[Steelwire]

Over at firingsquad is a write up from a former 3dfx employee (it doesn't specify whether they ended up at NV) on the last days at 3dfx before it folded. The source wishes to remain anonymous which of course means there can be no confirmation of details in this article. Anyway you can find it here.

[Laa-Yosh]

I expected something more that this. About 90% of the article is useless text, and even the remaining 10% did not contain any information that I haven't known before (okay, except that 3dfx spent $40K per month on lunches, whoa). No details on Rampage or Fear other than what every website has listed years ago, no explanations, no nothing. The screenshots looked familiar, too...

[Joe DeFuria]

It also didn't help that they sorta mixed together Rampage and Napalm. It seems that they used the Rampage code name not only for Rampage itself, but for VSA-100.

Napalm was the code name for VSA-100 architecture. (Voodoo4 / 5).

[Dave Baumann]

Here you go, a little trivia/history for ya - Napalm as it was initially sold as to developers:

http://www.beyond3d.com/news/images/napalm.ppt

[Anonymous]

From the text only two things are interesting :

a; the unknown "soldier" worked for the testing department

b; the engineers at 3dfx seem to be not so topnotch. Seven revisions for one chip is really a lot for a chip with the complexity of an 64bit DDR - VSA100 (Daytona). What was the problem? engineering, tools...????

[Anonymous]

Originally Posted by DaveBaumann

Here you go, a little trivia/history for ya - Napalm as it was initially sold as to developers:

http://www.beyond3d.com/news/images/napalm.ppt

thanks, dave
that was an amusing read - developers-targeted PR.. i guess it was even more funny for those guys who actually watched it at immersion '99.

and boy, was the avenger made look advanced in that .ppt!
..the integration of 7 discrete chips into 1... -- heh, pumping clockspeed (last but not least due to the significant advancement in fabbing tech compared to '96) is not exactly the same as actually putting 2 chucks and 4 bruces into one silicon.
..32-bit texture pipeline.. - the original sst1's texture pipeline was already 32-bit.. and avenger still had no 32-bit texture formats

[darkblu]

doh, didn't notice i wasn't logged in.

[Steelwire]

Dear me! Who in their right mind puts PowerPoint presentations on the web!!! It even has an export to HTML format IIRC.... ne'ermind.

Did seem 3dfx where throwing cash away at a frantic rate. $50K on lunches is just flat out silly. I wonder if they had too many projects running by the time they shut down. That last page shows an awful lot of projects. This is all well and good as long as you have enough engineers and a very good communication system.

I can tell you of one others graphics chip that had that many revisions. By the time we got to revision F of the silicon the company involved just ditched out of graphics completely. There was a G but it never appeared to us. In the end we just patched round the probz we could in the drivers and let sleeping dogs fall into a coma.... Note however that although 3dfx had to go back and change designs it does say that they did not actually produce silicon between revisions which probably saved a fortune but doesn't actually get you anwhere....

[Tagrineth]

Originally Posted by Anonymous

thanks, dave
that was an amusing read - developers-targeted PR.. i guess it was even more funny for those guys who actually watched it at immersion '99.

and boy, was the avenger made look advanced in that .ppt!
..the integration of 7 discrete chips into 1... -- heh, pumping clockspeed (last but not least due to the significant advancement in fabbing tech compared to '96) is not exactly the same as actually putting 2 chucks and 4 bruces into one silicon.
..32-bit texture pipeline.. - the original sst1's texture pipeline was already 32-bit.. and avenger still had no 32-bit texture formats

Try setting 32-bit textures in a 16-bit frame in Q3A or Serious Sam on V3. It should work

[darkblu]

Originally Posted by Tagrineth

Try setting 32-bit textures in a 16-bit frame in Q3A or Serious Sam on V3. It should work

yep, just the same as when dragging the tex res slider to the max would force the v3 use >256*256 tex

[Colourless]

The reason why setting 32 bit textures in OpenGL applications works on a Voodoo 3 is because the ICD must be able to support the all of the standard OpenGL texture formats. However, that doesn't mean that the format requested to be used for the texture is the actual format that is used. You can even tell the ICD to use the GL_RGBA16 format for textures on a Voodoo 3 (that a 64 bit texture with 16 bits per component) and it will still work, but you will not be getting that quality.

[Anonymous]

I'm pretty sure V3 supports *display* of 32-bit full-quality textures albeit only with a 16-bit frame buffer. Someone with a Voodoo3 care to test this? (screenshots of 16-bit textures vs. 32-bit textures)

[Colourless]

Tagrineth i've owned a Voodoo 3, I've programmed for the Voodoo 3 (in Glide, OpenGL and Direct3D) and I have the Glide sources for the Voodoo 3, I guarentee you that it does not support 32 bit textures.

[Added]

I will add this however, if you choose to use 32 bit textures things will probably look slightly better, I haven't tested this, but it's reasonable. The reason is in OpenGL you can not specify to use the 565 texture format (you can only use 555), however, if you ask for 888 the V3 will use 565 since that is it's closest format. Also the ICD might dither textures down, but I have seen no evidence that it does.

[Tagrineth]

Ah, OK, that makes sense. My mistake!

[Anonymous]

I thought that the V3 legitimately supported internal (ie prior to frame buffer) 32bit color, so that the quality diference between 32 and 16 bit on the V3 was simply a matter of wether the dithering occured prior to or after the texture was applied.

Then again, I did type the subject line as drunk idiot alert.

[darkblu]

Originally Posted by LwoodPDowd

I thought that the V3 legitimately supported internal (ie prior to frame buffer) 32bit color..

yes.
IMO both chuck's (i.e. voodoo's pixel) and bruce's (voodoo's TMU) pipelines had (at least) 8-bit-per-channel precision. as sst1's triangle setup interface took fixed-point channel values of even higher precision, i suppose the internal channel precision may have been higher than 8-bit. anyway, that's speculation on my part, i assume colourless is the only person around who could shed light on the matter.

..so that the quality diference between 32 and 16 bit on the V3 was simply a matter of wether the dithering occured prior to or after the texture was applied.

dithering occured at framebuffer writes only, and was independent of texture format/bitness. what colourless meant by tex dithering was sw dithering by the ICD prior to texture download* i.e. passing a 32bit texture to the ICD would (i.e. could) cause the former to be dithered by the latter, and only then downloaded to tex mem. FWIW, another form of encoding, and subsequently reconstruction, of higher bitness in less bits was 3dfx's narrow channel tex compression.

Then again, I did type the subject line as drunk idiot alert.

sounded pefectly sober to me


* loading a texture into tex mem is 'download' in 3dfx terms

[Anonymous]

If 3dfx only had the chance to design fear/mojo for the XBox, 3dfx might have survived. And XBox might now be selling more units, since it would have much better specs than PS2, than it does now.

[Anonymous]

Correction:

Should be "And MS would be selling more XBoxes..."

[Vince]

Originally Posted by DaveBaumann

Here you go, a little trivia/history for ya - Napalm as it was initially sold as to developers:

http://www.beyond3d.com/news/images/napalm.ppt

The pain....

Another piece of Trivia, who was the 3rd party whose programmable processor they wanted to use? What was the name of it?

Dave, did you ever hear why the cut the T&L chip? I assumed it was compatability reasons... was kinda shocked that they did it at the time because this left the higher end cards without AGP support (didn't knopw the 6K bridge idea yet).

[Laa-Yosh]

Originally Posted by xboxbought(bbot)

If 3dfx only had the chance to design fear/mojo for the XBox, 3dfx might have survived. And XBox might now be selling more units, since it would have much better specs than PS2, than it does now.

I don't think so. Better Xbox sales would be possible through a better software library...

[Steelwire]

Glad someone else jumped in! 3dfx in xbox = more xbox sales??? Complete and utter arse mate. Can't see it happening... get yer head straight - a VAST majority of the xbox target "audience" have NO idea who 3dfx are (where) OR VL (sorry IMGTEC) OR ATI or... etc... They know the branding on the box though...

It would make NO difference. XBox would sell more if it wasn't MS branded. A lot of console peeps have loyalties (strong loyalty) to Sega, The BIG N and Sony way before any desire to buy more MS products.

That is and was MS' biggest problem - they still haven't found a way around this. The irony of course is that in terms in international corporations with a control freak bent Sony and MS are fairly similar... it's all marketing... perhaps MS are worse!

[Anonymous]

Originally Posted by Vince

The pain....

Another piece of Trivia, who was the 3rd party whose programmable processor they wanted to use? What was the name of it?

Dave, did you ever hear why the cut the T&L chip? I assumed it was compatability reasons... was kinda shocked that they did it at the time because this left the higher end cards without AGP support (didn't knopw the 6K bridge idea yet).

Motorola... can't remember the chip designation. But it would've been slightly faster than a vanilla GeForce2's geometry...

Yes, it was compatibility. The cards with the Motorola bridge had an endless list of render issues, so it had to be dropped...

Even the 6K's pure bridge (no T&L) went through some five revisions, and then never hit the market at all anyway.

[darkblu]

Originally Posted by Tagrineth

Motorola... can't remember the chip designation. But it would've been slightly faster than a vanilla GeForce2's geometry...

Yes, it was compatibility. The cards with the Motorola bridge had an endless list of render issues, so it had to be dropped...

Even the 6K's pure bridge (no T&L) went through some five revisions, and then never hit the market at all anyway.

most intriguing. could you provide some more info/links?

[Dave Baumann]

Originally Posted by Tagrineth (can't log in)

Motorola... can't remember the chip designation. But it would've been slightly faster than a vanilla GeForce2's geometry...

Yes, it was compatibility. The cards with the Motorola bridge had an endless list of render issues, so it had to be dropped...

Actually, I'm fairly sure it was the Mitsubishi IMPAC-GE, which was supposed to have some level of load balancing with the CPU. Why was it dropped? AFAIK low performance in comparison to the cost - GeForce's T&L outperformed it, and probably so did CPU's by the time V5's were sceduled to be released.

A Google Search yeilds some information on the chip.

[AzBat]

Originally Posted by DaveBaumann

Actually, I'm fairly sure it was the Mitsubishi IMPAC-GE, which was supposed to have some level of load balancing with the CPU. Why was it dropped? AFAIK low performance in comparison to the cost - GeForce's T&L outperformed it, and probably so did CPU's by the time V5's were sceduled to be released.

A Google Search yeilds some information on the chip.

Dave, I believe you are correct. From what I heard Mitsubishi axed that division around the same time the Voodoo5 was launched. I suspect 3Dfx were pretty much screwed when that happened. All IMHO.

Tommy McClain