carmack on Cg, P10 and apology to Matrox about DM

[Doomtrooper]

It is amazing how Marketing PR brainwashes the General public, if you think Nvidia has been the only company inovating new features into graphic cards over the past two years then pigs can fly.
There has been significant contributions by other graphic companies like ATI with truform and PS 1.4, Matrox with FAA . ATI has been slowly getting back the desktop Market it lost to Nvidia, and now 3Dlabs looks like a very strong contendor, SIS Xabre, Via stepping up and Matrox.
You factor all these facts in then the latest analyst reports of Nv30 being late and more competition then ever, these comments are laughable.



Two intersting comments from John Carmack hints whats happening @ Nvidia:

http://www.msnbc.com/news/758337.asp?0dm=B15LT&cp1=1#BODY

"NVidia has been stellar in terms of driver quality and support and doing all of the things right," says Carmack, who has been an outspoken evangelist for NVidia?s GeForce technology. "For the past few years, they have been able to consistently outplay ATI on every front. The problem is that they are about one-half step out of synch with the hardware generation because they did Xbox instead of focusing everything on their next board. So they are a little bit behind ATI."

http://slashdot.org/comments.pl?sid=34863&cid=3784210

Multi chip and multi card solutions are also coming, meaning that you will be able to fit more frame rendering power in a single tower case than Pixar's entire rendering farm. Next year.

Note the only active consumer level video card manufacturer presently with multiple chip solutions is ATI with the Radeon 8500 for Evans and Sutherland, the bridge chip was working as far as I know and proven in this statement.

http://www.ati.com/companyinfo/press/2001/4434.html

The RADEON 8500 is the ideal graphics processor to use in the multiple chip high-end Ensemble and simFUSION product lines. The RADEON 8500 chip has best-of-class graphics and video acceleration

 

[Gollum]

Note the only active consumer level video card manufacturer presently with multiple chip solutions is ATI with the Radeon 8500 for Evans and Sutherland, the bridge chip was working as far as I know and proven in this statement.

ATi the only one? * cough* Since I don't see any product to buy I hardly think that statement qualifies, even if we have little reason to doubt MAXX has matured based on the comments you showed. If talking hypothetically:
- While not yet entirely on the consumer level, 3D Labs has a lot of experience with multi-chip technology, they've been using multi-chip solutions for years in their high-end workstation boards and could probably use that experience for future generations.
- Also, don't forget Nvidia has 3dfx' still flexible and powerfull SLI technology (up to 16 chips on Quantum 3D's alchemy products, Quantum 3D also offers Nvidia based products, but I'm not sure if those are multichip or SLIable) and also has many of their former engineers. I have little doubt they'd be able to offer fully functional multichip technology in the future, if they so desired.
- PowerVR has provided flexible multichip solutions for arcade boards in the past, so they should also have the technological competence to handle a multichip consumer board, should someone be interested in licensing the technology.

I doubt this specific comment from JC ("multi chip and multi card solutions are also coming") should relate to ATi, or even just one company. I can imagine a number of companies developing such technology ...

 

[Doomtrooper]

I guess my point I was trying to make is not getting across is it, I'm not stating Nvidia nor PowerVR couldn't release a multiple chip solution, but ATI is by FAR the most experienced at this time as they have continued on developing multiple chip boards while most others have not, Nvidia has gone on record many TIMES that they were not interested in multi-chip solutions.
3Dlabs was not a consumer level Graphic company until 6 months ago, I wouldn't mind seeing the PowerVR multiple chip graphic boards..you have a link ?

 

[Tagrineth]

Note the only active consumer level video card manufacturer presently with multiple chip solutions is ATI with the Radeon 8500 for Evans and Sutherland, the bridge chip was working as far as I know and proven in this statement.

ATi the only one? * cough* Since I don't see any product to buy I hardly think that statement qualifies, even if we have little reason to doubt MAXX has matured based on the comments you showed. If talking hypothetically:
- While not yet entirely on the consumer level, 3D Labs has a lot of experience with multi-chip technology, they've been using multi-chip solutions for years in their high-end workstation boards and could probably use that experience for future generations.
- Also, don't forget Nvidia has 3dfx' still flexible and powerfull SLI technology (up to 16 chips on Quantum 3D's alchemy products, Quantum 3D also offers Nvidia based products, but I'm not sure if those are multichip or SLIable) and also has many of their former engineers. I have little doubt they'd be able to offer fully functional multichip technology in the future, if they so desired.
- PowerVR has provided flexible multichip solutions for arcade boards in the past, so they should also have the technological competence to handle a multichip consumer board, should someone be interested in licensing the technology.

I doubt this specific comment from JC ("multi chip and multi card solutions are also coming") should relate to ATi, or even just one company. I can imagine a number of companies developing such technology ...

Sorry hun, nVidia don't own no SLI tech.

It's privately owned by Hank Semenec, formerly of 3Dfx

 

[Doomtrooper]

EDIT: whoops that's not juicy enough . I'm hearing about some company supporting something that another company is championing that we were wondering if any other company is going to support.

http://www.theinquirer.net/28060208.htm

Even though this is from them , there is truth to that statement from what I'm hearing

 

[Tagrineth]

Erk... o.o

But this could cause problems because if ATI designs its own shading language, this would put developers into an invidious position, as they would have to decide which language they should use.

Into an... invidious position? nVidia... invidious... nVidia... nVidious? o.o; Coincidence? You decide. o.o

Right now Nvidia is winning this game since it has many people employed just to help developers and most of developers community program with "Geforce 4 MX" in mind.

OK. So. nVidia is encouraging developers to programme with GF4MX in mind?!?!?! Gawd, we can blame nVidia for both pushing fragment shaders too fast, AND for stifling their proliferation knowingly?!

 

[Crusher]

I don't see how you extract the idea that NVIDIA is encouraging people to develop for the GeForce 4 MX. I think that came out of your own anti-NVIDIA head. If developers are designing for the GeForce 4 MX, it's because they feel that is what a large percentage of their potential customer base uses. Would it sound better if he said they were developing with the GeForce 2 in mind? It's basically the same thing, and given the fact that I don't recall seeing a single game with a GeForce 2 listed as a minimum requirement yet, I don't understand how you feel that would be holding anyone back. It still sounds like progress to me.

Besides, that entire statement had no facts to back it up, and was merely the opinion of someone who writes for The Inquirer, which, IMHO, is about as valuable as a plastic ring from a box of Trix.

 

[pascal]

Second, IMHO the case is not closed. We have information directlly from JC himself. 3DLabs performed very well from quality standpoint, in fact it was very encouraging for him.

I don't there is much in the way of a comparison to be made here as, as I said earlier, the board given to JC (being only a week ago) was pretty much a final board. As JC said about 8500 and also the R300 he's used these were both very beta boards and drivers so they are much more likely to be in a 'raw' condition.

I understand the difference.

But there is all those developers claiming problems (id, Epic, Legend, ...) with ATI drivers in the recent past. What I see is "smoke" in the air (I dont see fire directlly). ATI is the one capable to change all this "smoke" situation which is very uncomfortable for consumers like me.

 

[Docwiz]

wow, Doomtrooper you are really reaching here...

I know you like ATI because you are from Canada which is basically the 51st state of the USA anyway.

You post negative news about Nvidia because you know you can't handle it that they are doing really good selling their products and creating good products. Their stock is down, but ATI's stock was down to 8 dollars long before Nvidia's stock was down to 17.

The sad thing is that I come here to beyond3d to read some good information and all I get is Biased BS.

I am convinced that just about all of you don't know much about anything and the few that do hide behind the excuse of NDA.

The rest is made up crap about what you want to happen. Just like Matrox's new card. Many of you wanted it to blow away Nvidia because you can't stand that they are in the Leader Position and you want to take them down.

It comes down to being biased or just guessing thats all this board does and it disapoints me for people supposed to be in the "know".

Instead of posting crap from Inquirer which is very anti-nvidia (I have not read one postive article about Nvidia on there) and they are almost 100 percent wrong all of the time and are basically a tabloid, you should not be posting at all.

Since when do british websites know whats going on in Canada and California?????????

The more I read of this forum, the more I get sick of it. Its not about facts anymore, its what you want to happen.

 

[Simon F]

Well if Santa Claus is entering into the 3d market I hope he is getting a better welcome than Matrox's return. AND do not give me any of that crap Santa is from Finland!

Completely off-topic, but Saint Nicolas came from Turkey.

 

[Simon F]

ATi the only one? * cough* Since I don't see any product to buy I hardly think that statement qualifies, even if we have little reason to doubt MAXX has matured based on the comments you showed. If talking hypothetically:
- While not yet entirely on the consumer level, 3D Labs has a lot of experience with multi-chip technology, they've been using multi-chip solutions for years in their high-end workstation boards and could probably use that experience for future generations.
- Also, don't forget Nvidia has 3dfx' still flexible and powerfull SLI technology (up to 16 chips on Quantum 3D's alchemy products, Quantum 3D also offers Nvidia based products, but I'm not sure if those are multichip or SLIable) and also has many of their former engineers. I have little doubt they'd be able to offer fully functional multichip technology in the future, if they so desired.
- PowerVR has provided flexible multichip solutions for arcade boards in the past, so they should also have the technological competence to handle a multichip consumer board, should someone be interested in licensing the technology.

As Gollum said, several companies have produced multi-chip solution (eg. the PowerVR-developed Naomi2 architecture). It's not that it is difficult.

The problem is that such systems are unlikely to be cost effective for consumer applications. For example, let's say an arcade system costs around (picking a figure out of the air) $1000. This is too much for a home user but doesn't matter in the arcade as it will pay for itself very rapidly provided it 'delivers the goods'. Gollum also mentioned the "Quantum" systems based around multiple 3dfx chips - these were not aimed at consumers either.

 

[andypski]

Not revealing information because you are under NDA is not called hiding - it is called integrity. This is an important distinction.

- Andy.

 

[Saem]

Docwiz,

You're summary judgement is poor by far. Your spin on why people say the things they do speaks volumes about YOUR preconceptions and biases. What seems to me is that Nvidia is your darling and you can't see it get bashed. But heh, I can't really say conclusively since I haven't read much from you.

That said, what you say as anti-Nvidia, I would think is optimism that this industry won't turn into a MS like "we ownz you and dictate what's good for you" one man show. Thanks to largely ATi this is likely not going to be the case any time soon, if at all. Also the statements are made due to a basic desire for some excitement in the industry/technologies. A lot of the people here get a kick out of the new developments in the industry and thus hope for more than what happened in the past while, which was a rather one-sided victory.

If you choose to take statements in such the negative light that you do, well then I'd recommend you go else where, perhaps, nvnews and their unbiased prespective is more to your liking. BTW, your previous post would best be in site feedback.

BTW, do you really think that websites which are run from California are allowed to frequent the companies they cover? I seriously doubt it, it's only on rare occaisions that the big guys get invited. So just because a few site coordinators are from the UK and other parts of Europe doesn't mean much.

 

[darkblu]

Thus quoth John Carmack:

I do need to get up on a soapbox for a long discourse about why the upcoming high level languages MUST NOT have fixed, queried resource limits if they are going to reach their full potential. I will go into a lot of detail when I get a chance, but drivers must have the right and responsibility to multipass arbitrarily complex inputs to hardware with smaller limits.

Could someone who is more into these things please explain the bearing this comment has on current alternatives?

Entropy

my read into JC's saying is that he believes there are still actions that need to be taken in order to persuade the HLSL vendors to get less ego(hardware)-centric and more abstraction-driven. apparently Cg was not very likable to him, maybe he took some particular lack of abstraction in it as a worryng trend/step in the wrong direction.

 

[DemoCoder]

What JC is saying is that a developer should not have to query the API and worry about how many register you have available, how many texture units there are, how many instructions long your problem should be.

Right now, when I code C, I don't care how few registers the Intel architecture has vs PowerPC, RISC, et al. The compiler takes care of that for me.

I think JC is advocating a language that can be compiled automagically into as many passes as needed. He specifically references the OpenGL paper which demonstrated how to take ANY Renderman shader and convert it into OpenGL via (a very large number) of passes.

 

[darkblu]

What JC is saying is that a developer should not have to query the API and worry about how many register you have available, how many texture units there are, how many instructions long your problem should be.

Right now, when I code C, I don't care how few registers the Intel architecture has vs PowerPC, RISC, et al. The compiler takes care of that for me.

I think JC is advocating a language that can be compiled automagically into as many passes as needed. He specifically references the OpenGL paper which demonstrated how to take ANY Renderman shader and convert it into OpenGL via (a very large number) of passes.

exactly. i was just iterating further on his thoughts. as i see the queried resources approach to be vendor-favorable in one way or another.

 

[Simon F]

Right now, when I code C, I don't care how few registers the Intel architecture has vs PowerPC, RISC, et al. The compiler takes care
of that for

I would agree with that for 98% of my code but sometimes I do need to consider the target platform(s) and (after profiling of course) make some #if versions. The number of registers or the type of instructions in the CPU can influence these decisions.

 

[darkblu]

Right now, when I code C, I don't care how few registers the Intel architecture has vs PowerPC, RISC, et al. The compiler takes care
of that for

I would agree with that for 98% of my code but sometimes I do need to consider the target platform(s) and (after profiling of course) make some #if versions. The number of registers or the type of instructions in the CPU can influence these decisions.

but simon, you do that target-specific optimizations at the final stage of coding, not at the stage of initial laying down of the algo in code. which you would have to do otherwise if you had no way to circumvent the target specs at all. need we quote knuth on optimizations here?

 

[DemoCoder]

Yes, as Knuth said in the Art of Computer Programmer "premature optimization is the root of all evil".

 

[KimB]

Thus quoth John Carmack:

I do need to get up on a soapbox for a long discourse about why the upcoming high level languages MUST NOT have fixed, queried resource limits if they are going to reach their full potential. I will go into a lot of detail when I get a chance, but drivers must have the right and responsibility to multipass arbitrarily complex inputs to hardware with smaller limits.

Could someone who is more into these things please explain the bearing this comment has on current alternatives?

Entropy

I'm reasonably certain that he's commenting specifically about Cg. I don't know if the other languages have this limitation (Hopefully with JC's comments here, Cg will be updated, before its final release, to support automatic multipass). Unfortunately, I can't find the exact quote at the moment, so I am just operating from memory...I'll continue to look a little bit more.