An Interview with NVIDIA's David Kirk

[NFS4]

Just recently we had the chance to speak with NVIDIA's Chief Scientist, Dr. David Kirk. Over the course of the interview, we discussed a range of topics on the GeForce FX architecture, how it compares to RADEON 9800, as well as some of the early benchmark results with games such as Half-Life 2 and Tomb Raider.

http://firingsquad.gamers.com/hardware/kirk_interview

 

[AndrewKu]

Well the complete verdict will have to hold off till the public release of the Det 50s.

 

[orion7144]

It was a very good interview. Sheds some light on what is going on.

 

[Luagsch]

nice insights on the cg and fp24/32 stuff.
but either way fp24 seems to be the way to go until dx10 is out (probably toghether with longhorn). i will be interesting to see what nvidia is gonna do.

 

[SilentRunning]

Originally posted by: orion7144
It was a very good interview. Sheds some light on what is going on.

Sheds some light? Sure seemed to me to be some fancy dancing around the questions for damage control.

 

[hominid skull]

The insight of this interview is to see how far the denile goes. They still think that their hardware is better, one of the reasons that the Radeon 9xxx series is faster, they say, in todays games is that developers used radeon cards to develop on. That has a little to do with it but not the huge margine between the two cards.

They are still going to use optomisations in the new 50.xx dets, it's the only way that they can get the card to perform at a reasonable rate - replacing FP24 shaders with FP16 shaders. Good point he made about the FP32 architecture, meaning that running FP24 code the FX has half the performance, roughly speaking, than the radeons.

 

[yhelothar]

interesting interview.. thanks

 

[Confused]

That seemed to be another "there's nothing wrong with this" notion.

He mentioned that you can't use one game as a benchmark, (probably implying HL2), but that is one of the most anticipated releases this year, and if the NV3x doesn't stand up to it, then they're gonna get burned.


Confused

 

[Avatar007]

I don't think the interview was so good, and I would have preferred that the interview was taken by a person who had more knowledge about DirectX9 programming and Shader techniques.
In between lines I read more PR stuff then really technical evidence.
'Performance issues with Half-Life 2 and other DX9 games will all be solved in software...'
'hardware coded sin en cos functions are so good...'. I've heard that so many times.
'FP32' is the way to go, 'FP24' only temporally and not giving enough resolution for geometric calculations...".

Dirk loves DirectX9 because it makes it easier for the developer to use advanced hardware features on his cards.
Half-Life 2 uses Directx9, PS 2.0, HDN with 24-bit FP geometric calculation and it does this beautifully, on ATI hardware.
The only way to make it work on Nvidia boards with comparable performance is by 'optimizing' code and 'tweaking' shader programs to fit them in FX hardware, making it much harder for developers to use these advanced features.

I hope one day Nvidia will admit that for current hardware the FP24 path is the most optimal and that they have made (another) mistake with their nv35 generation, wanting to jump too fast.

FP32 will be the future, sure, but not for now and certainly not with an nv35 board.

Just an opinion.

 

[Regs]

I don't really want to believe what he has to say while he is still on Nvidia's payroll. Same with ATI. I'll bet once Nvidia gives Dave the pink slip he'll open up much more.

David also sounded constipated throughout the interview, or that could just be due to the lack of writing skills the author of that interview possessed. No Larry King material here.

Not one thing David stated sounded clear or without some opinionated assumption. He didn't sound defiant. But then again, I can't expect every one to be a good interviewee.

Does that mean I should not open my eye's to read what he has to say? No. It does somewhat put a point of view on things; could get brainwashed from reading all the fanAtics material.

 

[Goose77]

All i got to say is, " WOW, that is the biggest pile of doooodi i have ever seen!!!"

FiringSquad: One of the things that ATI has kind of said, or least they were suggesting at Shader Day is the fact that they can do more floating-point operations than you guys can. How would you respond to those types of statements?


Kirk: Well I guess the first response would be of course they would say that. But I don?t really see why you or they would think that they understand our pipeline, because in fact they don?t. The major issues that cause differing performance between our pipeline and theirs is we?re sensitive to different things in the architecture than they are so different aspects of programs that may be fast for us will be slow for them and vice versa. The Shader Day presentation that says they have two or three times the floating point processing that we have is just nonsense. Why would we do that?

i love this! It SEEMS like he is saying that if you optimize your code for NV it will work faster and vice versa, well if you optimize any code to a specific architecture of course it will perform the best. Problem here is that ATI architecture will work with any generic code really well, and Nvidias SEEMs to only work well with optimized code!!

it looks like Nvidia is using a frontend loader instead of a shovel for all that KAKA!

 

[NFS4]

Merrill cuts Nvidia to 'sell'; stock falls

 

[thorin]

OMG what's with these guys (FS) it seems like they can barely even type.

"Could you give us specific examples of where maybe you feel you guys, you mentioned you guys can do some things better than they can, can you give us some specific examples of that?"

"Do you feel that fact that you guys, your hardware came out later -- does that also contribute to the initial performance that?s coming out in terms of the DX9 titles that have been benchmarked with?"

"Do you feel that in terms of the Half-Life 2 performance numbers that were released recently?do you feel that maybe you guys were, I don?t want to say given a bad rep, but maybe an unfair deal?"

I'm surpised David Kirk could even figure out WTF he was asking about without laughing his a$$ off so hard that he couldn't even respond. It looks like the FS author/interviewer really likes feely guys ????

Thorin

Edit: You can just ignore me if you don't agree, I've just been really "moody" (for lack of a better word) about the world's inability to communicate well lately.

 

[Alkali]

Originally posted by: NFS4
Merrill cuts Nvidia to 'sell'; stock falls

This will finally get the message across, even if they have ignored it so far in nVidia management. Money talks.

 

[Acanthus]

Originally posted by: Alkali

Originally posted by: NFS4
Merrill cuts Nvidia to 'sell'; stock falls

This will finally get the message across, even if they have ignored it so far in nVidia management. Money talks.

NVIDIAs stock rubberbands around daily... i wouldnt base their stock price on anything...

It went from $13 to $41 in under a week before, then was back $26 in a couple days. I bailed out at around $26 :brokenheart: :brokenheart:

 

[Acanthus]

It does however explain some of the DX9 performance issues in the interview. The FX line is running 24 bit precision in emulation because the chips are designed to run 32 bit precision natively.

It "just so happens" that the 2 DX9 titles out now are in 24bit, and running in emulation, which is a large performance hit.

Of course thats just what "I" gathered from this interview, and it does make sense.

 

[AmdInside]

Originally posted by: Acanthus
It does however explain some of the DX9 performance issues in the interview. The FX line is running 24 bit precision in emulation because the chips are designed to run 32 bit precision natively.

It "just so happens" that the 2 DX9 titles out now are in 24bit, and running in emulation, which is a large performance hit.

Of course thats just what "I" gathered from this interview, and it does make sense.

NV3x supports 16-bit and 32-bit FP. They were under the impression that Microsoft would go with 16-bit FP, possibly even 32-bit. At last minute, MS decided to go with 24-bit FP precision for DX9 spec which really screwed NVIDIA. It would be as if AMD Athlon 64 designed their chip with the understanding that Microsoft would make a 64-bit XP OS for them and then at last minute, Microsoft said, nah, we'll make it 48-bit instead. AMD can't easily go back and redesign their hardware architecture which runs at 32-bit and 64-bit so they end up screwed.

 

[Acanthus]

Originally posted by: AmdInside

Originally posted by: Acanthus
It does however explain some of the DX9 performance issues in the interview. The FX line is running 24 bit precision in emulation because the chips are designed to run 32 bit precision natively.

It "just so happens" that the 2 DX9 titles out now are in 24bit, and running in emulation, which is a large performance hit.

Of course thats just what "I" gathered from this interview, and it does make sense.

NV3x supports 16-bit and 32-bit FP. They were under the impression that Microsoft would go with 16-bit FP, possibly even 32-bit. At last minute, MS decided to go with 24-bit FP precision for DX9 spec which really screwed NVIDIA. It would be as if AMD Athlon 64 designed their chip with the understanding that Microsoft would make a 64-bit XP OS for them and then at last minute, Microsoft said, nah, we'll make it 48-bit instead. AMD can't easily go back and redesign their hardware architecture which runs at 32-bit and 64-bit so they end up screwed.

Exactly what i thought. Explains volumes.

 

[DaveSimmons]

> At last minute, MS decided to go with 24-bit FP precision for DX9 spec

I've read that nV walked away from working with MS on DX9, then came back and was surprised that MS hadn't done what nV wanted. So it was nV's own fault for not participating fully during the DX9 process.

 

[grant2]

If you look at processors FP24 doesn?t exist anywhere else in the world except on ATI processors and I think it?s a temporary thing. Bytes happens in twos and fours and eights -- they happen in powers of two. They don?t happen in threes and it?s just kind of a funny place to be.

This guy doesn't know what he's talking about, why did FS waste their time asking him technical questions?

a) colour precision is in BITS not BYTES
b) there's no inherent reason that precision HAS to be in powers of 2. 3 is a natural "place to be" if you have 3 channel colours! (RGB)

 

[jbond04]

Originally posted by: grant2

If you look at processors FP24 doesn?t exist anywhere else in the world except on ATI processors and I think it?s a temporary thing. Bytes happens in twos and fours and eights -- they happen in powers of two. They don?t happen in threes and it?s just kind of a funny place to be.

This guy doesn't know what he's talking about, why did FS waste their time asking him technical questions?

a) colour precision is in BITS not BYTES
b) there's no inherent reason that precision HAS to be in powers of 2. 3 is a natural "place to be" if you have 3 channel colours! (RGB)

First of all, I'm pretty sure that David Kirk knows what he's talking about. He's won several awards, and you don't get to be nVIDIA's chief scientist by "not knowing what he's talking about". Although color precision is expressed in bits, I believe he was talking about bytes in general computer terms, not necessarliy "bytes of color precision". Finally, although 24bits of color precision is logical since you have 3 different channels of color, there is a fourth channel used in computer graphics: the alpha channel. The alpha channel deals with transparency using 8bits of grayscale precision, totalling 32bit (RBGA). I'm not sure how pixel shaders work, so I really can't comment further on why 24bits may be better or worse than 32bits, but Dr. Kirk does have a point when he asks, "what else in the computer world has 24bits of precision?". If you look at the color settings for your desktop, most likely the choices are 16 or 32bits; and in games, you can choose a color depth of 16 or 32 bits.

Again, I don't have enough knowledge to draw any sort of conclusion here, but Dr. Kirk's argument seems logical enough. Also, DaveSimmons do you have a link that shows that nVIDIA walked out on the development of the DX9 spec? Where did you 'read' that?

 

[ElFenix]

i'd like to point out that 32 bit color is 3 8 bit color channels. 24 bits. with an 8 bit alpha.

 

[jbond04]

Originally posted by: ElFenix
i'd like to point out that 32 bit color is 3 8 bit color channels. 24 bits. with an 8 bit alpha.

Yeah...I'm aware. That's why I was wondering if the extra 8bits of precision have any sort of benefit for pixel shaders...but I don't know enough about how they work to comment on that. Does the alpha channel do anything for pixel shaders?

 

[arcenite]

Originally posted by: jbond04

Originally posted by: ElFenix
i'd like to point out that 32 bit color is 3 8 bit color channels. 24 bits. with an 8 bit alpha.

Yeah...I'm aware. That's why I was wondering if the extra 8bits of precision have any sort of benefit for pixel shaders...but I don't know enough about how they work to comment on that. Does the alpha channel do anything for pixel shaders?

In otherwords... You don't need 32bit.. 16bit is all you'll ever need. Love, 3Dfx.

Bill

 

[grant2]

Originally posted by: jbond04 I'm not sure how pixel shaders work, so I really can't comment further on why 24bits may be better or worse than 32bits, but Dr. Kirk does have a point when he asks, "what else in the computer world has 24bits of precision?". If you look at the color settings for your desktop, most likely the choices are 16 or 32bits; and in games, you can choose a color depth of 16 or 32 bits.

My desktops have had 16, 24, & 32 bit colour modes for as long as I can remember.

Dr. Kirk's argument seems logical enough.

the only "logic" he's presenting is that 32 bit is a "round" number (in binary terms). Hey may be a smart guy but it sure doesn't show when his comments are just specious appeals to laymen!

I'm not be a chief-anything, but whenever I've programmed with colours (java, web pages) they've been expressed in 3-byte (i.e. 24 bit) values.

Kirk's argument "that's how it's always been done, so that's how we'll do it now" doesn't carry much weight with me, graphics companies are supposed to be INNOVATIVE, so that's the worst excuse they can give for any decision.