Carmack on Parhelia, revisited!

[afterburn]

Snipped from his latest .plan:

June 27, 2002
-------------
More graphics card notes:

I need to apologize to Matrox -- their implementation of hardware displacement
mapping is NOT quad based. I was thinking about a certain other companies
proposed approach. Matrox's implementation actually looks quite good, so even
if we don't use it because of the geometry amplification issues, I think it
will serve the noble purpose of killing dead any proposal to implement a quad
based solution.

I was so sure I had registered here like ages ago.. guess not..

[P5ycho]

http://forums.murc.ws/showthread.php?threadid=34158

[Zed_L]

My english ain't that good so could you in other words explain what does Carmack mean by "I think it
will serve the noble purpose of killing dead any proposal to implement a quad based solution." ???

Is that a positive statement towards Matrox's hardware displacement mapping?

And how is it possible that Mr.Carmack hasn't followed what happens in the business, I mean how could he not know what HDM means?

[Zed_L]

Now that I've read the line one hundred times and really thought about it I think he says: The good thing is that Hardware displacement mapping puts an end to developement that is going on with quad based solutions, because HDM is better.

Am I right?

[afterburn]

I think you are, yes.

[Buback]

I know that he is a smart guy and all, but how can Carmack try to make a special case where parhelia will win aginst gf4 and yet not know what hardware displacement mapping is?

it makes me wonder if he really knew the features of the card in order to exploit them correctly.

buback

do you like Led Zeppelin Zed?

[Fenrir(AVA)]

I am an avid readre of Carmack's .plan updates (he needs to update more often), he is the type of guy that thinks there is only one correct way of doing something. With hardware from what I noticed he is more concerned with the pipeline then anything.

[MrGaribaldi]

Originally posted by Buback
I know that he is a smart guy and all, but how can Carmack try to make a special case where parhelia will win aginst gf4 and yet not know what hardware displacement mapping is?

it makes me wonder if he really knew the features of the card in order to exploit them correctly.

He only cares about technology that's directly related to his work...

So he makes a test in Doom where he optimizes the code as much as possible for the P, but he doesn't use HDM, as it has no relevant use in Doom... It (I guess) relies heavily on shaders, and thus the GF4 is able to beat it...

Not very good wording in that last part... but I can't think of another way to write it, so...

[WaltC]

Right...

Originally posted by Fenrir(AVA)
I am an avid readre of Carmack's .plan updates (he needs to update more often), he is the type of guy that thinks there is only one correct way of doing something. With hardware from what I noticed he is more concerned with the pipeline then anything.

Carmack is rather tunnel visioned at times. If something isn't directly applicable to his current needs his tendency is either to overlook it or else misunderstand it. This comes from having his nose buried too long in nVidia's posterior, IMHO... Or maybe it comes from nVidia having its nose buried too long in Carmack's posterior, I'm not sure. Maybe, like a couple of dogs they're running in circles behind each other....NAH....

Seriously, he's been optimizing around nVidia for quite sometime, and most of his code is so graphics-card-engine specific that even with an SMP implementation of Q3 it hardly runs better on 2 cpus and in some cases runs worse on 2 cpus than on 1. If he'd write his software to be much more evenly divided between the cpu and gpu he'd see quite substantial gains from his SMP code.

In a way this is kind of symbiotic, because nVidia driver code is extremely cpu dependent, often monopolizing 100% of the cpu. But this still doesn't help Q3's SMP performance any and won't unless nVidia produces an SMP-driven videocard driver...

In this instance I think his apology has to do with the fact that Matrox's approach here doesn't push a standard he personally opposes, as he first thought. Even though he's not going to use what Matrox offers in the way of displacement mapping, he's relieved to see that it falls in line with what he thinks is a non-negative direction for the 3D market.

Notice that his comments really don't concern the 3D market from within the context of ID Software products, but rather usually are universal comments having to do with his opinions on the directions all other 3D coders should be taking.

[DosFreak]

"Optimizing for Nvidia for quite some time"

heh. What's your definition? The ONLY engine that he has released that you could MABYE make a case for Nvidia optimization is the Quake 3 engine. That's it.

I do not remember what cards they were using back then for Quake 3 development but it sure as heck wasn't a TNT1. Most likely a Voodoo 3.

Well, There goes your theory.....