Added: 2 years ago
From: BCEMCOCATb
Views: 40,860
Sort by time | Sort by thread (beta)

Link to this comment:

Share to:
see all

All Comments (53)

Sign In or Sign Up now to post a comment!
  • I like the idea of not needing to toss out a video card and use that for processing power... as to the original physx card... yep.. people got burned on that one.

    I also like the idea of project denver on an nvidia graphics card.. There are lots of possibilities without the bus in the way.. run the renderer on the graphics card.. almost no hit from physx.. physics/3d sound system.. better performence from GPGPU on smaller data chunks.

  • This guy can talk for hours.

  • @SamSpade2010 Long-therm hard ware is the way to go? I'll tell you this :) :

    Long-term, there will be only ONE piece of hardware on Earth! We will all use it.

    To be less abstract; software will run on as SMALL as possible hardware on clients, and big calculations ETC will go in the cloud, a.k.a. they will be done FOR you by a 3rd side.

  • Finally I get to hear the famous "mmmm" grunt that John makes. He's gotin' much better in the latest interviews I've seen. I can just imagine what he talked like when he was makin' Doom with Romero or even Commander Keen.

  • The guy knows what he is talking.

    If ww3 came, I would want to be somewhere near where this guy is!

  • The audience should be bending down & worshipping this man.

  • PhysX would be fine if they didn't use x87 instruction set (which runs horribly on x86 CPUs) and oh i don't know used something that x86 was good at (i.e. SSE etc etc)

  • @blake3443 he talks about HW accelerated PhysX.

    What you are talking is software PhysX and it is really good. As you said, it lacks optimization, but it is free for developers and it is pretty powerful in terms of easiness of use and what can be achieved.

  • PhysX was Owned

  • The only way hardware physics would have worked was to optimize it to run well on the onboard gpu's most gamers shut off when they upgrade to a dedicated card. I mean come on, the hardware is just sitting there doing nothing and those integrated 8300/9300 geforce chips were good enough to run hardware physx on early physx titles. Why not now?

  • Long-term, hardware is obviously the way to go because the algorithms are intrinsically parallel.

  • @SamSpade2010. Physics algorithms can use hardware, but it won't be a dedicated hardware nor close standard like PhysX that will gain any popularity. I don't see developers spending time and money to feed the growth of inhomogeneous platform in the future, as they try to develop game on the common-denominator configuration.

    Multicore CPUs with more advanced FPUs is the way to go with game physics.

    Ofc there will be games that support HW based physics, but only as a gimmick.

  • @BCEMCOCATb I don't understand how that's related to what's going on right now. IT seems obvious to me that when nVidia tells you to buy a physx card they tell you to do so, so you can run the current games out there that use physx. If something else is used in the future, cool, it's not going to matter because in that future none of our current hardware will be useful anymore. It's as if I regretted buying my SNES now that I have a DX11 card.

  • @elpolentatlp imagine if your SNES had only 5 games released total and they didn't much vary from your older NES. That is what Hardware Accelerated PhysX is about. Read the comments for further explanation.

  • he says i alot its so awesome :)

  • john effin carmack!! he's so anti-capitalizing on software. good man!  seriously!

  • I really do not understand why you guys hate physx so much. I personally love it and have enjoyed it in the few games that it has been in. I especially enjoyed it in the few UT3 levels that utilize it, Batman, and Mirrors edge. Physx on graphics cards allows for far more physics calculations at once in real time, I do not see how this is bad. It looks great, and is fun to watch and interact with objects with physx. What is so wrong about that?

  • What he says is that HW PhysX is useless, because it is proprietary and there are no gaming dev team that decided to use it, unless Nvidia pays them for the development.

    CPU is a perfect alternative for that:

    - it has 100% penetration

    - much easier to develop and port

    - results are the same with CPU, because HW PhysX takes significant amount of resources out of GPU. What you see in demos is not what you play.

  • @BCEMCOCATb

    The thing is that a GPU is just perfect for physics calculations, where CPU's are relatively extremely poor at it. Plus modern Graphics cards in the upper end of the spectrum are pretty overpowered and can handle both graphics and physx at the same time. Plus you can always add another GPU in to do physx. This makes things very complicated for the consumer though and most people will not take advantage of it. From my point of view I love physx and would like to see it pushed on gpu.

  • As you said, this makes things very complicated for the customer. But not only for the customer, but it also becomes ridiculously inconvenient platform for a developer too.

    Just look at physics engine in Crysis running on a single CPU core. 2010 Q2 6 core desktop CPUs coming. Quads are now cheaper than a $100.

    And again GPUs are not that powerful as marketing makes you think. These teraflops apper to be 10-20 times less when you apply a task like calculating physics.

  • I just read about the 6 core CPUs comming out and also about a 40 core cpu that intel is working on. This has me quite excited for the future. I do not question the power of parallel computing as much as you do though. I know for a fact that my GPU is far better at calculating physics than my quad core. How do I know? I have a quad core and an 8800gtx. I have seen how different physics programs work on my cpu, and nothing even touches physx... not even close. But with 40 very powerful cores...

  • What physics programs are you talking about?

    Please show me any existing game that can run better PhysX effects on your GPU than on your CPU. What I ask is a direct comparison, "here's a game running on 8800GTX with enabled HW PhysX performing XX FPS and here's the same game running with the same effects on Quad-core CPU with all cores loaded performing YY FPS".

  • There is no question. I do not mean to be rude, but you do not seem to understand what kind of calculations are being done to solve physics calculations. To be honest i cannot give you the actual code but i do understand that parallel processing is the most efficient way to do these calculations. My 8800gtx has around 130 (shader units) cores working on these problems. 4 cores can not even get close. It is not the power of the cores, but the number that is needed for physics processing.

  • Be attentive, I didn't say CPUs are faster than GPUs for physics effects processing. But running both a game with high settings and processing physics at the same time is another story.

    Look, you asked what's wrong about HW PhysX, I gave you more than enough arguments why it won't be utilized and can be called a dead-born technology.

    BTW I have my own highly efficient RTS path finding engine, so I know a little bit about "physics calculations" (;

    c the link in info 4 further reading, gl

  • Personally i have an 8800gtx for graphics and a 9600gso with 96 shaders dedicated to physx (and F@H). Most people will not have such a set up, i just hope that physx will become good enough to push set ups like mine. Also cards like the GTX 280 can handle graphics and physx just fine on its own. Cards are only getting faster with more shader units. And physx is being used increasingly. Ut III, Mirrors edge, The new batman game, Cryostasis, dark void, and ghost recon to name some.

  • @Iwillkillyou35 You go know that the 9600 is a rebrand of the 8800 right?

  • @BCEMCOCATb : Oh, and I would just like to add (didn't read below, so you may have already said this) that now we have compute shaders. We're basically going to be doing physics on Raedon HD5700+ series and Fermi without much problem. With CUDA and OpenCL support (even DirectX's less sophisticated compute shaders), that's going to be how we do physics. That's what Carmack meant when he said PhysX would be supplanted by more sophisticated "compute resources on there."

  • @BCEMCOCATb

    And has asbysmal perfotmance compared to a PPU or GPU.

  • Please try to be more specific. It is very hard to read others thoughts basing on phrase cut out from the context.

  • yesssssssssss go carmie, eff PhysX, dirty cheap trick

  • Unless we are talking about Tetris, proper 3D Accelerator must accelerate calculations with: (a) object shapes, (b) light reflections and (c) interactions. It depends on combinations of 6 elementary particles and 4 forces between them. Accelerators based on physical laws wont suffer complete revamp because Universe physical laws dont change. There is no other way to create convincing virtual reality than to utilize every known physical law of real World.

  • DirectX 11 and OpenCL will be the way to go with physics.

  • I won't bother with any hardware physics as long as its only accessible through one brand.

  • Physics is unesseccary bullshit, hes right!

    That was just a way for Nvidia, to get dumb people buy their cards.

  • Physics is unesseccary bullshit ? Are you kidding all of us or what? :P.Games will become more and more life-like and that includes physic calculations...There is no go-around-this cuz people won't bother with 2D tetris for a couple 'o generations.Physics do improve the visual appearance of games and they also add new ways of possible gameplay.The only "sucky" thing about "physics" in games right now is,either it's cpu-processed or "Nvidia only" - the idea itself is but pure epicness! :P

  • Not physics, but HW PhysX. Hope you can distinguish between two.

  • Physics do not improve gameplay what-so ever.

    There has not been more than a handful of games which required or used Physics to a degree that has effected gameplay dramatically.

    The day Nividia port PhysX to OpenCL, we will start to see innovative ideas to the mass's.

  • You've misunderstood him. He is saying a *dedicated physics hardware* solution isnt smart.

    nvidia acquired it when it wasn't really necessary, as dedicated physics cards were never going to really catch on, or at least have enough staying power over future integration. Ageia's program was merely a gimmick that, as John said, was started up looking to get acquired before technology left Ageia in history. sort of a waste of resources on nvidia's part.

  • niceguy391987, i think you missed the point of what he was saying. He's not saying physics in general is stupid but hardware accelerated physics is. He thinks physics should remain software rendered

  • WHats the difference between the two?

  • Physics is a general term for physics effects in games.

    PhysX is an API similar to Havok.

    HW PhysX is an HW accelerated execution of PhysX code on Nvidia GPUs or AGEIA cards.

  • Right on! We demand fair competition and price decrease.

  • if nvidia is smart PhysX should be made open-source and supported via OpenCL.

    as a middleware it's a bad idea.

    opensource the bastard and get on with it!

  • Oh, I see misinformation here.. what's the point to name this video "JC on hardware accelerated (GPU) PhysX" while he is talking about dedicated Ageia PhysX PPU stuff ?

  • Listen carefully, he was asked about "where HW PhysX vs software PhysX heading" in other words what's the future of HW PhysX.

    And he answers that the whole idea of HW accelerated physics was wrong from the start. Beginning from Ageia PPU, and currently embedded in Nvidia's GPUs.

  • Carmack didn't say a clear word about GPU PhysX, only about Ageia "that was a really bad idea.. whole idea for that was to do a startup to be acquired.. NVIDIA didn't pay a whole lot of money for them"

    Quoting Carmack from 2007 (google "John Carmack reckons PhysX is useless") - "Multiple CPU cores will be much more useful in general, said Carmack, but when GPUs finally get reasonably fine grained context switching and scheduling, some tasks will work well there" - he likes GPU PhysX idea ?

  • No, what he talks is a GPGPU with flexibly programmable pipeline, something like Larrabee implementing x86 architecture. Not about dedicated HW.

    BTW he says "and I HOPE NVIDIA didn't pay a whole lot of money for them", so don't cut the words out of context.

  • Ok, I see clear PhysX hater here.. pointless polemic isn't my goal. Just don't delete may comments, let the people decide. Thanks for good talk.

    "BTW he says "and I HOPE NVIDIA didn't pay a whole lot of money for them", so don't cut the words out of context. "

    You know that comments allows only 500 characters to be posted ? I cut that, to fit more valuable content.

    BTW, good nick) радикальные фанаты ATI ваше видео несомненно оценят)

  • I don't understand how can anybody hate a technology? So you're in love with HW PhysX or what?

    If your goal isn't pointless polemic, why did you post a comment, which I deleted for your own good, that says that JC is a "mediocre developer" and his opinion isn't valuable?

    Don't be an angry fanboy, judge sober. Good luck.

  • Трезвомыслящие люди это видео оценят. А любой фанатизм, по определению, является отсутствием здравого мышления.

  • No, John talks about GPU in general (like AMD, NVIDIA or Intel) when he says: "Multiple CPU cores will be much more useful in general, said Carmack, but when GPUs finally get reasonably fine grained context switching and scheduling, some tasks will work well there"

    Read this interesting interview from pcperspective "John Carmack on id Tech 6, Ray Tracing, Consoles, Physics and more"

  • it's not embedded in nvidia's gpu's.

    it's done using cuda, which is a nvidia general compute language.

    there is no special PhysX hardware in nvidia's gpu's, also because all the GPU's which also support were shipped before nvidia aquired it.

  • You misapprehended what I was saying. I was talking about technology, not about piece of silicon "soldered in" GPU.

    English is my third language :)

  • GPU-physics is Nvidia's new way of selling more cards, cause they can't seem to bullshit more people into buying SLI.

    Now they're telling you that with their GPU can compute everything a billion times faster than any CPU Intel can throw at them, but that's simply not true.

    And that's Carmack's point: Additional CPU-cores can be used for anything, including physics.

    Also notice how CPU load increases A LOT even using the gfx-card and how the PhysX-software runs on 1 CPU-core only.

  • Wait, so he expects things like fluid particle physics to be calculated on the CPU? I've seen plenty of particle physics demos, including ones that use PhysX (which performed far more smoothly), and it really doesn't seem like this is possible.

  • nvidia was pwned in this vid,lol :P

  • Nvidia realized the death of HW PhysX at least a year ago.

    It is all about marketing campaign making people think that running PhysX on GPU is a huge advantage of Nvidia cards, while actually it is totally useless.

  • Wow you are so wrong. But then again, you don't have the ability to see the upcoming PhysX things that I have. I wish I could say more...

  • Whoop de doo, there are plenty of tech demos out there that show what PhysX is capable of but only a handful of major games that actually use it. Even then you'd need a new Nvidia card to run it otherwise the game runs like crap.

    Havok FX is going to be OpenCL based so it will run on any GPU, when / if that comes out we may see a resurgence of HW based physics but PhysX will be long dead.

  • Havok FX(made in collaboration with NVIDIA and ATI) was killed by Intel when it acquired Havok, a few years ago, despite was almost complete.

    At this moment there's no available version of havok tath support GPU acceleration

Loading...
0 / 00Unsaved Playlist Return to active list
    1. Your queue is empty. Add videos to your queue using this button:
      or sign in to load a different list.
    Loading...Loading...Saving...
    • Clear all videos from this list
    • Learn more