John Carmack on PhysX

[Aug 21, 2009, 11:46 am ET] - Share - Viewing Comments

There is a video on the YouTube highlighting comments John Carmack made during a Q&A session at QuakeCon 2009 where he was asked about hardware-based physics solutions such as PhysX, now owned by NVIDIA (thanks game\phys). The id Software technical director was already on record as being against dedicated physics processing units, but in his answer he elaborates. Here is our hand-crafted transcript:

Okay, hardware physics. I think I was fairly public about my thinking that that was a really bad idea, and in fact it was pretty clear to me from early on that the whole idea for that was to do a startup to be acquired.

I didn't feel it was actually... I actually had a really quite negative opinion about stuff like that because they went out, they evangelized, they got some people to buy a piece of hardware that I didn't think was actually a good technical direction for things on there; certainly was going to be supplanted by later generations of more integrated compute resources on there. I don't think it was a good idea, I certainly wasn't a backer of the company, and I hope NVIDIA didn't pay a whole lot of money for them.

View

34 Replies. 2 pages. Viewing page 2.
< Newer [ 1 2 ] Older >

14.

Re: John Carmack on PhysX

Aug 21, 2009, 14:15

DangerDog

PhysX turned out to be a joke of a particle effect enhancement, shoot the ground an 1000 chunks of rock go flying around. The software based solutions from the likes of DICE and Crytek completely out class anything PhysX based.

13.

Re: John Carmack on PhysX

Aug 21, 2009, 14:13

Hazard

Oh and by the way, Carmack denounced Direct3D early on and that's why they stuck with OpenGL but now we know that Direct3D had a bright future because it would evolve faster and be better targeted for gaming.

The reason why he denounced Direct3D is because it was a pretty bad API. But it steadily improved and by D3D7 Carmack was on the record as saying that it has finally caught up to OpenGL in usability.

That's the reason: it used to be bad, then it got better. And JC definitely was right about that.

Don't know about the PhysX stuff, though. I guess it depends on how expensive it is for nvidia to include the technology. If integrating it is not much of a technical problem then I it is a good thing, even if it is a gimmick for cool looking smoke.

If on the other hand a considerablly large portion of the graphics card sales price is used to pay for it, then I think JC is right and it is not worth it.




Fully automatic backups with Ocster Backup Pro 3 http://www.ocster.com

12.

Re: John Carmack on PhysX

Aug 21, 2009, 14:08

Creston

More than 2 for sure: http://en.wikipedia.org/wiki/Physx#Games Do those new games still support the actual CARD, or do they only support Nvidia's PhysX implementation? I was referring to the games that would detect that you had the actual card installed. If that still works even with new games, then okay. Creston

11.

Re: John Carmack on PhysX

Aug 21, 2009, 13:55

Wolfox

In the end, I think 2 games support the damn thing, and do so poorly. More than 2 for sure: http://en.wikipedia.org/wiki/Physx#Games Other than that, I agree with you.

10.

Re: John Carmack on PhysX

Aug 21, 2009, 13:52

LittleMe

How does the OGL/D3D/Glide/etc. war relate to this topic? Nobody is talking about a standardized physics API...

It does relate.. And you are false, there is definitely talk about standardizing compute shaders with a focus on accelerating physics for gaming.




The real zombie game

Re: John Carmack on PhysX

Aug 21, 2009, 13:50

LittleMe

With the direction technology is moving, it's an incredibly stupid idea to have hardware accelerated physics.

DirectX 11 is including the capability to do physics on the GPU so your statement is incorrect with "the direction technology is moving."

Here I'll spell it out for you and your categorically false statement:

Target: Interactive Graphics/Games
• Image/Post processing:
– Image Reduction, Histogram, Convolution, FFT
• Effect physics
– Particles, smoke, water, cloth, etc.
• Advanced renderers:
– A-Buffer/OIT, Reyes, Ray-tracing, radiosity, etc.
• Gameplay physics, AI, etc.

Source:
http://s08.idav.ucdavis.edu/boyd-dx11-compute-shader.pdf

MS page on DX11 compute shaders:
http://www.microsoft.com/downloads/details.aspx?FamilyID=9f943b2b-53ea-4f80-84b2-f05a360bfc6a&DisplayLang=en




The real zombie game

Re: John Carmack on PhysX

Aug 21, 2009, 13:49

Creston

Carmack denounced Direct3D early on and that's why they stuck with OpenGL but now we know that Direct3D had a bright future because it would evolve faster and be better targeted for gaming. Direct3D "won" because of two very simple reasons 1) Microsoft was on the OpenGL board 2) Direct3D was the default API on the one operating system that 99% of the target market was running. Not because of any inherent supremacy in quality. Also, Carmack stuck with OpenGL because at the time they were porting all their games to Linux, and obviously Direct3D isn't going to work on Linux. I think dedicated physics acceleration is not a bad idea, it just needs to be handled by a CPU core, rather than yet more separate hardware. The whole idea of a PhysX card was just plain stupid. People want FEWER fucking add-on cards, not more. It's fun to remember all the monkeys who bought one bleating how awesome it was, and how they'd made a smart 300 dollar (or whatever it was) investment. In the end, I think 2 games support the damn thing, and do so poorly. Nvidia's idea of putting it on their GPU, but then letting it take horsepower from the GPU itself, isn't very smart. Do people want cooler physics? Sure. Not at the cost of 30% of their framerate, however. Creston

Re: John Carmack on PhysX

Aug 21, 2009, 13:44

PHJF

For the past few years games have become more and more reliant upon the GPU while CPU demand hasn't really moved forward. CPUs are getting faster and getting more cores, and games are doing nothing to utilize all that extra power. Let the CPU handle physics.

How does the OGL/D3D/Glide/etc. war relate to this topic? Nobody is talking about a standardized physics API...

This comment was edited on Aug 21, 2009, 13:46.




Steam + PSN: PHJF

Re: John Carmack on PhysX

Aug 21, 2009, 13:43

LittleMe

Oh and by the way, Carmack denounced Direct3D early on and that's why they stuck with OpenGL but now we know that Direct3D had a bright future because it would evolve faster and be better targeted for gaming. OpenGL was OK for gaming but that wasn't what it was really designed for, irc. (opengl was more for rendering 3D on UNIX workstations)

Carmack isn't always right.. He's no god, unlike Phil Conners.




The real zombie game

Re: John Carmack on PhysX

Aug 21, 2009, 13:41

ASJD

"Yeah people said that about dedicated 3D graphics, too" Never heard anyone say that back in the day. I thought it was wonderful at the time, taking the calculations off the CPU. And it was obvious to anyone technical enough to have seen stuff like Silicon Graphics workstations, or even what they could do. Kind of a weak argument really. 'People said'... With the direction technology is moving, it's an incredibly stupid idea to have hardware accelerated physics.

Re: John Carmack on PhysX

Aug 21, 2009, 13:30

LittleMe

It's still dumb and I still agree with Carmack.

Yeah people said that about dedicated 3D graphics, too. I think you are judging it on the poor implementations so far but not on its potential. In the 1980's CPU's didn't have math co-processors. Now they are standard and games use it. Also the MMX/SSE instructions are another example of functions that can be accelerated on the CPU and games do use those and the idea is very similar.

The primary problem with accelerating physics for games is that it needs to be an industry-wide standard accessible to all.




The real zombie game

Re: John Carmack on PhysX

Aug 21, 2009, 13:24

ASJD

He was asked what he thought about hardware physics at the show because he had commented about PhysX similarly before. Now instead of hardware physics on a specific card, it's hardware physics on a GPU. It's still dumb and I still agree with Carmack. Besides PhysX requiring so much power from a Nvidia card (have a GTX260 216), it scales horribly (my framerates are instantly equivalent to a 8800GTX running PhysX when I'm running hardware PhysX). It's not a great physics system (very poor for normal stuff like boxes or character clothing, only really impressive in terms of particle water or stuff like banners, wrapping, as seen in Mirror's Edge).

Re: John Carmack on PhysX

Aug 21, 2009, 13:18

LittleMe

The idea of advanced physics processing for games I think is a really great one, but imo, it needs to be a standard implemented by DirectX or by an open standard like OpenCL, so that the industry can run with it without a monopoly on it.




The real zombie game

Re: John Carmack on PhysX

Aug 21, 2009, 13:12

vermin

kind of a weird thing. no idea why he was asked that question, since hardware physx as in the separate card is dead and buried since quite a while ago and that seems to be the only thing he's talking about. the way that kind of physx is implemented now, albeit somewhat pointless, is good and seamless vs nvidia sli stuff. and the physx physics engine is very widely used and seems to be a good alternative to havok and whatever. physx as a company definitely wasn't without merit, even without the hardware bullshit.

34 Replies. 2 pages. Viewing page 2.
< Newer [ 1 2 ] Older >