3dfx Rampage interview

[NFS4]

http://boards.fool.com/Message.asp?id=1250006014790000&sort=postdate

 

[Czar]

And it has absolutely no information about Rampage, hehe I allmost wrote Rambus, just we cant say anything at this moment.

 

[Soccerman]

indeed

that doesn't tell us anything! just that they're keeping tight lipped about everything this time. We'll see what happens is pretty much all we know about the Rampage.

 

[NFS4]

Made ya look

 

[Sephiroth_IX]

Just me, or is he kind of redundant throughout the interview?

 

[Michael]

Glad you liked it, Brandon <grin>.

They answered the question I wanted answered (as an investor). I didn't ask about specs, they would have either either 1) not told me, or 2) made me sign an NDA so I couldn't say anything.

Michael

 

[NFS4]

<< We feel much better on the bank-end process for Rampage than we did on the VSA-100 (which was very complex). >>


Hmmm, well isn't the Rampage just gonna get even more complex?

 

[Michael]

Brandon - think - if Rampage is more complex (more gates and more features) and yet they are saying the back-end (which is laying out all the routings for the chip itself) is less complex, what does that mean? Add on one word that has been bandied about to describe the &quot;Rampage&quot; and maybe you'll get it.

There is a bunch of info in just that short interview. I'm the one who &quot;broke&quot; the story about the VSA-100 delay last year and I did it by asking the same type of questions.

I'm going to write up some speculation later based on what I know (as opposed to official info from 3dfx).

Michael

 

[NFS4]

OK, now I'm confused

 

[Soccerman]

me too.. Michael, please specify!

 

[NFS4]

Spill the beans oh mighty one

 

[Michael]

I don't know, Czar seems to think that 3dfx's PR group in annoying and some of my info came from them. I'm scared he might call me names and hurt my feelings.

j/k

I'm working on it, I'll let you know as soon as I post something on TMF (since the questions I asked were from an investor's point of view, that's the most appropriate place to post something).

Michael

ps - for al those who think I'm just for 3dfx, read this post:

http://boards.fool.com/Message.asp?id=1250006014680000

 

[Mr.IncrediblyBored]

Add on one word that has been bandied about to describe the &quot;Rampage&quot; and maybe you'll get it.

Umm...tile rendering? Please explain what on God's green earth you are talking about. Deffered rendering is the only possibility i could think of.

But i thought they stated that Rampage was too far along in its development cycle to incorporate it. Hmm...

<----Hes confused

 

[Edwardo]

Michael...I am disturbingly intrigued by your hintings. Dammit...I want that damn answer. ANSWER MEEEEEEEEE!!!!
Seriously though...3dfx isn't going back to one-chip type designs are they? Surely that would make everything less complex...no?

 

[Mr.IncrediblyBored]

Oops i just realized tile rendering isnt one word



[Edit] Edwardo poses an interesting idea. One chip designs would be a good idea, sorta. It would be a good idea to have one chip be very powerful, but still have enough scalability to make a dual chip solution. That way instead of getting a monster 4 chip design, you get a more elegant 2 chip design and still have the power of the 4 chips.

Damn i could have worded that a bit better. Just think what ATI will do with the Radeon. One Radeon chip is pretty fast, but having 2 of them will be monsterous. [Edit]

 

[Soccerman]

Well the Rampage will be more powerful. they (3dfx) couldn't increase the rendering power of the old Voodoo architecture I guess.

In any case, it wouldn't make sence to continue on a chip with the same power as the VSA-100 now would it? of course all you nVidiots would seem to think so I guess.

My point is, 3dfx saw that they needed more power per chip, and they of course won't be dropping their scalability. That would be stupid, becuase if they did drop it, in the event of an emergency that they need alot more speed, boom, we have dual chip video cards.

In any case, this chip for this card will not be weak.

 

[Becks]

huh?

Dude like everyone is making no sense at all to me... aaaaaaaahhhhhhhhhhh

 

[Michael]

The word is &quot;modular&quot;.

Here's my speculation:

http://boards.fool.com/Message.asp?id=1250006014798000&amp;sort=postdate

Michael

 

[Mr.IncrediblyBored]

Something ive been kinda worried about. If Sage in fact is the separate T&amp;L chip that has been rumored, wouldn't that mean that the actual boards will end up getting bigger on account of the extra chips needed? I know the die shrink will make the chips smaller, but will they be small enough?

My last concern is the repeated texture data for each memory bank. 3dfx will need to move to 64 megs per chip relatively soon (sometime next year at the latest). 64 megs of DDR ram is expensive enough, but 128 megs of DDR?!?! :Q WoW that will not be cheap.

I guess thats why 3dfx released FXTC for free, hoping that developers would adopt it quickly, so that they [3dfx] can skimp on the amount of RAM needed.

4 pixel pipelines @ 250mhz on .18um process, making a gigapixel. Dual chip making it up to 2 gigapixel. Quad-chip --> 4 gigapixel :Q :Q

 

[Mr.IncrediblyBored]

Oh BTW, I don't quite understand your speculation Michael. I guess im stupid

 

[Michael]

One chip isn't going to make all that much of a difference in board size. There is no confirmation that &quot;Rampage&quot; will be multichip, but even so, board size shouldn't be much of an issue unless you go to a four chip version. The V5 5500 should be able to fit in all my cases. I have two smaller cases where the hard drive enclosure might be coming down too far and block something the size of a V5 6K, but the bottom drive bay is empty anyways and it was that big an issue I would just cut the bottom off the enclosure with metal shears.

I would guess that the Rampage boards would be smaller, but it would just be a guess. As far as I know, no Rampage chips exist yet.

Memory requirements would be identical for all the board makers. 3dfx's compression method can do S3TC as well. NVIDIA's current drivers automatically enable S3TC to try and pack as much into 32 meg as possible.

Michael

 

[Futuramatic]

It seems to me taht all that CFO said was &quot;don't be suprised if it doesn't make it for Christmas. In fact, don't expect it.&quot; Typical

 

[Thanatopsis]

I'm guessing what Michael is saying ( I haven't read the posts you made at the fool yet) is that the board layout for the VSA-100 was pretty complex, being a multichip solution. If 3dfx is saying it is less complex for a Rampage, it probably means they are looking at a single chip solution, hence less traces to draw on the PCB.

 

[Pete]

I think tape-out refers to chip complexity, not the complexity of constructing the actual board, PCB and all. First, Rampage should be .18mu, which should be simpler than 6-layer .25mu, though I'm only using logic here, not actual knowledge ;p. Second, since Sage will be another chip entirely, it simplifies both, as they have less to do per chip.

Hopefully with the move to .18mu they can trim power consumption as well. I'm actually waiting for a .18mu 200MHz core/200MHz DDR Voodoo5--now that would kick ass.

 

[rahvin]

<<It seems to me taht all that CFO said was &quot;don't be suprised if it doesn't make it for Christmas. In fact, don't expect it.&quot; Typical>>

Yes it it typical for a CFO to say that. They MUST say that or be liable for investor class action law suits. Read a 10K sometime.

<<I'm guessing what Michael is saying ( I haven't read the posts you made at the fool yet) is that the board layout for the VSA-100 was pretty complex, being a multichip solution. If 3dfx is saying it is less complex for a Rampage, it probably means they are looking at a single chip solution, hence less traces to draw on the PCB.>>

No, what Michael is speculating is that 3dfx is now designing their cores in a modular fashion. It has nothing to do with the board. Much like a CPU has multiple &quot;units&quot; 3dfx's next chips could have highly optimized modular sections as part of the silicon. Think of it this way, you design a really great pixel pipeline and lay the silicon traces out. In the next generation you want to add another pipeline so you simply add another one of the already designed layouts. If you plan ahead the whole design can be modular, I'm sure this is how nvidia designed the TNT to allow them to modify stuff quickly and effectively. It makes drivers easier and everything else work better.