GPU Physics Initiative Dead?

[Vipasnipa]

DAAMIT appears to be canning its plans for GPU physics, after all the talk of the technology being yet another big application for graphics cards.

At Computex this year, a big focus was put on the ability to run multiple ATI cards and have some compute visuals and some physics - making use of all that horsepower for more than just extra layers of anti-aliasing.

Unfortunately for AMD (and Nvidia, for that matter), all this was being done with a variation of the Havok physics engine called Havok FX. That was, of course, until Intel bought Havok in September.

So what do you think, are we still going to see GPU Physics from Big Red and Big Green? What does this mean for Physx (which a lot of you seem to despise for no other reason then that it costs money), or for hardware-based physics in games in general?

[ViperJohn]

So what do you think, are we still going to see GPU Physics from Big Red and Big Green? What does this mean for Physx (which a lot of you seem to despise for no other reason then that it costs money), or for hardware-based physics in games in general?

It's not that people hate Ageia PhysX so much. It's the game makers are not supporting it which makes it useless.

Viper

[Omsion]

I don't see such a huge need for hardware physics with the direction of current CPUs (aka many many cores). Also, games have no real need of perfectly realistic physics simulations - realistic enough is good enough, so the benefit of highly acccurate hardware physics isn't really needed.

[WiglyWorm]

I kind of don't get the whole physics computations on video cards thing anyway. If the bottleneck for modern gaming is the GPU, when why should we put the physics processing load on the GPU as well?

Why not use the CPU?

If I had to guess, that's exactly what intel is planning to do. Put it on the CPU, or build it in to intel motherboards.

This might help Agea out, but only if games start supporting PPUs. I'm thinking about picking one up for UT3, but it's a way down the road back burner thing, because there's just no good reason as of yet.

That's the problem with trying to invent a market for something, which is exactly what Agea is trying to do. Unless they can convince some big name games to put something worthwhile in to a game that only PhysX can do, it's just not going to be a must have. Why should I pay for another peripheral just so I can have a couple more particles on the screen or an extra wavey flag?

[Mr.Guvernment]

I don't see such a huge need for hardware physics with the direction of current CPUs (aka many many cores). Also, games have no real need of perfectly realistic physics simulations - realistic enough is good enough, so the benefit of highly acccurate hardware physics isn't really needed.


"isnt needed" is your own personal opinion, some people like it to be a real as possibe.

[Omsion]

"isnt needed" is your own personal opinion, some people like it to be a real as possibe.Can you tell real from "close to real"? We've really seen almost no benefit in games that use the Aegia card.
Does a few pieces of debris landing +- a few cm/inches away from where a fully realistic simulation would have them land really affect the gameplay/graphics that much?

Take CoH, where physics actually play quite a role (fully destructable enviornments affecting gameplay ftw). That's an example of "good enough" - the major pieces of physics (the ones that effect gameplay) are taken care of, and the minor stuff is eye-candied.

[deathman20]

Reason behind this is because Havok wasn't going ot release support for onboard GPU phyiscs as well since well Intel now owns Havok.

CPU's are powerful as is doing quiet a bit of phyisics work as well and surely the mroe CPU's the more can be dedicated to Physics.

[WildMonkey]

Physics accelerators will probably never have a future.

Why put effort in coding effects and drivers, etc.. for a specific card (not to mention the extra cost) for something that can be done by an extra CPU?

[EmAn]

Physics accelerators will probably never have a future.

Why put effort in coding effects and drivers, etc.. for a specific card (not to mention the extra cost) for something that can be done by an extra CPU?


isnt that pretty much all it is?

[simba1621]

Physics accelerators will probably never have a future.


that's a scary thing to say. you never know what's around the corner in the electronics world.

[Mr.Guvernment]

i am sure physics in some form are already highly in use in things like training for say army simulations, who knows what other massive corporations are already using this stuff for things.

[dipspit]

i am sure physics in some form are already highly in use in things like training for say army simulations, who knows what other massive corporations are already using this stuff for things.

Not a chance. Real life simulation physics equations cannot be handled by a $175 add on card, or even 10 of them. They use super computers to compute live physics scenarios, these things lack the horsepower to have any application in realistic physics models. They just write software to allow their CPUs to crunch the numbers.

[ryanmartini]

heres some agaea demos. I kinda like the realistic touches it puts into games potentially.

[Immortal_Hero]

@ryanmartini - where are the links?

The current GPU and CPU makers have the ability to add specific physics processing to their products. They choose not to because of the low support in software, the extra cost to implement, the low return on investment (little change in games that do support it), and the small customer base. There is just no "bang-for-buck" in it for the major companies mostly becasue game makers are not programming for it. There will have to be a universal solution that is easy to implment before it will really catch on.

There is also a big negative for gamers. Adding the physics card slows game play down which is bad. Gamers could and do spend that extra $200 or a better CPU/GPU b/c it is a smarter investment.

Stand alone Physics cards fall in a very nitch market because of limited support and high cost. Until they can come up with something to appeal to the masses they will never "blow up."

For Ageia to keep going they need to get out there and sell their products to game makers. Then we need to see a big improvement is many games when the Ageia card is in the system. At this point they have a market and can sell.

IMO is it more cost effective for game makers to use the universal solution and offload physics to the CPU. We are looking at more and more cores setting there and it won't be long before game makers start to dip into them.

[Vipasnipa]

I was under the impression that the whole point of the PPU was its parallel processing architecture, much like a GPU. To expand the argument; now that we have an abundance of processing cores in our CPUs, why do we still have GPUs? Why not do all the graphics calculating on those extra cores, which seems to be the main argument against PPUs? We all know that answer to this: a general CPU cannot match the GPU when it comes to doing massive numbers of parallel math calculations necessary in graphics.

[Immortal_Hero]

I think we are heading to the point where we simply have CPU's. There are multiple sockets on a board and we can designate what each CPU will do. I think AMD/ATI will play a big role in this. This is just my opinion.

You are right with the idea of parallel processing with a PPU but I don't think it does a good job of parallel processing. With parallel processing there will be a certain amount of overhead even if done right b/c data has to get to and from the unit doing the processing.

[WildMonkey]

I think we are heading to the point where we simply have CPU's. There are multiple sockets on a board and we can designate what each CPU will do. I think AMD/ATI will play a big role in this. This is just my opinion.

You are right with the idea of parallel processing with a PPU but I don't think it does a good job of parallel processing. With parallel processing there will be a certain amount of overhead even if done right b/c data has to get to and from the unit doing the processing.

It's interesting that you bring this up.

In the olden days (10 years or so) GPUs were needed as CPUs were not powerful enough to run the simplest of games, thus started the video card industry.

I see no reason why a chip or CPU in a motherboard couldn't eventually be able to replace the GPU card.

The questions are:
1. Is it possible at this time to do a mobo with an integrated GPU that competes with current cards?

2. Are mobo producers just so used to the idea of video cards they don't really see a need for it, cost or effort wise?

[Immortal_Hero]

Here is my vision. I see us having 4 socket boards and CPU makers selling specialized CPUs to drop in the sockets. Say you are a gamer you drop in 2 CPU's optimized for crunching numbers and 2 CPU's for GFX. Say you are a server you drop in 4 CPU's optimized for virturaliation, number crunching, etc. I think we are going to see more and more integrated into the mainboard over the next 10 or so years.

To answer you questions AMD/ATI has everything in their own shop so they can do whatever. They can make the CPU, make the chipset, make the GPU and optimize it all to work on a single board.

[thideras]

It's interesting that you bring this up.

In the olden days (10 years or so) GPUs were needed as CPUs were not powerful enough to run the simplest of games, thus started the video card industry.

I see no reason why a chip or CPU in a motherboard couldn't eventually be able to replace the GPU card.

The questions are:
1. Is it possible at this time to do a mobo with an integrated GPU that competes with current cards?

2. Are mobo producers just so used to the idea of video cards they don't really see a need for it, cost or effort wise?...and modern processor still lack the power to run 3d graphics at a decent framerate.

There is a HUGE difference between a GPU and CPU.

A CPU is a "do everything" type of processor. It gives up speed (optimizations) for being able to calculate all types of data.

A GPU is a dedicated floating point processor. It gives up versatility to gain speed (tons of speed).

The CPU is very much capable of doing floating point, but no where near as fast as the GPU.


Try this, run a program that will use software (using the CPU) to render graphics, tell me how it runs. I'd be surprised if you got 10fps average in a decently new game. Compare that with a normal card that will get at least 5x that. Why is it slower? The CPU is not optimized to run floating point operations and therefor is slower.

[Neural Net]

The CPU and GPU is a good combination (as two separate products) - it works and allows for more custimisation in terms what you demand from your machine. Offering the same versatility with an AIO solution would just lead to an unnecessary and ridiculously complicated product lineup, as if the naming schemes of CPUs and GPUs isn't silly enough. It's not going to happen, well, maybe when quantum computers become mainstream. :p

[Immortal_Hero]

I don't see it being a real solution until we saturate our extrnal buses (PCI-E, AGP, PCI, or whatever new they come up with).

[dipspit]

I've always been a fan of specialization for the best performance, whatever the application may be (computers, cars, business models, lab equipment). Keeping this in mind I would much rather see individual companies specializing in specific components (as we have now), than having a 'jack of all trades' type setup. I don't want to see Intel making GPUs (high-end) or see nvidia making CPUs. Specialization prevents the manufacturers from skimping in one area and allows total focus and dedication to their single product.

As far as quality goes, it just seems like specialization always improves overall product quality. For the consumer's sake, I hope it stays this way.

[>HyperlogiK<]

...and modern processor still lack the power to run 3d graphics at a decent framerate.

There is a HUGE difference between a GPU and CPU.

A CPU is a "do everything" type of processor. It gives up speed (optimizations) for being able to calculate all types of data.

A GPU is a dedicated floating point processor. It gives up versatility to gain speed (tons of speed).

The CPU is very much capable of doing floating point, but no where near as fast as the GPU.

Try this, run a program that will use software (using the CPU) to render graphics, tell me how it runs. I'd be surprised if you got 10fps average in a decently new game. Compare that with a normal card that will get at least 5x that. Why is it slower? The CPU is not optimized to run floating point operations and therefor is slower.

Well firstly the Pixomatic renderer has always run quite well and I think it's current lack of features is more the result of developers lacking interest than anything else. It was used in UT2004 and would run quite acceptably at 1024x768 on a mid range AXP or Northwood system.

Secondly the raytraced version of Quake 4 has been running fine and dandy (90FPS at 1280x720) on dual Clovertown Xeons. I realise that Quake 4 is a little past cutting edge and few people have 8 core machines, but I bet that a single Penryn quad would run it OK.

Also, unlike multiple GPU solutions for raster graphics, raytracing is supposed to scale almost perfectly the more cores you run it on.

With the general purpose processing power of computers increasing so quickly and not much in the way of software to take advantage of it I think there is something to the claims that we may see raster graphics hardware replaced by raytracing algorithms implemented in software.

[thideras]

Well firstly the Pixomatic renderer has always run quite well and I think it's current lack of features is more the result of developers lacking interest than anything else. It was used in UT2004 and would run quite acceptably at 1024x768 on a mid range AXP or Northwood system.

Secondly the raytraced version of Quake 4 has been running fine and dandy (90FPS at 1280x720) on dual Clovertown Xeons. I realise that Quake 4 is a little past cutting edge and few people have 8 core machines, but I bet that a single Penryn quad would run it OK.

Also, unlike multiple GPU solutions for raster graphics, raytracing is supposed to scale almost perfectly the more cores you run it on.

With the general purpose processing power of computers increasing so quickly and not much in the way of software to take advantage of it I think there is something to the claims that we may see raster graphics hardware replaced by raytracing algorithms implemented in software.Well, that still kind of proves my point. You said it took dual quads to run the game at 90fps at 1280*720. What kind of graphics card (single card) would it take to run that? Not much....

[>HyperlogiK<]

Sorry I should have been clearer and I think you may have missed my point. If you want to see what a half decent software raster renderer can do, look at Pixomatic. It runs acceptably on older hardware and there is no reason I can think of why an updated version sporting DX8.1 or DX9 features couldn't be developed for more recent games.

Software raster rendering is more than dooable even with fairly modest computers.

The stuff about the raytraced Q4 was a different point altogether (even if I didn't make that especially clear). I was pointing out that some pundits believe that raytracing may shortly replace raster graphics, and since general purpose CPUs seem so good at raytracing it might become more economical for PC gamers to buy high end CPUs than mid range parts and discreet 3D cards.

[Prisoner1138]

So what do you think, are we still going to see GPU Physics from Big Red and Big Green? What does this mean for Physx (which a lot of you seem to despise for no other reason then that it costs money), or for hardware-based physics in games in general?

I don't despise Physx because it costs money, I just find it totally useless.

In the few games that actually support it, it just adds more realistic movement when kicking boxes around(or other objects) without actually affecting or adding to gameplay. Strike 1

Also, there's the reports I've seen involving performance hits of a few FPS, but without adding anything to my gaming experience. Strike 2

Plus, especially in the realm of multiplayer games, even if developers wanted to add support for the physx cards and have it affect gameplay, they couldn't do it unless everyone had it, because to give a significant advantage to the gamer who has the physx card over the gamer who doesn't, is not going to win that developer any fans. Strike 3

So that's 3 significant issues right there preventing me from bothering to buy a physx card, and developers from doing much of anything with them.

[Maverick0984]

I don't despise Physx because it costs money, I just find it totally useless.

In the few games that actually support it, it just adds more realistic movement when kicking boxes around(or other objects) without actually affecting or adding to gameplay. Strike 1

Also, there's the reports I've seen involving performance hits of a few FPS, but without adding anything to my gaming experience. Strike 2

Plus, especially in the realm of multiplayer games, even if developers wanted to add support for the physx cards and have it affect gameplay, they couldn't do it unless everyone had it, because to give a significant advantage to the gamer who has the physx card over the gamer who doesn't, is not going to win that developer any fans. Strike 3

So that's 3 significant issues right there preventing me from bothering to buy a physx card, and developers from doing much of anything with them.

Well, I don't have a PhysX card and don't plan on getting one anytime soon, but I find your logic against them a bit flawed. They can be used for other things then "box motion" when you kick them. Perhaps Particle Dynamics during the rendering of an explosion. That being the little pieces of debris and where they fly to. Or any other piece of environment. Perhaps the splitting of wood when you shoot through it? Something as simple as the human hair has always been somewhat of a Holy Grail for game developers because of it's complexity in depiction since every head of hair has hundreds of thousands of hairs (probably? :p). Something like the inclusion of a PhysX card to handle something like this would be a huge improvement and allow for more "realism" to be added to process data, and not just polygons.

Your issue with it not adding to the game experience is completely relative to your personal opinion on what you want in a game. How many games nowadays are pushing for better/high quality graphics? (Rhetorical: Answer - All of them :beer:). According to you, you don't care about that type of stuff. The ever increasing quality of graphics is exactly the same thing with the ever increasing ability for movement in objects. MANY people feel that graphics and particle/environmental interaction is in fact, a huge, if not major part, of their gaming experience. Just because you don't feel that way, doesn't make it a "reason" to discontinue the development of the tech :p.

Your entire 3rd reason is just silly TBH. PhysX isn't meant to add something you can "only do with the tech" like picking up a box you wouldn't normally be able to pick up to complete an objective. (Dumbed down example I know, but still relevant). It is meant to enhance the visual experience.

I mean this with the utmost respect, but it seems as though you are on the side of "Performance" when it comes to gaming, and you should be aware that there is an entire other group that wants to be catered to "Quality". There is nothing wrong with wanting max performance at the cost of quality. Just your personal preference.

Sidenote: Could developers start using the CPU more efficiently when they have 4 cores to work with to accomplish some of these tasks? Sure, and they will. However, there is only so much you can do with straight CPU cycles. In either case, that is a bit of a tangent and perhaps a topic for a different discussion entirely.

[thideras]

Sorry I should have been clearer and I think you may have missed my point. If you want to see what a half decent software raster renderer can do, look at Pixomatic. It runs acceptably on older hardware and there is no reason I can think of why an updated version sporting DX8.1 or DX9 features couldn't be developed for more recent games.

Software raster rendering is more than dooable even with fairly modest computers.

The stuff about the raytraced Q4 was a different point altogether (even if I didn't make that especially clear). I was pointing out that some pundits believe that raytracing may shortly replace raster graphics, and since general purpose CPUs seem so good at raytracing it might become more economical for PC gamers to buy high end CPUs than mid range parts and discreet 3D cards.Ah, sorry, I did miss that ;)

I am also talking about efficiency though. If it took dual quads running full out to do 90fps (which is very respectable) and a decent graphics card could do that same, then it kind of defeats the purpose.

I am far from saying ditch software rendering. I am saying though, from an efficiency standpoint, software rendering (at this point) is not really close to what a dedicated GPU can do ;)

[Prisoner1138]

Well, I don't have a PhysX card and don't plan on getting one anytime soon, but I find your logic against them a bit flawed. They can be used for other things then "box motion" when you kick them. Perhaps Particle Dynamics during the rendering of an explosion. That being the little pieces of debris and where they fly to. Or any other piece of environment. Perhaps the splitting of wood when you shoot through it? Something as simple as the human hair has always been somewhat of a Holy Grail for game developers because of it's complexity in depiction since every head of hair has hundreds of thousands of hairs (probably? :p). Something like the inclusion of a PhysX card to handle something like this would be a huge improvement and allow for more "realism" to be added to process data, and not just polygons.

Your issue with it not adding to the game experience is completely relative to your personal opinion on what you want in a game. How many games nowadays are pushing for better/high quality graphics? (Rhetorical: Answer - All of them :beer:). According to you, you don't care about that type of stuff. The ever increasing quality of graphics is exactly the same thing with the ever increasing ability for movement in objects. MANY people feel that graphics and particle/environmental interaction is in fact, a huge, if not major part, of their gaming experience. Just because you don't feel that way, doesn't make it a "reason" to discontinue the development of the tech :p.

Your entire 3rd reason is just silly TBH. PhysX isn't meant to add something you can "only do with the tech" like picking up a box you wouldn't normally be able to pick up to complete an objective. (Dumbed down example I know, but still relevant). It is meant to enhance the visual experience.

I mean this with the utmost respect, but it seems as though you are on the side of "Performance" when it comes to gaming, and you should be aware that there is an entire other group that wants to be catered to "Quality". There is nothing wrong with wanting max performance at the cost of quality. Just your personal preference.

Sidenote: Could developers start using the CPU more efficiently when they have 4 cores to work with to accomplish some of these tasks? Sure, and they will. However, there is only so much you can do with straight CPU cycles. In either case, that is a bit of a tangent and perhaps a topic for a different discussion entirely.

It's not flawed at all. And yes, I'm well aware that you can do more with it than move some boxes around, but it's still the same thing. "Oh hey, those bricks fell down so realistically when I blew up that wall in such and such game", whoopdeedoo.

And actually, with decent physics processing, there's a LOT that can be added to the game. For example, say that brick wall I just mentioned. You're playing a first person shooter, some guys are behind some kinda bunker thing, and you fire some uber missile that blows up the wall behind them sending debris flying. I think it'd be GREAT if that realistically moving debris did say, damage to those guys you couldn't shoot directly at.

How about something as simple as realistic grenade bounces? Bullet ricochets? Those are things that could make use of something like the physx processor in a game, and actually make an impact on gameplay without just being graphical fluff like debris and ragdoll corpses. But what if you and I are playing a game, I have the physx, you don't. Wouldn't that suck if I could ricochet bullets or something else and you can't?

That's precisely why game developers can't add something like that using the physx card. They'd end up excluding a LOT of folks who simply don't have the card, potentially leaving them at a disadvantage in whatever game they were playing vs the folks that did have it.

And no, I'm not the kind of guy who cranks all the settings down all the time just for the sake of having a high framerate.

[Maverick0984]

It's not flawed at all. And yes, I'm well aware that you can do more with it than move some boxes around, but it's still the same thing. "Oh hey, those bricks fell down so realistically when I blew up that wall in such and such game", whoopdeedoo.

And actually, with decent physics processing, there's a LOT that can be added to the game. For example, say that brick wall I just mentioned. You're playing a first person shooter, some guys are behind some kinda bunker thing, and you fire some uber missile that blows up the wall behind them sending debris flying. I think it'd be GREAT if that realistically moving debris did say, damage to those guys you couldn't shoot directly at.

How about something as simple as realistic grenade bounces? Bullet ricochets? Those are things that could make use of something like the physx processor in a game, and actually make an impact on gameplay without just being graphical fluff like debris and ragdoll corpses. But what if you and I are playing a game, I have the physx, you don't. Wouldn't that suck if I could ricochet bullets or something else and you can't?

That's precisely why game developers can't add something like that using the physx card. They'd end up excluding a LOT of folks who simply don't have the card, potentially leaving them at a disadvantage in whatever game they were playing vs the folks that did have it.

And no, I'm not the kind of guy who cranks all the settings down all the time just for the sake of having a high framerate.

You kind of missed most of the point to what I said. That being, that it has the potential to include graphical improvements that make it "look" and "feel" better that otherwise would not have been obtainable. That is the part that is flawed in your argument. You assume that something merely looking better isn't a welcome improvement to any game. When in fact, is one of the major driving forces for the industry right now.

As for "damaging" debris. Yea, that'd be great. But the situation you create where hardware must be available to take advantage of it isn't a very real option, which it seems like you are aware of. So the made up analogy doesn't exactly work. It wouldn't be that hard to place that added stress onto whatever hardware components you do have, and the PhysX tech would just help it run "better" with higher framerates, just like everything else...I just don't see how this is any different than any other optional component in a system. It would provide improvements on what you already have. Whether they are graphical or realism inclined isn't really an issue.

Of course a developer won't force the user to buy an extra card, but who says they have to do that?

[Prisoner1138]

You kind of missed most of the point to what I said. That being, that it has the potential to include graphical improvements that make it "look" and "feel" better that otherwise would not have been obtainable. That is the part that is flawed in your argument. You assume that something merely looking better isn't a welcome improvement to any game. When in fact, is one of the major driving forces for the industry right now.

As for "damaging" debris. Yea, that'd be great. But the situation you create where hardware must be available to take advantage of it isn't a very real option, which it seems like you are aware of. So the made up analogy doesn't exactly work. It wouldn't be that hard to place that added stress onto whatever hardware components you do have, and the PhysX tech would just help it run "better" with higher framerates, just like everything else...I just don't see how this is any different than any other optional component in a system. It would provide improvements on what you already have. Whether they are graphical or realism inclined isn't really an issue.

Of course a developer won't force the user to buy an extra card, but who says they have to do that?

Ok, so as I said, you seem to want useless graphical fluff that doesn't add to the game, I want physics processing that adds to the game. Since the physx card can't do anything but add fluff, because the game developers can't add the physics "stuff" that does something using the physx cards.

[Maverick0984]

Ok, so as I said, you seem to want useless graphical fluff that doesn't add to the game, I want physics processing that adds to the game. Since the physx card can't do anything but add fluff, because the game developers can't add the physics "stuff" that does something using the physx cards.

lol, again, there are alot of people that feel graphical improvements are not "useless".

I was merely trying to show you that you and your opinion do not represent the everyone's and many people would love the card if ALL it did was add realism through increased graphical enhancements and the overall "feel" of the game.

[Karadhas]

Lets look at it this way, my pentium 4 can run Cell Factor with soft cloth and everything. It's nearly a slideshow but it still runs. I doubt a c2q would even care in all but the most extreme situations.

[Prisoner1138]

lol, again, there are alot of people that feel graphical improvements are not "useless".

I was merely trying to show you that you and your opinion do not represent the everyone's and many people would love the card if ALL it did was add realism through increased graphical enhancements and the overall "feel" of the game.

Yes, but that doesn't cover the thread topic, which mentioned the GPU physics. If nvidia and ati got off their butts and pushed GPU physics processing, game developers could include that, more users would have it, and then they could start adding physics effects that aren't just graphical fluff.

I'm pretty sure folks would rather see THAT, than more physx support.

[Maverick0984]

Yes, but that doesn't cover the thread topic, which mentioned the GPU physics. If nvidia and ati got off their butts and pushed GPU physics processing, game developers could include that, more users would have it, and then they could start adding physics effects that aren't just graphical fluff.

I'm pretty sure folks would rather see THAT, than more physx support.

Yea, I'm guilty of merely skimming most of the first few posts :-/ after the OP and then yours which was the last one at the time. :bang head

[Prisoner1138]

So what you're saying is, my posts are so awesome you forget the rest of the thread? Sweet.

Now imagine how much more awesome my post would have been if you had a physx card installed and could see the tornado dust cloud blowing around boxes :p

[Maverick0984]

So what you're saying is, my posts are so awesome you forget the rest of the thread? Sweet.

Now imagine how much more awesome my post would have been if you had a physx card installed and could see the tornado dust cloud blowing around boxes :p

Well, it started out as merely an aside to your post :p Initially there wasn't a reason to read the rest ;)

On Topic: So how about those PhysX cards? eh?

[Mr Alpha]

Warmonger: Operation Downtown Destruction (http://www.fileplanet.com/182564/180000/fileinfo/Warmonger-Gold-Install-(Free-Game)) just went gold. It has some nice destructible houses and stuff. It also says that a PhysX card is only "strongly recommended", so one should be able to run it completely on a CPU. Could somebody with a nice quad-core download it and see if it runs fine?

I believe one of the big issues with doing both physics and graphics on the CPU is memory bandwidth.

Also, I'm still hoping for some really good AI, and that won't happen if you guys move all the physics and graphics processing onto the CPU.

There was a lot of talk in the early reviews about how enabling PhysX in GRAW reduced the performance and this, of course, turned all enthusiasts against it so that nobody read the follow up articles which showed that a few driver and game updates later the PhysX card actually sped up GRAW.

With the Unreal Engine 3 using PhysX SDK and Ageia cutting the PhysX card to $99 this is probably the last chance for it to take of. Or it'll just die or be bought by somebody.

[Maverick0984]

Warmonger: Operation Downtown Destruction (http://www.fileplanet.com/182564/180000/fileinfo/Warmonger-Gold-Install-(Free-Game)) just went gold. It has some nice destructible houses and stuff. It also says that a PhysX card is only "strongly recommended", so one should be able to run it completely on a CPU. Could somebody with a nice quad-core download it and see if it runs fine?

I believe one of the big issues with doing both physics and graphics on the CPU is memory bandwidth.

Also, I'm still hoping for some really good AI, and that won't happen if you guys move all the physics and graphics processing onto the CPU.

There was a lot of talk in the early reviews about how enabling PhysX in GRAW reduced the performance and this, of course, turned all enthusiasts against it so that nobody read the follow up articles which showed that a few driver and game updates later the PhysX card actually sped up GRAW.

With the Unreal Engine 3 using PhysX SDK and Ageia cutting the PhysX card to $99 this is probably the last chance for it to take of. Or it'll just die or be bought by somebody.

I can install it, but I won't be home for 9 hours or so :-/

[BobcatDan]

Downloading now...

[BobcatDan]

OK, well I wasn't exactly sure how to test it, since I'm not sure if one of the features affects physics. So what I did was to turn the res down to 720x480 (lowest 16x10 res) and leave everything else as high as possible, this should prevent the vid card from becoming the bottle neck. Running the game I got as low as 10 FPS but it seemed to hover in the 20fps rage (going as high as 40s and as low as mid teens). When I hosted a multiplayer game, it hovered around 10fps and dipped as low as 6, I think the added stress of controlling 29 bots complicated things. The game did load all four cores, though not full (most around ~80% and one at ~60%).
Dan

[Maverick0984]

OK, well I wasn't exactly sure how to test it, since I'm not sure if one of the features affects physics. So what I did was to turn the res down to 720x480 (lowest 16x10 res) and leave everything else as high as possible, this should prevent the vid card from becoming the bottle neck. Running the game I got as low as 10 FPS but it seemed to hover in the 20fps rage (going as high as 40s and as low as mid teens). When I hosted a multiplayer game, it hovered around 10fps and dipped as low as 6, I think the added stress of controlling 29 bots complicated things. The game did load all four cores, though not full (most around ~80% and one at ~60%).
Dan

Wow really? and at that low of a res? Pretty poor performance if you ask me :-/

[BobcatDan]

Well, I decided to drop the CPU clocks down to stock 2.4 (1.2GHz underclock from my everyday) to see what would happen. There were two extended spots where I was at a fairly stead 6 fps! Some spots of ~10 fps, and actually a few spots of 20-30 fps. So it seems that this game is very CPU limited (with the lack of an AGEIA card at least). BTW this was playing the tutorial, as was the first part of my last post. Also, the CPUs were more like 95% loaded as opposed to ~80% as with the higher CPU clock...
Dan

[BobcatDan]

One last test, CPU back to my normal overclock and GPU to stock clocks (513 core 900 mem). Ran the tutorial again, virtually no change from the first time. Around 20 fps, dipping into the teens and spiking normally no higher than low 30's. The CPUs loaded to about 80-90%.
Dan

[Prisoner1138]

Well, I decided to drop the CPU clocks down to stock 2.4 (1.2GHz underclock from my everyday) to see what would happen. There were two extended spots where I was at a fairly stead 6 fps! Some spots of ~10 fps, and actually a few spots of 20-30 fps. So it seems that this game is very CPU limited (with the lack of an AGEIA card at least). BTW this was playing the tutorial, as was the first part of my last post. Also, the CPUs were more like 95% loaded as opposed to ~80% as with the higher CPU clock...
Dan

If I were to take a guess, it's probably just not optimized at all. I mean, there's no sense in Ageia optimizing the software to let it run decent without the physx card, when that's what they're trying to sell.

[BobcatDan]

If I were to take a guess, it's probably just not optimized at all. I mean, there's no sense in Ageia optimizing the software to let it run decent without the physx card, when that's what they're trying to sell.

That is most probably the case, though to AGEIA's defense I'd imagine optimizing a physics engine to be scalable to multi-core CPU's wouldn't be the easiest thing to do...

[Maverick0984]

Downloading...

[Prisoner1138]

That is most probably the case, though to AGEIA's defense I'd imagine optimizing a physics engine to be scalable to multi-core CPU's wouldn't be the easiest thing to do...

Of course not, otherwise we'd be seeing far more games that supported more cores. But I wouldn't believe for a minute that they put too much effort into it. I haven't checked it out myself, but it sounds like a "look what your uber quad-core can't do without the help of our $100 card!"

[>HyperlogiK<]

If I were to take a guess, it's probably just not optimized at all. I mean, there's no sense in Ageia optimizing the software to let it run decent without the physx card, when that's what they're trying to sell.

It's not all that they are trying to sell, remember that they licensed a software implementation to Sony for the PS3. Since Havok for GPUs seems effectively dead, Ageia could probably licence an implementation to AMD and Nvidia for a fairly large sum if they are desperate to carry on with GPU physics. Yes they want to sell PhysX cards, but they want to make money even more.

That is most probably the case, though to AGEIA's defense I'd imagine optimizing a physics engine to be scalable to multi-core CPU's wouldn't be the easiest thing to do...

Not sure about this, I wouldn't be surprised if intel relaunches Havok as a highly optimised CPU physics engine. It could be better to push a GPU version, particularly if it could be more powerful; competing with intel is a dangerous game.

[Prisoner1138]

Actually, Intel launching Havok as a highly optimized CPU engine wouldn't be a bad idea at all. It'd already include more folks with multiple core CPU's than physx cards, plus if the physics engine ran on a core, that's one less thing the game developers have to do for themselves to implement multi-core support in games, rather do something with the extra cores.