3dfx Rampage ;)
- Thread starter Ante P
- Start date
3dhq Revenge "specs" - http://web.tiscali.it/mayhem/01 REVENGE.htm
My choice selections -
"TNL units hyper threaded for optimal CPU like performance." What?
"32 fully featured pixels per clock." Riiight.
"500 + MHZ internal and DDR clock." Reasonable but...
"3200MP/s fill rate" ...so 32 x 500 = 3200. Remember that.
"Twin Risc TNL engines for Advanced pixel/Vertex shading calculations" Huh?
"Upto 78GB/s fill rate." - my personal favorite.
Apparently it was also a 16x2 with 36 million transistors. Yay.
EDIT: These were "released" towards the end of that charade (Apr 03), after everyone sane had already lost faith. You have to hand it to "Aqueel," for not much work at all he made himself quite a fortune (in "donations") from what I've heard.
My choice selections -
"TNL units hyper threaded for optimal CPU like performance." What?
"32 fully featured pixels per clock." Riiight.
"500 + MHZ internal and DDR clock." Reasonable but...
"3200MP/s fill rate" ...so 32 x 500 = 3200. Remember that.
"Twin Risc TNL engines for Advanced pixel/Vertex shading calculations" Huh?
"Upto 78GB/s fill rate." - my personal favorite.
Apparently it was also a 16x2 with 36 million transistors. Yay.
EDIT: These were "released" towards the end of that charade (Apr 03), after everyone sane had already lost faith. You have to hand it to "Aqueel," for not much work at all he made himself quite a fortune (in "donations") from what I've heard.
Wouldn't what he did be illegal?
BTW, even the first post on that link has me calling BS, and I'm a hopeless optimist.
If 3dfx really had something that would have changed the industry as much as the voodoo2 or even more so, they would have stayed in business, it's not like they couldn't get loans or something, but it looks like they still would have been in deep trouble even with a 100% successful rampage launch. Well, unless the management was really greedy and nvidia offered them a much better deal than they'd get working for 3dfx.
BTW, even the first post on that link has me calling BS, and I'm a hopeless optimist.
If 3dfx really had something that would have changed the industry as much as the voodoo2 or even more so, they would have stayed in business, it's not like they couldn't get loans or something, but it looks like they still would have been in deep trouble even with a 100% successful rampage launch. Well, unless the management was really greedy and nvidia offered them a much better deal than they'd get working for 3dfx.
His Quake 3 claim is one for the ages. 254 FPS at 1280x1024 with 8x SSAA applied. In other words this card is rendering 2.7 Gpixels/sec, assuming no overdraw. Quake 3 had IIRC an average overdraw factor of around 3, so our requirement jumps to ~8.1Gpixels/sec. Also remember Q3 forced multipass with more than two texture access, so the raw fill rate requirement actually could be much higher. For comparision a 9800 Pro only can write around 3Gpixels/sec.
Fox5 said:
Loans were an issue because of the .com bust and any bank's reasonable concerns that the company would inevitably file bankruptcy. According to one story I was told, 3dfx did have a significant investor lined up late '00 but someone fairly high in management slipped in front of him and spilled that they'd also been talking to potential buyers. The would-be investor immediately bolted at that point.
Interestingly enough, VIA made an offer and it was widely expected throughout the company among those in the know that the board would go with that offer and keep their doors open. The announcement that Friday to go with nVidia's offer came as a complete surprise because it obviously meant the end of the company.
akira888 said:
That forum still exists?
I had the chance to talk on other boards with Aquoes, Devin and the likes and it was bleedingly obvious after the second post if not the first, that they hardly knew what they were actually talking about. I don't know and frankly I don't care either what was behind all that smoke and mirrors, all I could see in Devin's case (or some other individuals present in that thread) is a bunch of foulmouthed minors that couldn't get over their own sandbox pipedreams.
akira888 said:3dhq Revenge "specs" - http://web.tiscali.it/mayhem/01 REVENGE.htm
My choice selections -
"TNL units hyper threaded for optimal CPU like performance." What?
"32 fully featured pixels per clock." Riiight.
"500 + MHZ internal and DDR clock." Reasonable but...
"3200MP/s fill rate" ...so 32 x 500 = 3200. Remember that.
"Twin Risc TNL engines for Advanced pixel/Vertex shading calculations" Huh?
"Upto 78GB/s fill rate." - my personal favorite.
Apparently it was also a 16x2 with 36 million transistors. Yay.
EDIT: These were "released" towards the end of that charade (Apr 03), after everyone sane had already lost faith. You have to hand it to "Aqueel," for not much work at all he made himself quite a fortune (in "donations") from what I've heard.
Heh, we had a 12-page thread about *that*. Makes quite and interesting read in retrospect.
http://www.beyond3d.com/forum/viewtopic.php?t=5189
Ailuros, Geeforcer: Thanks for the info/thread.
Tagrineth: The Rampage had MSAA, correct (albeit done somewhat weakly IIRC). However the 254 fps figure seems to refer to the modified board "Revenge" although it's rather unclear like everything else there. If you look at the specs you see - "Super Sampling upto [sic] 8X jittered grid," which would require 8x fillrate, not just 8x Z-ops. Even if it was really MSAA Rampage only had one Z op unit per pipeline, and while it could loop concurrently with the fragment/texture combiner Q3A still uses only two textures per pass anyway IIRC, so your speed drop would be atrocious at 8x AA anyway.
Tagrineth: The Rampage had MSAA, correct (albeit done somewhat weakly IIRC). However the 254 fps figure seems to refer to the modified board "Revenge" although it's rather unclear like everything else there. If you look at the specs you see - "Super Sampling upto [sic] 8X jittered grid," which would require 8x fillrate, not just 8x Z-ops. Even if it was really MSAA Rampage only had one Z op unit per pipeline, and while it could loop concurrently with the fragment/texture combiner Q3A still uses only two textures per pass anyway IIRC, so your speed drop would be atrocious at 8x AA anyway.
akira888,
One point I recall from the old B3D thread above, is that recent GPUs utilize a combination of sophisticated compression/bandwidth saving schemes and are unable to reach those exaggerated numbers even with 4xMSAA enabled.
Granted 3dfx did have a couple of interesting ideas which have been used by NVIDIA since NV25, yet they're based on those ideas and have seen some understandable evolutions too. Two significant ones would be the way ROPs are being handled, where oversimplyfied pixel fill-rate gets sacrificed in order to leave texel fill-rate intact and thus improve MSAA performance further and the "filter on scanout" trick which has been used up to 2xMSAA in NV25 and 2/4xMSAA in NV3x.
In fact the latter had been present since the VSA-100 already. I don't recall if it had any "conditionals" yet there are on NV2x/NV3x. Filter on scanout saves bandwidth only AFAIK if on NV25 performance won't drop below 2/3rd and on NV3x 3/5th of the resolutions refresh rate you're running in. Conditionals apply for the ROP thing too and that's exactly the point where you're right on spot with the multitexturing scheme of the Q3a engine.
One point I recall from the old B3D thread above, is that recent GPUs utilize a combination of sophisticated compression/bandwidth saving schemes and are unable to reach those exaggerated numbers even with 4xMSAA enabled.
Granted 3dfx did have a couple of interesting ideas which have been used by NVIDIA since NV25, yet they're based on those ideas and have seen some understandable evolutions too. Two significant ones would be the way ROPs are being handled, where oversimplyfied pixel fill-rate gets sacrificed in order to leave texel fill-rate intact and thus improve MSAA performance further and the "filter on scanout" trick which has been used up to 2xMSAA in NV25 and 2/4xMSAA in NV3x.
In fact the latter had been present since the VSA-100 already. I don't recall if it had any "conditionals" yet there are on NV2x/NV3x. Filter on scanout saves bandwidth only AFAIK if on NV25 performance won't drop below 2/3rd and on NV3x 3/5th of the resolutions refresh rate you're running in. Conditionals apply for the ROP thing too and that's exactly the point where you're right on spot with the multitexturing scheme of the Q3a engine.
Here , you'll find a little "Rampage review" with a lots of screenshoots...
http://www.atris.it/Forum/showthrea...263b8f1044fa&threadid=33674&highlight=rampage
This "review" was made by an italian proud Rampage owner...
BYE from MaX
http://www.atris.it/Forum/showthrea...263b8f1044fa&threadid=33674&highlight=rampage
This "review" was made by an italian proud Rampage owner...
BYE from MaX
3Dfx Rampage + SAGE die-shots
3Dfx Rampage + SAGE die-shots
the SAGE core is BIG
http://www.falconfly.de/cgi-bin/yabb2/YaBB.pl?num=1235950780
3Dfx Rampage + SAGE die-shots
the SAGE core is BIG
http://www.falconfly.de/cgi-bin/yabb2/YaBB.pl?num=1235950780
