3dfx Rampage ;)

No, not at all.

 

Oh well, I look like a moron once again. Don't know where I got that idea.

 

I'm not that familiar with powerpoint, but does that 2/19/2004 at the bottom left hand corner of the slide mean it was modified today?

BTW, while I suppose sli makes up for it, I was hoping a single chip would reach the 1 gigapixel barrier, but I guess between geforce 2 gts and ultra performance would be good for the low end.

And pretty nvidia like performance numbers for the rampage there, comparing the banshee to the voodoo3 to the rampage at super high resolution(why is the banshee in there? just lumping 2d/3d chips in there? I wanna see the voodoo 5 in there!) with FSAA on, something which I think the banshee and voodoo 3 would suffer severe performance hits on, and would they have to use the cpu for it? I think rampage single chip at best would be about 3x as fast as v3 at none bandwidth limited reses with fsaa off.(333 MP versus 800 MP)
Nice that it supported HDTV though.

 

The presentation is available for download at a fan site. It's quite an old presentation, first shown before the Napalm delay. The transistor count of the final silicon was also much higher than expected at presentation time (more around 30M than 18M).

An interresting detail of this presentation is that with Rampage it was the first time 3dfx used emulation tools for chip development ...

 

All of the 3dfx powerpoints have the "current date" in the corner, rather than any specific date.

The thing is much more efficient than the classic example of sheer brute force that you list there.

The low end single Rampage was intended to go toe to toe with GeForce3, and likely would have (may not have necessarily defeated it, but it would've definitely compared extremely well).

Yes, odd that the V5 is in there. I suspect they had a hacked working multisample buffer working on V3 and Banshee, but not on VSA-100.

Anyway, that alone shows quite respectable performance.

 

Except according to that slide show, it would have only done so in 16-bit color. Additionally, remember that one of the reasons that the GeForce3/4's performance was much higher than that of the GeForce2 was due to better memory efficiency. Did 3dfx plan to use similar advances? If not, then the low-end product could not have competed with the GeForce3.

 

If that would have been the case, then according to simple logic there wouldn't had been much reason to waste resources on a dual chip rampage design after all, would it?

It always depends under what conditions those hypothetical scenarios are being thought of. The GF3 wasn't designed to show it's muscle in dx6 applications after all either.

 

that's a rather old slideshow. The Rampage could do 13-bit signed per component, and it had some very good bandwidth reduction features to allow for the higher bit-depths. I'll let tag go into it in more detail (but honestly, if you would just read that link she gave earlier it would answer most questions you have...)

edit:
does someone need to leak the Imm2k doc again? I did it last time, it's someone elses turn...

 

And yet the slide show did still specify 16-bit for the higher fillrates. It may well have just been a memory bandwidth issue.

 

pay more attention before wasting more of my time.

 

Damn, getting pissy, aren't we?

Besides, that doesn't mean anything to me. Old compared to what? The hardware was never released...

 

srry, read my sig


edit:
old meaning outdated. it was for a version of a chip that was supposed to be out instead of the V5 (if I remember correctly) and not the one that ended up at the fab.

 

As Sage said, that slide show is outdated.

Not only that, but Rampage did indeed support FX13 - effectively 52-bit colour. Why do you think 3dfx would include that if it was completely worthless, speed-wise? And if Rampage supported FX13 at a decent rate, don't you think it might've been able to handle 32-bit just fine? Hehehe.

FX13 is kinda like the 'ace up the sleeve' of Rampage - it would've given it hands down better IQ than nearly any other DX8 card.

 

In what way is FX13 internal precision related to performance with a 32bit framebuffer as opposed to a 16bit one?

Oh, and R200 supports FX16, NV2x has FP32 in the texture shader and in FX9 in the color part. Together with the rather restricted pixel shader capabilities, I don't think FX13 would have been anything like "the 'ace up the sleeve' of Rampage".

I feel like repeating myself, but I don't believe for a second that a single chip Rampage would have performed close to a GF3.

 

With half the multi-textured fill rate of a GF3, no way. SLI was Rampage's ace up its sleeve (near 13GB of bandwidth in early 2001? impressive), and a rather costly one (for the consumer) at that.

 

I think it was said they did, and I think the rampage was only using 16bit color because the voodoo3 and banshee could only use 16 bit.(if you're talking about the performance graph)

 

Well, if I remember correctly, PS 1.0 was meant for the Rampage, so the Rampage was to have a similar shader architecture to the NV2x: texture instructions separated from color ops. This meant that the Rampage may well have had much higher precision for texture ops (as it should have: FX13 would be pretty inaccurate for textures, with only about 256x256 addressable reasonably-well).

 

heh, i'm flattered.

that slide show was from 8/31/99. incredibly outdated.

also, that article i wrote for vnroundup had a lot of incorrect data on it (but for the time, it was the closest public information). it wasn't until that was published that i started getting emails from ex-employees and the like. my 3dfx site has all the updated rampage stuff on it.

i only know of three working cards with rampage on it that exist today. one is even rumored to be a dual rampage with sage, but i'll beleive it when i see it.

ps. i hate when these threads come back to life, but i can't stay away when they do.

 

Agreed. Especially when practically every possible facet of the damn thing has already been discussed twice! X_X;;;

 

Ah, but all that multitexture fill rate isn't worth much if you run out of bandwidth.

And besides... go look up recursive texturing.