MelonSplitter
[H]ard|Gawd
- Joined
- Aug 6, 2006
- Messages
- 1,142
I just played the Batman demo which looked pretty good on my 4870x2. When will ATI develop something like Physx.
Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
well I have a physx card and I have yet to buy a game that uses it so I consider it worthless. and with cpu physics coming along nicely I don't see the need. but as said open cl and DX11 will both allow it.
After finishing the demo, I have yet to be really blown away by this PhysX thing.
Let me bring up the new Ghostbusters game as an example: This was some of the most fun i've had with game physics since HL2 came out 5 years ago. Nearly every object aside from the walls themselves could be destroyed or manipulated. The game also runs impressively well on my aging rig so my CPU can't really be TOO strained by the physics.
Except the most powerful 8 thread/4 physical core CPU processes on the order of ~85 GFLOPS. A mid-range GPU processes between 10 times that and 13. Do the math regarding CPU physics and you'll see that for large interactions, you will need a co-processor assist.
PhysX is implemented using CUDA by nVidia. ATi had the chance and opportunity to do the same using their Stream GPGPU solution, but refused and are still stuck with CPU-only Havok.
DX11 will include no physics API, nor will OpenCL. OpenCL and DX11 (OpenGL GLSL equivalent) however can be used to implement a physics API like PhysX.
I think that at this point we'll see PhysX begin making Havok irrelevant as hardware accelerated physics will be the next (only?) big new thing in games. We have pushed graphical realism as far as we can so far, it's time to make the physics side catch up, and the only way to do that right now is through PhysX.
Indeed. If I recall correctly it was something around 5,000 physics objects/particles for a modern CPU and 500,000+ for a decent GPU in a test we ran with our in-house game engine (using PhysX) before we hit a wall. I completely disagree that PhysX slows down a game, even using just a CPU for PhysX is plenty fast. If the game is slow with PhysX-effects enabled, the implementation just sucks as we have encountered no such issues with our games.
To add to the statistics, current-gen GPUs are ~1 TF for high-end models (HD4870x2, GTX285), the upcoming cards (870, GT300) are expected to double that number. It'll be a long time off before CPUs will get anywhere near that number on integer or floating point arithmetic.
PhysX is implemented using CUDA by nVidia. ATi had the chance and opportunity to do the same using their Stream GPGPU solution, but refused and are still stuck with CPU-only Havok.
DX11 will include no physics API, nor will OpenCL. OpenCL and DX11 (OpenGL GLSL equivalent) however can be used to implement a physics API like PhysX.
I think that at this point we'll see PhysX begin making Havok irrelevant as hardware accelerated physics will be the next (only?) big new thing in games. We have pushed graphical realism as far as we can so far, it's time to make the physics side catch up, and the only way to do that right now is through PhysX.
I would say not too much yet. However, after just watching the youtube comparison of PhysX on and off, the steam and fog would make the game much more immersive for me.
Steam and fog have been in games for years. It's the fact that when you walk through it with Physx, it reacts. However, the dev for the batman game linked above decided to remove things completely if Physx is not enabled.
For that matter, it seems like physx is trying to be forced, by either using it, or not replacing it with a counterpart if not using physx.
Havok is an AMD/ATi-licensed technology, owned by Intel. Havok, the company, was bought by Intel a while ago. Havok (the company) was developing Havok FX, but after they got bought out by Intel all work on this API stalled. A version of Havok (API) has been shown to work on GPU by AMD, but it was only a proof of concept, and there are AFAIK no plans to develop it commercially.Havok is not an ATI owned technology, its owned by Havok (the company) and Intel, and they were/are developing a GPGPU variant which will run nVidia or ATI GPUs (and ATI/Intel are working on an OpenCL/Stream SDK variant to utilize Havok SDK as well (I see no reason for them to shaft nVidia, unless nVidia themselves simply say no))
Correct. Havok is the older physics API of the two, first becoming popular in 2006 with Half Life 2 and has been used in many games. PhysX didn't enjoy popularity until 2008 when the company which created PhysX got bought by nVidia. Since then there has been a huge uptake of PhysX, with it becoming available for SDKs on all current consoles and huge publishers/developers like EA licensing it as well. A major advantage PhysX has over Havok especially for smaller developers like my company is that using Havok requires a license, which isn't cheap. PhysX can be used commercially without any charge. This is why I think that PhysX will enjoy more growth than Havok the coming years.ATI is not "stuck" with anything, as a physics API is something a developer chooses to utilize, only a few games actually use PhysX, even less on GPU, and there are MANY more games with Havok support as opposed to PhysX (so I'm calling fud on your assertions), because any of these games will run fine with nVidia or ATI hardware, basically getting the same performance
my thought on PhysX is that its the last in a line of retarded dot-com ideas that sadly made enough noise to get pushed into production, just like A3D, EAX, GLiDE, and so on, a highly specialized proprietary API will always run "better" on custom fitted hardware, and it looks great in reviews, but has little to no general appeal or application, before you argue with that, none of my examples are still considered "in development" or "active"
You don't seem to realize that it is incredibly hard to 'scale' the amount of physics in a game dynamically. You'd have to swap out physics-enabled objects for static ones or ones with a less fine mesh or so, or remove them all together. The latter option is far easier and thus is chosen because the former option would add months to the development time.well if you spent thousands to millions of dollars on licencing fees paid out to a company that spent millions of dollars to buy up the technology in a bid to own a larger stake of computer gaming, only to learn that it isn't really going to pay out, what do you think you'd do? be reasonable about it? or try to force it down everyone's throats with an iron fist?
here is a comparison for Cryostasis http://www.youtube.com/watch?v=kufpPwho9Ec&NR=1
this game of course is very sluggish on highest settings with advanced Physx effects turned on.
PhysX is implemented using CUDA by nVidia. ATi had the chance and opportunity to do the same using their Stream GPGPU solution, but refused and are still stuck with CPU-only Havok.
DX11 will include no physics API, nor will OpenCL. OpenCL and DX11 (OpenGL GLSL equivalent) however can be used to implement a physics API like PhysX.
I think that at this point we'll see PhysX begin making Havok irrelevant as hardware accelerated physics will be the next (only?) big new thing in games. We have pushed graphical realism as far as we can so far, it's time to make the physics side catch up, and the only way to do that right now is through PhysX.
I would say not too much yet. However, after just watching the youtube comparison of PhysX on and off, the steam and fog would make the game much more immersive for me.
The problem for Physx is that developers are not intergrating it enough. it is not being used to to the extent that it has to be used in the game. Developers are doing that with havoc.![]()
Your imagination is wrong. Interactive steam would hit a CPU very hard. A lot of people make the mistake of equating "faked" effects with the stuff PhysX is doing. For example, while the floor tile destruction is pretty lame visually it's fully dynamic and you can interact with the broken pieces of tile (break them into smaller pieces). In the past if you did that you'd get a "broken tile" texture decal that was flat and static.
Indeed. If I recall correctly it was something around 5,000 physics objects/particles for a modern CPU and 500,000+ for a decent GPU in a test we ran with our in-house game engine (using PhysX) before we hit a wall. I completely disagree that PhysX slows down a game, even using just a CPU for PhysX is plenty fast. If the game is slow with PhysX-effects enabled, the implementation just sucks as we have encountered no such issues with our games.
To add to the statistics, current-gen GPUs are ~1 TF for high-end models (HD4870x2, GTX285), the upcoming cards (870, GT300) are expected to double that number. It'll be a long time off before CPUs will get anywhere near that number on integer or floating point arithmetic.
Honesty, I have to say that PhysX is really cool, but it only adds about 1% of content to any game and isn't necessary to run on a GPU or PPU specifically. For the cost/performance ratio to specifically buy a GPU JUST for this isn't really required yet. A CPU can do just fine as long as enough cores are available.
Steam and fog have been in games for years. It's the fact that when you walk through it with Physx, it reacts. However, the dev for the batman game linked above decided to remove things completely if Physx is not enabled.
For that matter, it seems like physx is trying to be forced, by either using it, or not replacing it with a counterpart if not using physx.
LOL
OpenCL had adopted Havok's API (for GPU and CPU). It will load balance varying on the hardware detected using the balance which provides the best performance.
PhysX never really got a start and it seems it will die out in the next 1-3 years.
That is awesome.
Honesty, I have to say that PhysX is really cool, but it only adds about 1% of content to any game and isn't necessary to run on a GPU or PPU specifically. For the cost/performance ratio to specifically buy a GPU JUST for this isn't really required yet. A CPU can do just fine as long as enough cores are available.
If a spare GPU is available, I totally think it's worth it and if it REALLY makes a difference in a game.
I'm just saying, for the cost/performance ratio, it's not worth it.
But yeah, the CPU could never do what a GPU could do in terms of raw performance.
In the end, having physics calculated on the GPU means that graphics and physics are directly competing with one another... which is a messy situation. If PhysX ever becomes a "big thing" I fear we'll all need SLI setups to handle it without wrecking graphical performance. Which would be great news for NVIDIA, of course!
Aside from that, though, I don't like the situation we're in at the moment where NVIDIA cards do PhysX acceleration and ATI cards don't. There needs to be univeral support for physics acceleration. If there isn't universal support, I'm tempted to say that I'd rather not have physics acceleration at all (though I know that's a very argumentative thing to say!). Basically I don't think having proprietary physics acceleration is a much better situation than having no physics acceleration. With things the way they are at the moment, something has to give way... and I don't think ATI getting muscled out of the market would be a better outcome than PhysX dying out.