pouët.net

if by 'MIPS' you mean the result of the industry-standard Dhrystone test, indexed to the 1-MIPS reference VAX, then I have to object against these figures. Neither the size, nor the scalability of these figures is realistic. Where did you get these figures from, and what should we deduce from them?

Another thing that you seem to forget completely is that 486 demos ran in 320x200 (or less), in 8 bit modes, with 1000-ish poly objects (generally only one object on screen aswell), and single-texturing (texturesize generally 32x32 or 64x64).
Shading (if any) was performed with 5ish-bit gouraud or envmap, which acted as an index into a precalced pre-lit palette, so no actual computations were required for blending/shading. There was also no z-buffering, so intersecting objects were generally not possible, and some poly-configurations were also rendered wrongly.

Compare that to more recent demos, where you use 32 bit colour, 1024x768, 256x256 32 bit textures, multitexturing with true blending, z-buffer, multiple lightsources etc etc etc, and all that on enormous scenes with huge polycount... and go figure out for yourself where a lot of today's processing power is used. It's not as simple as you want to make it sound.

And Optimus: PCs rock, when used properly :)

Oh, and I forgot... 486-engines generally used linear or bilinear texturemapping, no perspective (who'll notice the difference anyway when envmapping :))

Besides of being poor - I can't code. :)

Scali, you're no one I would recommend to. But I'll answer to "some" questions.

MIPS: This is a very reliable way to calculate the performance of a processor. And MIPS are still used by the overall industry leaders to measure the performance of the competitionning computers. My source BTW is from a software called Sandra by SiSoftware. Ever heard of that? Hmmm? Download the software and look for yourself. You might notice that a 486 is even weaker "even" than what I have mentionned in FPU....

None the less, I have to inform you that complex scenes existed in 1997 (for example) and I will ASK YOU to go download JIZZ or any other demo from TBL so you can see that you're an idiot. And this is just an example.....

http://www.tbl.org/pcdemo1.htm

Scali: ...not to mention that no matter how fast your cpu runs it still has to stop, light up a cigarette and read the newspaper while waiting for the data to come from the ridiculously slow memory subsystem...

33: Sandra is a synthetic benchmark that shows you how fast a piece of code that you never actually run on your computer executes. Unless of course you use Sandra every day for purposes other than benchmarking...

An (IMO) interesting article on how CPUs have really evolved since the Pentium 100 can be found here.

1995: DOPE by COMPLEX (893kb) Optimized scenes with Gouraud and Phong shaded objects with blur mapping, environment mapping and whatever you want. A simple 486dx is enough for the dream!

Yes, as it appears, SiSoftware seems to have 'forgotten' to divide by the reference computer, but just divide by 1000 or so... (bump the decimal point?)
You can download the Dhrystone source and compile it yourself... This gives you WAY lower ratings than what Sandra reports.
Use a more reliable source please (which was my original point).
Furthermore MIPS only indicates ALU performance, where FLOPS (calculated with Whetstone) indicate FPU performance.
While these have been used by the industry to indicate performance in the past, in recent years we've moved to the SpecINT/FP benchmarks, which give a more diverse workout to the processor, and therefore give a more complete image of performance.

As for 487s being slow... yes every coder knows that, so people didn't use FPU in innerloops... the engines were limited so that fixedpoint maths could be used.
When Pentium arrived, 3d engines changed, and matrix/vector math and poly setup were now done on FPU, allowing for larger scenes, more stable math etc (I already covered such points in my first post).
The Pentium FPU could also allow for scanline subdivision with perspective correction, so we could finally have 'perspective correct' texturemapping.

And yes, I've never seen Jizz, Stash or Astral Blur...
Let's see what we have there? Oh dear... a textured sphere... and on the inside we have a 32^3 grid with a handful of metaballs or other simple procedural volumes? Gee, incredible!
As for the renderer used there... If I'm not mistaken it's indeed z-sorted, not zbuffered (spot the sort-errors on the toadstools for example). And it does look like the texturemapping is linear aswell, no perspective... Note though that I described a 486-engine, and the TBL-stuff is aimed at... P166 or so (and some of the MC-stuff is so slow, that you need way more than that to get anything remotely resembling a decent framerate... Say 500 MHz).
What exactly did you want to say with this example, and how does this relate to me being an idiot?

Moral of my first post, and also this one: Engines have changed a LOT over the years, and they DO take advantage of the new hardware in general... That's not to say that it always shows this very clearly. While 486 demos may have looked great, their underlying technology and philosophy was significantly different from demos on other hardware.

When coders in my group will learn how to push hardware to the limit, new features on new gfx cards will be available... You should have cards in advance (like many game programmers have) to keep the pace.

Or just fuck that and do demos which look good on average hardware (which is challenging anyway).

33: have you ever heard about 'faked' effects?

and people writing 12398239583 polies next to an object that had only 900?

besides the fact that a lot of visuals were achieved by pure trickery.

have you ever coded anything? you sound like a fool.

"None the less, I have to inform you that complex scenes existed in 1997 (for example) and I will ASK YOU to go download JIZZ or any other demo from TBL so you can see that you're an idiot. And this is just an example....."

jizz is *not* that complex. not at all. it was low-res, it did a few nice-looking and for that time 'hip' effects combined with a few smart tricks to generate/compress data (so yes, it was kinda woo-hoo back then), but it certainly wasn't a wonder. (and for the record: i know the creators quite well and have seen both jizz and stash source in the past, so yes: i do base this on facts).

if someone would show up in a 64k compo with jizz today, we'd be laughing our arses off.

usually i'm not with scali when it comes to discussions, but i'd like to second him for a 100% this time.

Uhhh... that phong was fake.
Envmap, as I said in my earlier post. There was no gouraud in there afaik, the envmap did all lighting, via the precalced palette (or even via a static bluescale palette), as I described earlier.
The 'blur' was simply done by adding the previous frame to the current one (where previous frame was faded out a tad). Easy to do with monochromatic palettes. Try it with truecolour on a 486...

Don't get me wrong, Dope was a great demo for its time... But technically it's not all that advanced, compared to later engines... And it couldn't be, because it had to run on 486 :)

Steve-Dave, you're most likely an idiot.

Check this out: Jizz is a 64K demo written in mid 1997 for first generation PENTIUM machines. I know because in jan. 1998 I bought a Pentium 166 MMX (2000$ CDN). Was running pretty smoothly, in DOS and DIDN'T use 3D ACCELERATION. It has a real time SOFT SYNTH (did you notice????) . Where the heck does the music come from, eh? So FR-08 is no surprize......... Me and some buddies in Montreal have designed principles for developing software synthesizer in 1995 that would run on a 486 DX machine, something that would be the NEXT GENERATION of tracking. Reading your comments, doesn't seem you give a shit about this, right?

SCALI: You're a little fuckwit French dude that want's to show how big of a dick he's got. Please.... You haven't invented anything and will never. And before commenting on DOPE, try first running it.

=============================================

The best thing the scene has done since 1998 was these simple steps:

- Move from DOS mode to W32 mode
- Move from Software rendered 3D scenes to Hardware rendered 3D scenes
- OpenGL and Direct X usage

This has been done in 1998 as a huge step foreward. But since then, not much has happened, besides REAL TIME RAYTRACE.

stop talking shit 33.. the softsynth is a sample generator, not realtime. Since your only answer to a rational, well documented post is to insult the poster, you're opinions are instantly as pure gut-generated garbage.

instantly reduced to pure gut-generated garbage, I mean.

There is also a Windows version, which also runs 'pretty smoothly', and I believe there are BeOS and linux versions aswell, who cares what OS it uses?



You can't use what isn't there. In 1997, 3d accelerators were still pretty much a novelty... I didn't have one at that time anyway. I doubt any of the TBLers had :)



Actually, it precalcs the samples at startup (that's partly why it takes so damn long :). During the actual demo, it's basically an XM playing with the generated samples. No different from regular XMs, as far as CPU usage goes.



Doh! You mean I've been living in the wrong country all this time?!



I never claimed that I did. What's the relevance of this to the current discussion anyway?



I can dream Dope, it's one of my favourite 486-demos, and one of my all-time favourite demos aswell. And as you see, I can reasonably accurately describe how it works on the inside. But what are you trying to say?



It appears that a lot of the evolutionary steps just aren't noticed by people like you. That doesn't mean they aren't there. A lot of them are just very subtle.

Anyway, I don't really know what you're trying to say here, perhaps you should try again, without the flames.

where is the NEXT GENERATION to tracking now then? Or did it fail because you tried to render the samples in real-time instead of pre-render it? ;)

Oh, since you like Jizz so much:
Here's an evolutionary step for you:
TBL used a MarchingCubes table generated from the 15 original base cases. These have bias-problems, which can lead to holes in the surface.
Later demos do not have holes in them, because they also use the 8 additional cases that deal with ambiguous bias in a cell.

and in stash, they had some obvious problems when mirroring tables :)

anyway

"Check this out: Jizz is a 64K demo written in mid 1997 for first generation PENTIUM machines. I know because in jan. 1998 I bought a Pentium 166 MMX (2000$ CDN). Was running pretty smoothly, in DOS and DIDN'T use 3D ACCELERATION. It has a real time SOFT SYNTH (did you notice????)"

i'm not even gonna answer this. i just happen to feel so much better than you ;) go take some english classes btw. nitwit.

TBL, Contour, Psychic Link, Acme, Pulse, Oxygene, Complex, Matrix, Halcyon, TPOLM ..........

They will always be remembered as the only innovators. R.I.P. 1995-1998

You can dig your hole Scali.

I believe an 'IMHO' is in order here...
This seems to be a severely biased and severely personal opinion.
There have been plenty of 'groundbreaking' demos before and after that period, and by plenty of people other than the ones you mentioned... And I say 'groundbreaking' because Jizz may have been one of the first to have realtime MarchingCubes, but the algo was known long before (also the bias-problem, ironically).

Would you not call eg. Animotion by Phenomena groundbreaking? It was the first ever demo to feature the spacecut effect... Had quite a wow-factor at the time. Or what about Enigma, also by Phenomena? A show of 3d excellence as never seen before.
There was a scene before 1995 you know. And there's a scene after 1998. Just because you don't like those productions doesn't mean they're crap.
I don't like all demos either. And lately there haven't been too many demos that I really liked... But every now and then, a demo still comes around and brings me joy... Variform by Kewlers is a nice recent example.



Exactly what is that supposed to mean? And what does it have to do with any of the points that we (I?) were discussing? I can't even deduce whether you agree with what I said or not, from this sentence. It's just... pointless... But perhaps that is the key :)

When this thread was started, arguably the Radeon 9700 Pro was the GFX leader. But a few days later and the 9800 Pro is nearing launch. I guess the entire scene needs to upgrade once again.

Seriously, take the inane ranting of 33 out of this thread and you have some of the most interesting posts I have read on pouet.net in a long time.

It’s a shame it takes a moron to stimulate some intelligent comment 8)

i dont care a shit about lamers who cant make their own money.


iam getting a parhelia now, no more ps2.0 demo shittalking from me.

ati r sux

Hmmm... OK, I'll move to the "next" lever or my ranting now ;°) ..........

Here is where you can get Radeon 8500 demos. Those demos are 3 years old. They are demos that feature PS 1.4, SMARTSHADER, TRUEFORM, etc etc

http://www.ati.com/developer/demos/r8000.html


Now here are the NEW demos for the already out Radeon R300 series and soon to come R350, both able to do DX9 and PS 2.0 etc etc. The R300 is out since OCT. 2002 ;°) Also in there are new NVIDIA demos for you NVIDIA fans don't feel left out.

http://www.presence-pc.com/news/newsv3.php?news=257
http://www.ati.com/products/pc/catalyst/dx9demos.html
http://www.nvidia.com/view.asp?PAGE=power_demos
http://www.nvidia.com/view.asp?PAGE=demo_archive

Yeah, those Nvidia demos sure are great. Nothing like downloading a 244 MB file just to see one effect. An ogre of a demo for sure!

33 then go away, we are not interested in your simplistic idolatry of feature lists.