Questions for Gary T (yeah, that Gary T)

[Tagrineth]

Definitely ask if nV would consider a multi-chip or SLI solution. Wouldn't it have been to their benefit, with the NV30 rumored to arrive late, to be able to just tell people to buy another GF4 and SLI for double the performance? That way chips would be in produciton longer, OEMs would make more money as their product cycles would lengthen, and consumers could choose either reasonable performance now, and more later, or extreme performance right off the bat, for an extreme price.

I thought 3dfx made quite a bit of money off the Voodoo 2. It surprised me that they didn't continue that dual-card mentality. Particularly now that dual-monitor PCs are almost commonplace--the benefits of two separate 2D engines, as well as two 3D engines that would be smart enough to operate as one (via an SLI cable), would be fantastic. 3D speed, and the ability to simulatneously do some 2D work on your other monitor using the extra 2D portion of the second "GPU." Would the engineering hurdle be too high for such a solution? (They could always offer a dirt-cheap model without SLU capability, for the mass market.)

First off, 3dfx did use SLI with Voodoo4 and Voodoo5. Not dual-card, but still dual-chip. The trouble here was AGP... it is possible, but extremely complex, to SLI an AGP with a PCI. Timings are too different.

Second, nVidia does not own SLI. Hank Semenec does. And he didn't go to nV

 

[Colourless]

As I understood it, Brian Hook was the person who designed the initial Glide Implementation (v0.9), however that was all he did. He wasn't involved in any further developement of the API.

 

[Tagrineth]

::giggles::

2 was 3dfx mojo a result, or maybe a port from the Sega gamecube Blackbelt project to the Pc market?
I mean, The blackbelt project was about a GPU which was desined around Tiltled architectue, and had i thought 63 bit colors, don't remeber good.

Sega Gamecube?

Gamecube Blackbelt?

Uh.

Blackbelt was supposed to be for Dreamcast, and IIRC it became Voodoo2.

 

[elimc]

What are the exact complications of moving to RDRAM, and has nVidia ever analysed RDRAM for its GPUs?

 

[psurge]

There is a trend towards more programmability in the GPU, and letting a GPU handle more and more duties :
- triangle setup
- T&L
- HOS tesselation
- displacement mapping
- (future) NVidia has mentioned programmeable primitive assembly in some online paper (i lost the link)
- (future) there is talk of hardware scenegraph support being nice)

Also, increasingly large amounts of on board RAM is being shipped on graphics cards, hence allowing more and more scene data to be stored locally.

Given this trend, how would describe the relationship between a future CPU and GPU? Do you think that applications will eventually be able to execute their entire rendering algorithms on the GPU - i.e. no OpenGL/DirectX driver calls are involved - the CPU sends application specific state to the card, which is then interpreted by the rendering code running on the GPU?

In a game situation much of the data required for 3d rendering is required for other aspects of the game (collision detection, physics, visibility checks, etc...). Since much of this data will be local to the GPU, and the GPUs ability to render geometry far outstrips a CPUs ability to process it, do you foresee any of the afore-mentioned tasks moving into the domain of the GPU, or at least perhaps that of a some sort of DSP or general purpose MPU dedicated to them on the same board as the graphics chip?

 

[Vince]

I hate to get off topic, because Gary is an amazing person to ask about the more future 3D philosophy form of questions due to his foresight and knowledge, but...

Second, nVidia does not own SLI. Hank Semenec does. And he didn't go to nV

Right... Who the hell's Hank Semenec? I've never heard of him and nether has the US Patent Office. This sounds like some X3dfx BS that really needs to end and these people need to move on.

 

[Tagrineth]

I hate to get off topic, because Gary is an amazing person to ask about the more future 3D philosophy form of questions due to his foresight and knowledge, but...

Second, nVidia does not own SLI. Hank Semenec does. And he didn't go to nV

Right... Who the hell's Hank Semenec? I've never heard of him and nether has the US Patent Office. This sounds like some X3dfx BS that really needs to end and these people need to move on.

Hank's the creator of SLI and one of 3dfx's lead engineers

And now I shut my mouth before I shove my foot in it...

 

[Bonehed316]

got any docs saying who cleaned the bathrooms?

maybe gary knows. why dont you ask him. lets see if he knows anyone named "vince".

 

[Devo6486]

Excellent idea. They said no 3dfx questions, but just this one might be okay, since this "Vince" person didn't really work for 3dfx only says he did.

Just ask Gary,

"Was there anyone named Vince that you can recall ever working on anything important, or in a position to be given any special knowledge, about anything?"

 

[Typedef Enum]

Yeah, I can see that happening.

me: "Gary, can you please tell the viewers out there about Vince? wink-wink...You know, VINCE!?"

(meanwhile, 99.9999999% of those reading are like, "Who the hell is this Vince character?")

 

[Ailuros]

After the rather comedic interruption can we get back on topic please?

 

[Mephisto]

Second, nVidia does not own SLI. Hank Semenec does. And he didn't go to nV

Really?

http://groups.google.ch/groups?hl=de&lr=&ie=UTF-8&th=c18a94b8ab29bf87&rnum=3

I am one of the four systems engineers that worked on the
6000 and none of them made it into private (not engineers) hands .

Here he claims that he is a system engineer. He also claims that none of the Voodoo 5 6000 made it into private hands, which is definitly not true (ask Colorless or the guy here: http://www.maximumpc.com/features/feature_2001-12-05.html ).

Aside from this, even if Hank is what he claims he is, he definitly can't "own" SLI, even if he designed it. If he worked for 3dfx, his work is owned by the company. And as the company sold all the IP to Nvidia ...

 

[MfA]

3dfx never even filed a patent, noone owns it.

 

[Simon F]

3dfx never even filed a patent, noone owns it.

And I would think it would be unlikely that such a patent would be granted - I'm sure that some of the workstation graphics systems of the late 80s and early 90s were using this sort of technique. I think that the SGI Reality Engine, for one, must be very similar.

 

[MfA]

Had multiple-scanline interleaving been explicitly described in literature at the time? If not Im sure they could have gotten a patent on it. Or were patent examiners better at their job in the good old days?

 

[Simon F]

I just had a very quick look at "High-Performance Polygon Rendering" by Akeley and Jermoluk of SGI in SIGGRAPH 1988. Looks to me that they have 20 polygon/texturing units arranged so that each processes a unique screen pixel in each 4 scanline by 5 column block.

That would seem to constitute prior art to me.

Sarcastic Addition: As to whether patent examiners (in the US at least) would look at SIGGRAPH, well, I doubt it. I get the feeling that they only look at other US patents.

 

[It is I]

well...

Aah darned, i didn't read something not good... well if we can't ask something about 3dfx and his technologiy (where Nvidia made and is going to make very much use of it) I totaly do not see whhy this intvervieuw should be so important and/or usefull.

But again, STOP ARGUEING FFS!
You got the chance to ask questions, and you guys go argueing, comone now!
Go ask questions already!

 

[Johnny Rotten]

Aside from this, even if Hank is what he claims he is, he definitly can't "own" SLI, even if he designed it. If he worked for 3dfx, his work is owned by the company. And as the company sold all the IP to Nvidia ...

Exactly. I've never worked for, or heard of a company, that allows employees to keep patents of work they do for the company. Just doesnt happen.

 

[Devo6486]

Wrong.

All companies do it.

Hank Semenec owns the patent to SLI technology, and that's a fact.

He also owns(ed) some of the memory interface patents.

You've got to realize that "3dfx" didn't make technology, its core people did. Many of the patents 3dfx used actually belong(ed) to Scott Sellers and Gary T. personally.

The same is true for nVidia.

I have documents proving all of this, but I'm obliged not to release most of them. I'll check up and see if there are a few that are okay to release or not. Most likely not, but who knows.

However, anyway, this topic did sort of get hijacked, so sorry for that, Typedef Enum.

I suggest we get back on topic here.

Here, I'll help...

I'd say ask him about how he thinks "extra" features are going to affect the graphics market. For example, nVidia boasted and hyped their hardware TnL since the GeForce256. However it has been a consistent trend that developers ignore such features because nVidia and/or other hardware developers design new ones within one or two product cycles to get more buyers. Hardware TnL has now been around what is practically forever in the hardware world, and yet still even the newest games such as UT2003 don't require it, even if they do allow its use to help things. Most things that come about become required relatively quickly. Most new OpenGL games are now requiring at least OpenGL 1.2 and D3D games D3D8. However TnL came about in the earlier days of DX7, and yet no real trend of requirement exists.

 

[John Reynolds]

Wrong.

All companies do it.

I somehow doubt that. Otherwise, predatory recruitment tactics would eviscerate a company, IT-wise, that lost a long-standing or even founding engineer. Technology advancements are created on company time, using company offices and company equipment/tools.