• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Infernal Engine VELOCITY Physics Tornado PC Demo

Whoa! That is 50 kinds of bad ass! Thanks for showing us this Kyle. Finally a game that will justify my purchase of the I7.
 
Awesome. But how kind will the game be to us existing Core 2 Duo/Quad users? Will we be left out in the cold?
 
Nice! I agree doing physics in a AAA game engine or title on the GPU doesn't make much sense to me. You should be having it handle the T&L and rasterization etc... graphics stuff not off-loading processor physics or tasks. If there is free cycles on the graphics card that means the engine should be uping the detail settings rights?
: )
 
ferrisnox said:
Nice! I agree doing physics in a AAA game engine or title on the GPU doesn't make much sense to me. You should be having it handle the T&L and rasterization etc... graphics stuff not off-loading processor physics or tasks. If there is free cycles on the graphics card that means the engine should be uping the detail settings rights?
: )
Click to expand...

One thing I have been told about GPU physics, when a single GPU is doing both Physics and Graphics, there is a tradeoff in graphics performance, it does diminish. This wouldn't happen with a dedicated GPU for physics, only a single GPU doing both physics and graphics. Now, I don't know about you, but I would rather have that second dedicated GPU for physics running in an SLI or CrossFire connection to speed up my graphics rather than having it sit there just doing physics, which is what the CPU should be doing, IMO, unless there are real gameplay physical differences that make it worth it, and we haven't seen any yet. All physics on GPUs has so far been effect physics, and not gameplay physics. See here to learn about those differences.
 
Holy shit that is impressive. Smooth as butter, probably running all 8 threads on that i7.:eek:
They state that is infinitely scaleable to the amount of cpus. I although I have my doubts, I imagine running this on 16 cores is not out of the realm of possibility. I would love for them to build a 4S or 2S Intel w/ hyperthreading workstation box just for this purpose and show that off.

I wonder what it takes to reach a GPU bottleneck on that engine with that many objects on the screen? The cpu maybe doing the physics, but the gpu still has to render...
 
Impressive stuff, but I just have to wonder how it works when they've got AI and realplay and all that going on at the same time. Either way, even 1,000 physics objects is a lot by and standard, and that's presumably before optimisation (i.e. removing redundant or unnecessary objects).
 
Holy shit kyle i am impressed MORE PLEASE <3 oh since when do they have an 8 core i7:confused: do they not mean 8 thread i7 :p. again another prime example that amd's next cpu microarchitecture best be multithreaded:eek:

Fixed! thanks. - Kyle
 
Last edited by a moderator:
Yay, finally something that will push all 8 threads of my i7 besides the odd x264 encodes I do :D
 
+1 wondering how this will scale DOWN to a Core2Quad system. Could make a big difference in many ways (increased i7 adoption, increased ME PISSED OFF, hehe).

Funny that they have written their own physics engine, yet it's a TWIMTBP game.
http://www.infernalgame.com/us/index.html
 
It helps that the actual game looks like it's going to be pretty good too.

All the fancy technical bells, whistles and gimmicks in the world mean jack if the actual game is crud :p
 
Modred189 said:
+1 wondering how this will scale DOWN to a Core2Quad system. Could make a big difference in many ways (increased i7 adoption, increased ME PISSED OFF, hehe).

Funny that they have written their own physics engine, yet it's a TWIMTBP game.
http://www.infernalgame.com/us/index.html
Click to expand...

You will see some of this tomorrow in the next videos....

Ghost Busters is going to work fine with a Core 2 daul core IMO. We went in and were watching CPU utilization in the GAME (NOT A DEMO) and we never saw over 50% utilized on Core 2 Quad.

Keep in mind all of this is about how the Infernal Engine can scale. Not about how much it is leveraged in Ghost Busters....which was still impressive. :)

As for TWIMTBP, you still have to have a video card to play! :)
 
And there were people yesterday commenting on the PS3's "power". Heh, I really like my PS3 but so far it's really not close to being as visually stunning or powerful as my PC with a 285GTX. Not natively @ 1920 X 1200 at least.:eek:
 
Question: we've seen these kinds of videos from other engines for a while now (physics pimping and what not). Why is this engine special enough to show in the announcements on the [H]? By the way, they look dope!

edit: after watching the plinko video, I think I know why. Is it because its multi threading the physics across the CPU, and not doing GPU physics? Very impressive.
 
inotocracy said:
Question: we've seen these kinds of videos from other engines for a while now (physics pimping and what not). Why is this engine special enough to show in the announcements on the [H]? By the way, they look dope!
Click to expand...

Well here is my personal take on it and why I think it is news worthy.

We have been hearing this GPU=physics BULLSHIT for years and how it is going to change our gaming. The fact is that it has done NOTHING to change our gaming. GPUs are being leveraged harder and harder on top end displays now days. Comes down to this. Do you want your GPU doing shader ops to give you a better picture or doing the physics that we are showing you here today that can easily be done on a CPU now.

Here is Alan Wake physics demo from 2 years ago on a dual core CPU. Now we have four cores with 8 threads. Here is a demo of what a Tornado looked like on a quad core 2 years ago. Has changed quite a bit.

ATI and NVIDIA told us that CPUs would NEVER be able to do what GPUs are doing in terms of physics and that simply is not the truth. I would suggest that this Infernal Engine is showing us some damn good "real world" in-game physics.

What do you want your CPU doing and what do you want your GPU doing?
 
This looks like it could be a reason for me to upgrade. I've been threatening to since I went with the p5n-e + e4300 more than two years ago.

E3 should be exciting this year.

Maybe PD can borrow some of this idea and implement a damage system worthy of GT5.
 
Wow, all of these done on the CPU. I can't imagine if they can use GPU for the same calculation. This engine really makes PhysX look unoptimized.
 
This is way more impressive than the PS3 demo.

However, I think they should have done 1,000 ragdolls in the tornado - box physics are old hat ;)
 
Modred189 said:
Two things:
Will be interesting to see Red Faction Guerilla, as it seems to be boasting some pretty good physicsy stuff as well, and no sign of Physx.

Second, does this argument (what would you rather your CPU/GPU be doing?) bode well for the discreet PPU market moving forward? If the physics apps require so little of the CPU, then I would say no, but who knows...
Click to expand...

Well if you want to push realistic fluid physic we are looking at loads several orders higher then what any current physic engine is showing us. Cryostasis gave us water simulation with level of detail to that of a marble, to give semi realistic water the LoD needs to get down to sand grain level. GPU physic will only make sense when there is spare power on the GPU. It might not be crazy to see NV or AMD selling portless gfx cards as physic cards in the future if physics do manage to overcome the CPU and GPU.
 
Litfod said:
I don't know, I think that AW demo stands up surprisingly well for 2 year old tech. Makes me want to play it (if they'd only give us the chance). I'd like to see something that shows off the velocity stuff in the context of the game.
Click to expand...

I was more pointing out the number of physically simulated objects in the AW tornado compared to the Tornado in the video today. Pretty big different in my eyes.

Modred189 said:
Second, does this argument (what would you rather your CPU/GPU be doing?) bode well for the discreet PPU market moving forward? If the physics apps require so little of the CPU, then I would say no, but who knows...
Click to expand...

What is the "discreet PPU market" you speak of? ;) Someone had better let NVIDIA know about it and quickly. Seriously, I have said this for a good while now, the time for the PPU / GPU physics market is past us.

kllrnohj said:
This is way more impressive than the PS3 demo.

However, I think they should have done 1,000 ragdolls in the tornado - box physics are old hat ;)
Click to expand...

Yes, it is very impressive. Given his comments that he traded out 200 ragdolls for 2000 boxes and 3500 boxes was all he could push around the room total, I don't think the system would smoothly pull off 1000 ragdolls.
 
Brent_Justice said:
Pretty amazing, I love the infinite scalability that it can scale upwards with more cores, now that Intel 80 core CPU doesn't sound so crazy.
Click to expand...
Hmmm, that & DX11 spec quad sli/xfire (when cards become available) for Win7, imagine the size/resolution display possibilities.

Would we soon to see monitors like this (perhaps even larger & higher resolution) become gaming reality?
http://www.engadget.com/2008/12/18/alioscopys-40-inch-3dhd-autostereoscopic-lcd-headed-for-ces-no/.:cool:
 
Mr. Bluntman said:
Awesome. But how kind will the game be to us existing Core 2 Duo/Quad users? Will we be left out in the cold?
Click to expand...

We get the joy of upgrading obviously..:D I mean seriously..is this not the kind of cool stuff that finally makes even the most tight fisted gamer run to the store to gleefully tear their computer apart?
 
This is definitely cool stuff but i think that the demos don't really show what it can be used for in games. What about things like *dynamically* destroyable structures? Consider a building getting shot at by a whole team using m249's versus when they shoot at it with at4's? What about when half the building is destroyed, how much more does it take to send it crashing down? The building should come down in a reasonably real way. Or when I throw a grenade does it leave a crater in the ground? There is too much focus on bouncing small things around in every physics demo I have ever seen.
 
Very impressive. It's about time someone better leveraged the CPU in physics. I'm not so sure GPU physics is dead though...

I can't help but wonder - wouldn't it be sweet to be able to run physics on both the CPU and GPU simultaneously? I'm no mathematical genius or an expert in physics, but I would guess that some equations are better suited for the in-order nature of GPUs, while others would fit well with OOO CPUs and combining the power of the two, given that the amount of communication between them isn't too severe to handle, would certainly be amazing. I recognize that this would be quite a difficult feat, but I don't see it being impossible with OpenCL nearly finalized (and able to work on current GPUs). Even without OpenCL, this would seem to be right up Larabee's alley, with supporting x86 instructions and all.

I realize the first step in something like this would be to dictate that certain functions are pre-defined to be performed on one device or the other, but I would hope that as the tech matured it'd be able to dynamically load-balance between the CPU/GPU. This way, no matter what game or game level, it'd be able to leverage more of what we already own, which, of course, leads to a better gaming experience for all, as a "dedicated" GPU or powerhouse ultra-multi-core wouldn't be quite as necessary to enable USABLE physics - although they would obviously still help. Higher efficiency all around = win. Also, this way, rigs that are aren't "balanced," so to speak, using dissimilar generation CPUs and GPUs could have whichever more-modern processor doing a bit more work.

I know neither my CPU or GPU are 100% all the time, but I'm certain that any worthwhile physics running exclusively on either processor would degrade my overall experience by over-taxing them and maxing them out, thus, slowing everything else down, but I could see them working in tandem without penalty.
 
Last edited:
Elledan said:
I keep expecting you to state that the time of hardware-accelerated graphics is past us as well. Oh, and don't forget 3D and other sound effects. They're all passe. Nothing a regular CPU can't do without wasting power on useless extra hardware.
Click to expand...

That's completely different, however, as these videos SHOW that CPU physics is not only possible, but just as good as what the gpu physics proponents have shown...
 
Zok said:
I can't help but wonder - wouldn't it be sweet to be able to run physics on both the CPU and GPU simultaneously?
Click to expand...
That's how I would want to see it. I think the idea that one method is better than the other is silly when there's just as much reason to want improved physics for their gameplay improvements as there is for eye-candy improvements.

If one can offer better in one department then they should absolutely play to their strengths in a joint environment. Even if it means not actually interacting with each other
 
Sooner or later it will be a moot point GPU/CPU processing this that or the other thing. They will be on the same chip like the FPU, L1 cache, L2 cache, memory controller, & multiple cores and we'll carry it around in our uber smart phones.
 
GPU physics is dead as a dodo after this... :D I have to agree with Kyle and Brent. I don't want my two GTX285's doing physics bullshit and yielding significantly lower framerates.

I want them rendering and harnessing their shading power. I want my physics calcs on my QX9650 thank you. Someone needs to put this GPU physics shit to rest. Intel and AMD should aggressively push Havok CPU ONLY technology.

Mind you this infernal engine positively blows Havok and Physx out of the water. :)
 
Modred189 said:
That's completely different, however, as these videos SHOW that CPU physics is not only possible, but just as good as what the gpu physics proponents have shown...
Click to expand...

I didn't see anything in those videos I haven't seen already in games like HL2 and Red Faction with its destructible environments. Years ago, with CPU-based physics.

The problem comes when you start using more complex types of physics calculations and scale up the number of physics objects. When you have over 40,000 objects and complex effects in a scene, you'd need an awful lot of CPU SIMD units to handle this workload. A GPU will handle it with one arm tied behind its back and barely feels the strain.

It's exactly the same as with hardware vs software rendering. Both will go up equally for a while until you start scaling things up. Yes, with current CPUs a software renderer will offer respectable framerates on a game like UT4, but a hardware renderer will offer framerates 10 times higher.

Why do people here hate physics? It keeps amazing me. If it isn't PhysX-bashing while Havok is being glorified (why? Havok's API is worse than PhysX's), it's people proclaiming that they never need more than CPU-based physics.

Just go back to your software-rendered, software-physics Quake I/II or so and leave the people who want to see progress alone. Please.
 
I concur with Zak, physics should be run on CPU and GPU simultaneously, balancing out my system to maximum usefulness, but I know that is not an easy thing to ask. This will be more useful when the CPU/GPU line blurs even more over the next few years. Current games are gaining more of a photo-realism, but let’s face it, the physics are still crude. The graphics and shaders are great, but future visual rendering improvements will only be steps, not leaps. Graphics are already really impressive.

But physics? Look at this game, as advanced as it is see the destructible environment? The tables break at predefined points, there’s no wisp of smoke or small spontaneous fire from wood under tremendous energy load, no blackened or bubbled varnish/polyurethane from the beam impact, and don’t get me started that real wood tends to break with the grain, usually leaving curved debris. No, physics are at least a decade behind the graphics curve. Much like AI, physics can use as much processing power as you can afford to throw at it.

You want to make a game designer sweat over a common game example? Tell him to take a dry sand pile (beach invasion scenario) and have the sand act anywhere near realistically. Heck, just give me real-looking footprints in the sand. Now add rain and try again, in your mind you know the footprints should look different. Today, sand might as well be concrete in games.
 
Modred189 said:
That's completely different, however, as these videos SHOW that CPU physics is not only possible, but just as good as what the gpu physics proponents have shown...
Click to expand...

No, it really isn't, its no where close.

GPU physics gave us deformable objects and realistic cloth. CPU physics are still stock on rigid bodies and boxes. Seriously, box physics are about as easy as it gets.

Lets see a CPU pull this off in real time: http://www.youtube.com/watch?v=ok8ThRR-59Q
 
they kinda say, gpu is bottleneck, is that what im hearing ?

So true :p

But they can use 8 threads, so quadcores, will sell better soon? new unreal engine, this engine, and rage engine.. cryengine 2.. here we go !
 
Zok
I can't help but wonder - wouldn't it be sweet to be able to run physics on both the CPU and GPU simultaneously?
Click to expand...
Pfff.. how do you think PhysX engine works ?
You playing Mirror's Edge with GPU physics - character controller, ragdolls are running on CPU
Cloth banners, smoke, particles are simulated on GPU
You playing Warmonger - ragdolls, large rigid body debris - CPU side
Ash, small fragments and other particles, cloth - GPU side
Sacred 2, UT3, etc - same thing) In PhysX SDK this mechanism is called "compartments"
That's how it works basically)
----------------
As about Velocity engine.. pretty impressive rigid body solver :)
But it's pretty weird to hear how they are highlighting in every video "we are not using GPU physics.. your GPU stays for graphics" and demonstrating.. rigid body physics ?!
There is no physics engines currently with GPU rigid body calculations support, except prototype CUDA projected gauss seidel constraint solver for Bullet physics engine)
 
Last edited:
Zogrim said:
Zok

Pfff.. how do you think PhysX engine works ?
You playing Mirror's Edge with GPU physics - character controller, ragdolls are running on CPU
Cloth banners, smoke, particles are simulated on GPU
You playing Warmonger - ragdolls, large rigid body debris - CPU side
Ash, small fragments and other particles, cloth - GPU side
Sacred 2, UT3, etc - same thing) In PhysX SDK this mechanism is called "compartments"
That's how it works basically)
----------------
As about Velocity engine.. pretty impressive rigid body solver :)
But it's pretty weird to hear how they are highlighting in every video "we are not using GPU physics.. your GPU stays for graphics" and demonstrating.. rigid body physics ?!
There is no physics engines currently with GPU rigid body calculations support, except prototype CUDA projected gauss seidel constraint solver for Bullet physics engine)
Click to expand...

That's interesting, I wasn't aware of that. Glad it's being done to some extent. Still though, I highly doubt that the CPU is being used very efficiently, considering the current ownership of NVIDIA. I just don't see NVIDIA putting in quite as much effort into CPU physics as Havok (Intel) or Velocity. Considering Velocity's "neutrality," they should have a slight advantage in that - although I bet they lack the expertise to extract efficiency out of the CPU or GPU that Intel or NVIDIA can, but this would be a negated by the fact that they use the other processor to a much greater extent. Moreover, dynamic load-balancing between the two is still a dream - one I hope comes true.
 
Zok said:
That's interesting, I wasn't aware of that. Glad it's being done to some extent. Still though, I highly doubt that the CPU is being used very efficiently, considering the current ownership of NVIDIA. I just don't see NVIDIA putting in quite as much effort into CPU physics as Havok (Intel) or Velocity. Considering Velocity's "neutrality," they should have a slight advantage in that - although I bet they lack the expertise to extract efficiency out of the CPU or GPU that Intel or NVIDIA can, but this would be a negated by the fact that they use the other processor to a much greater extent. Moreover, dynamic load-balancing between the two is still a dream - one I hope comes true.
Click to expand...

I use CUDA in some projects, and I can tell you that the thought of even using the CPU for the largest part of the processing is considered plain insanity. When you get tens to hundreds of MPs (multi-processors) on a GPU with each having a pile of processing cores, all aimed at vector (SIMD) processing, the SSE SIMD units on the CPU seem like a complete joke, and thus the CPU is delegated to control tasks, like shuffling data around and doing I/O and other tasks the GPU can't do or can't do well. For PhysX and Havok (GPU version) the solution is the same as it involves the same kind of workload, processing loads of vectors (matrices).

I must say that seeing 10-200+x speed-ups on CUDA-enabled code compared to a CPU version (multi-threaded, even) always brings a big grin to my face :D
 
Dekoth said:
We get the joy of upgrading obviously..:D I mean seriously..is this not the kind of cool stuff that finally makes even the most tight fisted gamer run to the store to gleefully tear their computer apart?
Click to expand...

True, but times are hard and not everyone (myself included) has money to run to the store and drop a few hundred into their system...
 
You must log in or register to reply here.
Back
Top