PowerVR Series 4, with its 4 pipes and on-chip T&L or Vertex Shader(s) might have been a nice alternative to the GeForce3/GeForce4 variant, NV2A, in Xbox.

[Fafalada]

Hehe, that's cool. I guess this makes you unlike Carmack, who is a high-level algorithm whore instead. He'll gladly rewrite half his engine if it gets better in the process, but he'll only consider doing it in C++.
I could be wrong but from what I recall Carmack stayed away from C++ a fair bit longer then most of the development community.
Besides, if you have the time to spend on it, you can write C++ into doing things at very low level too, while still keeping far more generality then lower level lang. can afford.

Anyway, I also like to get familiar with how exactly hw does things, but if I was working on PC as primary platform I'd be reluctant to bother with it for every bloody chip out there too. :P

[Fox5]

With all the talk about PVR, how about "what if MS chose a variant of PVR, instead of Geforce3 for the Xbox"?

At best a dual chip kyro 2 then I suppose. Power may have been about the same as a dual chip rampage(or a voodoo5 6000, which would be a bit slower), but would it have been worth it with no pixel shaders at all, and possibly not any t&l?

Hehe, that's cool. I guess this makes you unlike Carmack, who is a high-level algorithm whore instead. He'll gladly rewrite half his engine if it gets better in the process, but he'll only consider doing it in C++.

I never knew this....wow, Carmack writes some of hte fastest and most versatile game engines, and he doesn't even code to the metal like other developers.

[DeanoC]

John Carmack gets as low as he can get on PC.

Doom3 is the first non C engine, he wrote the Jaguar version of Doom himself and he first innovation was finding interesting ways of scrolling back in EGA (register level hacking). He also did a fair bid of pioneering ModeX for Doom PC.

He works the same as all the best people do, try various algorithms until you find one you like and then you optimize it till it bleeds.

[archie4oz]

I like playing around at the register level, I like to understand how something is working , that and I'm a masachist when it comes to hardware.

Ah yes, I used to love sitting down with a bowl of squid jerky and a beer and get nice and intimate with a PC Engine, SFC or Saturn... Shame that deadlines, publishers, and the realities of living can sap all the fun out of that most of the time...

Besides, if you have the time to spend on it, you can write C++ into doing things at very low level too, while still keeping far more generality then lower level lang. can afford.

Meh, it's all about Lisp baby! ;) Actually I'd better shut up before you get on the metaprogramming soap box.. :P

Anyway, I also like to get familiar with how exactly hw does things, but if I was working on PC as primary platform I'd be reluctant to bother with it for every bloody chip out there too

Hehe, go to the Mac... :P 2-3 processor designs to worry about (and only one really cool vector ISA extension to deal with), and simpler configurations to deal with... :P Granted unless you work at like Omnigroup or something you'll have to "serve fries with that" part-time to pay for rent.. :P

Doom3 is the first non C engine, he wrote the Jaguar version of Doom himself and he first innovation was finding interesting ways of scrolling back in EGA (register level hacking). He also did a fair bid of pioneering ModeX for Doom PC.


And scores cool points with me for sticking to GL and being a NeXT fan back in the day... :)

[DeanoC]

I like playing around at the register level, I like to understand how something is working , that and I'm a masachist when it comes to hardware.

Ah yes, I used to love sitting down with a bowl of squid jerky and a beer and get nice and intimate with a PC Engine, SFC or Saturn... Shame that deadlines, publishers, and the realities of living can sap all the fun out of that most of the time...


There always enough time for quickie with the hardware :twisted:


Hehe, go to the Mac... :P 2-3 processor designs to worry about (and only one really cool vector ISA extension to deal with), and simpler configurations to deal with... :P Granted unless you work at like Omnigroup or something you'll have to "serve fries with that" part-time to pay for rent.. :P


Everybody wants a Mac these days, just not running OS X :wink:

[archie4oz]

Everybody wants a Mac these days, just not running OS X

Yeah I know, it's sad... :( Do they come with a can of disinfectant? :twisted:

[Fafalada]

Meh, it's all about Lisp baby! Actually I'd better shut up before you get on the metaprogramming soap box.
Lisp... what's that again? Oh right, I remember... it's a template metaprogram! ;)
You're just jealous I get a compiler out of the box and you'd probably have to write your own. :P

Hehe, go to the Mac...
Well I'm good with consoles for now. But maybe in future if EA decides to support Mac, who knows what could happen...

There always enough time for quickie with the hardware
One can always make time for that yeah. Just that it gets even better when you can take your time and do it slowly...
Ahem...

[archie4oz]

You're just jealous I get a compiler out of the box and you'd probably have to write your own.

Yeah but it'll generate faster, tighter code than the half-assed outa-the-box port of a BOCCE interpreter you call a compiler... :twisted:

One can always make time for that yeah. Just that it gets even better when you can take your time and do it slowly...
Ahem...

Yeah, but time for that sort of attention is usually reserved for 1st party blokes. :P 3rd party chaps get stuck with waiting in line for quickies... 8)

[darkblu]

He also did a fair bid of pioneering ModeX for Doom PC.

you mean he was sane enough to see the commersial viability of a VGA feature the demo coders had been using for ages before him. as for the EGA panning - yeah, i think he actually took key part in inventing that feature, but don't quote me on that. gosh, that stuff dates back to the commander keen days.

but one thing can be attributed to the guy unquestionably - the famous low-granularity-perspective-correction assembly texturing routine in quake1 ...or wait! was that Abrash's?

[archie4oz]

IIRC that was Abrash's...

[DeanoC]

you mean he was sane enough to see the commersial viability of a VGA feature the demo coders had been using for ages before him.

<snip>

but one thing can be attributed to the guy unquestionably - the famous low-granularity-perspective-correction assembly texturing routine in quake1 ...or wait! was that Abrash's?
Both Abrash I think, AFAIK xsharp was the first major user (fast filled polygons). But it was well known even before that, just nobody could think of a good use for them. It was Abrash that realized (or at least publicly wrote an article) you could chain the writes to write 4 pixels at once.

The point I was making was that John Carmack knows his hardware registers and he's certainly not a high level only coder. Anybody who played with Tom and Jerry can claim hard core status :-) (T and J are the Risc processors in Jaguar).

[Guden Oden]

Ah yes, I used to love sitting down with a bowl of squid jerky

Ggyyeesshh...! :? I dunno, it might taste good, but then I think about how squid look when they're ALIVE, and how they MOVE....! :lol:

I ate squid once, on Crete. But then it was nice and grilled, on a bed of rice and veggies n stuff! :D

Everybody wants a Mac these days, just not running OS X :wink:

Hey, I wouldn't mind a modern 2.5GHz dual-processor Mac running OSX, but it bloody well be a version of OSX that supports more than one friggin mouse button...! Jesus, that's stone-age, no wonder they used to call it the macintoy back in the day.

[archie4oz]

Ggyyeesshh...! I dunno, it might taste good, but then I think about how squid look when they're ALIVE, and how they MOVE....!

Wusssss... :P Not as bad as you think (doesn't even look like squid most of the time)

Hey, I wouldn't mind a modern 2.5GHz dual-processor Mac running OSX, but it bloody well be a version of OSX that supports more than one friggin mouse button...! Jesus, that's stone-age, no wonder they used to call it the macintoy back in the day.

Ummm... OS X has always supported more than one mouse button... Scrolling even works, imagine that! :roll: Actually the stone-age comments about the one-button are for the clueless since you've always had keyboard modifiers in prolific use on Classic... If you've ever seen and old skool hardcore mac user fly on the system with one hand on the mouse, the other on the keyboard doing stuff that's just simply not possible on a 3-5 button mouse, it's pretty amazing... (actually Maya and Houdini incorporate some of those UI/input paradigms as well)...

BTW, my dual 2.5 is on the way.. :P

Anyways, just to keep this mildly on topic, some more up Simon's alley, where do you consider counting from for determining the PVR's tri throughput? Tris coming out of the TA, or the output of the ISP?

[Fafalada]

Ggyyeesshh...! I dunno, it might taste good, but then I think about how squid look when they're ALIVE, and how they MOVE....!
Aww, how could you not love this
http://homepage.mac.com/archie4oz/.Pictures/Fafalada/squid.jpg
I just had to go grab one after all your talk :P Seriously they are a very delicious snack.

And speaking of the live things, squids got nothing on Sea Cucumbers as far as yuck factor goes. And they are edible too - granted when I first tried them I didn't even know what it was...

as for the EGA panning - yeah, i think he actually took key part in inventing that feature
I dunno, I think most people that played with register stuff for the first time would think of the idea, it's pretty straightforward(I know I did - back then it was not so uncommon to "not" know of already implemented stuff and reinvent the wheel many times over). The other matter is how many ever got a commercial products out of it, and out of those Keen was probably the first anyhow.

[DeanoC]

And speaking of the live things, squids got nothing on Sea Cucumbers as far as yuck factor goes. And they are edible too - granted when I first tried them I didn't even know what it was...


As you brought up Sea Cucumbers, can I ask a dumb question to anybody who might know?

While porting Silent Hill 2, the main creatures are called Sea Cucumber (or at least the auto translated comments say they are), the also had a enum name of SCU, so we think its there actual name but why? The actual creatures are the slimy guys that look like they are in straight jackets, why would they be called Sea Cucumbers?

Its something I've never managed to work out and it dominants my thoughts whenever sea cucumbers slip into a conversation (which to be honest, isn't very often).

The best I've come up, is that its some Japanese cultural thing that is completely missed by us silly Englanders...

(I think this may be a record for off-topic-ness but then maybe Sea Cucumber have some Dreamcast meaning as well :-) )

[Farid]

And speaking of the live things, squids got nothing on Sea Cucumbers as far as yuck factor goes. And they are edible too - granted when I first tried them I didn't even know what it was...


http://www.uog.edu/classes/bi-100/images/sea_cuke.jpg

Here comes the diner!

Bon appétit! :D

[Guden Oden]

Vysez,

Too bad for you French people that sea snails are typically extremely poisonous! :D

(Well, so I heard anyway hehe. :))

And YES Faf, I would actually try squid jerky, I'm not THAT squeamish. I just want to make sure my food won't suddenly slither off the plate and crawl back to the ocean again! :lol:

[Simon F]

Vysez,

Too bad for you French people that sea snails are typically extremely poisonous! :D
So, presumably, they are therefore available in expensive Japanese restaurants?

/me remembers (with a shudder) eating liver of anglerfish whilst in Tokyo. Bleagh. At least, I think it was anglerfish liver - it was green, had the consistency of cream cheese, and tasted utterly foul.

[Farid]

I just want to make sure my food won't suddenly slither off the plate and crawl back to the ocean again! :lol:

And whatcha gonna do when that squid will slither off the plate and start attacking the submarine you're onboard?

http://www.donatoart.com/20000.jpg

Squids are mean! Don't trust them! :twisted:

edit:Too bad for you French people that sea snails are typically extremely poisonous! :D

You should try some of our finest cheese, Guden... Sea Snails poison (one of the most dangerous poison on earth BTW) is nothing compared to that. :lol: