A CPU cannot match a PPU or GPU for physics!

[HOCP4ME]

Maybe it's because they don't want to buy anything, or maybe it's because they don't understand the technology, but every day there are several posts on this forum that follow the same old lines "I won't buy anything until they utillize my extra CPU core". So here are the reasons why that just doesn't work:

1. People keep saying "just use the CPU for physics". Well, that's what developers are doing right now! Physics are nothing new, and they do exist in today's games, calculated on the CPU. But we've hit a wall with this method of physics procesing. A CPU simply doesn't have the required power to do physics any better than what we have now. But what about dual-core? Well, take the number of objects a CPU can handle (around 2,000) and add 80% (dual-core CPUs aren't 100% efficient). Add 280% for quad-core. You still will be nowhere near the 30,000+ objects that a PPU or GPU can handle.

2. There are technical reasons why a CPU cannot do physics while a GPU can. It's all about pipelines. A typical GPU has between 24 and 48 pipelines. This allows it to do many calculations all at once or in parallel. A CPU, however, has very few pipelines. This means that a CPU must do instructions in a sequential manner. Physics calculations require massively parallel processing to perform well. Ideally, every object should be tracked at the same time. A GPU and PPU can do parallel processing, but a CPU isn't very good at it.

3. Here is a constructed mathematical argument:

We know that physics calculations and graphics calculations require the same type of processing power. This is why a GPU can be used to do accelerated physics.

It is obvious that we can make the statement "If a CPU can do physics, then it can do physics" using the reflexive property.

Since we know that graphics = physics as far as processing goes, we can substitute "graphics" for "physics" in that statment: "If a CPU can do physics, then it can do graphics".

We know that a CPU cannot do graphics in real time. You can verify this by attempting to run any game without a GPU installed, or with hardware acceleration disabled.

So then, we have this:

If a CPU can do physics, then it can do graphics.
A CPU cannot do graphics.
Therefore, it cannot do physics.

Now, for all you math geeks out there, that is called "denying the consequent" or the Modus Tollens argument method. It is one of two valid ways to draw a conclusion when given two premises, one of which is a conditional.


So that's why a CPU is no substitute for a GPU or PPU when it comes to physics. Can something at least be put in the FAQ about this? I have to restate this argument every day because someone else comes along and says "use the CPU'".

Disclaimer: When I say "a CPU cannot do graphics", I am not referring to raytracing or pong. I am referring to shader graphics of the quality and quantity found in today's games (not including turn-based and/or 2D games, obviously). I understand that raytracing technologies are being developed that work on a CPU and that very old games used the CPU for rendering. Likewise, when I say "a CPU cannot do physics", I am referring to accurate, realtime physics of the type and quantity promised by AGEIA, NVIDIA, Havok, and ATI (i.e. cloth, friction, tens of thousands of objects...). This has confused a few people, so I thought I would clear it up right here.

 

[gta95]

Very nice write up. I'm glad some people understand this.

 

[penguin]

I took a logic course last year, and I never thought I would use it again. Props for a quite valid and quite true argument

 

[CodeX]

You're bringing up bad memories of discreet math class... stop it you're scaring me

 

[Arcygenical]

Still, your argument is moot until they release something that uses this power .

I dont think very many people are doubting the power of the PPU... Except for the really uninformed bunch of course. Instead, they're doubting it's usefulness, and the performance/price ratio.

Right now I see the Killer NIC a better investment than the PPU.


**And my spelling is moot until I go back to english class**

 

[HOCP4ME]

Still, your argument is moot until they release something that uses this power .

I dont think very many people are doubting the power of the PPU... Except for the really uninformed bunch of course. Instead, they're doubting it's usefulness, and the performance/price ratio.

Right now I see the Killer NIC a better investment than the PPU.


**And my spelling is moot until I go back to english class**

I agree that it isn't useful today. But my argument is that a CPU cannot do physics like a PPU or GPU can. I didn't say anything about usefulness.

I took a logic course last year, and I never thought I would use it again.

Lol, it may seem like I've remember that stuff for several years, but I'm actually taking a logic course right now, so not really

 

[tvdang7]

is there even a game with that many objects that need calculating? cellfactor doesnt count.

 

[Adidas4275]

I wont buy a PPU until the games i like utilize it or require it... BF2

 

[Big Fat Duck]

1. People keep saying "just use the CPU for physics". Well, that's what developers are doing right now! Physics are nothing new, and they do exist in today's games, calculated on the CPU. But we've hit a wall with this method of physics procesing.

I stopped reading. What wall?

Show me this wall.

Even if a dev wanted to make every wall in his dream game breakable into individual bricks, there is no way GPU's (for graphics) can handle it.

 

[tornadotsunamilife]

There's a lot of technicality I don't understand, however I have reached the conclusion that pretty much any specific, focused piece of hardware (such as a GPU or PPU) will do a better job than for something which is more versatile (such as a CPU). But the question remains do we need a PPU? Do we need all that specific processing power for todays physics? I have yet to see something which has made me want to buy a PPU or wish they'd implement more physics to further the gameplay experience. Similarly, what happens when not doing something with your desktop which requires a PPU? Unlike a CPU or GPU (which are pretty much always being used on the desktop)...

Just some thoughts

 

[BinarySynapse]

A small correction. Today's CPU's have more than one pipeline and do not process information sequentially unless they have to.. The original Pentium had 2 pipelines dubbed U and V. Hyperthreading just took advantage that there were multiple pipelines. Granted though it's nowhere near the scale of an application specific processor.

Otherwise very good.

 

[hughJ]

The main thing that graphics chips had+have over CPUs is how well they perform texture filtering. Even back when 3d accelerators only had a single pipeline, only working on 1 pixel at a time, they were far faster than CPUs due to texture filtering.

Most games right now are bottlenecked by your graphics card. Newer, faster and more parallel CPUs are obviously not being utilized as well as they could be. We've got dual core CPUs out now that aren't taken advantage of very well, and we've got quad core CPUs just around the corner. Will they be as efficient as specialized hardware? Of course not. But it's definitely a step forward to use what you already have, before trying to force something new to come along.

"Hardware acceleration" is definitely a giant buzzword right now, so I don't think anyone around here are is implying that CPU core = PPU in terms of performance. But wanting your extra cores to be properly used for physics and other tasks in games isn't too much to ask before crying out for specialized hardware.

 

[jimmyb]

A CPU simply doesn't have the required power to do physics any better than what we have now. But what about dual-core? Well, take the number of objects a CPU can handle (around 2,000) and add 80% (dual-core CPUs aren't 100% efficient). Add 280% for quad-core. You still will be nowhere near the 30,000+ objects that a PPU or GPU can handle.

You are correct in saying that a PPU is vastly more effective at running physics simulations than a CPU.

You are incorrect in saying that a "CPU simply doesn't have the required power to do physics any better than what we have now". You mention, literally in the next sentence, dual core CPUs, and then go on to talk about how they will be 80% faster (per core). You have contradicted yourself. Newer CPUs (dual core or otherwise) will be able to do physics better than they can now.

Furthermore, any game that is GPU limited (i.e. most new games on common hardware configurations) has spare CPU cycles, that could potentially be used on physics.

 

[drizzt81]

I applaud the OP for an excellent post. I have tried saying the same thing in less words, but some people do not want to understand. I hope that your effort is not in vain.

I shall sum up the OP's post again, to make it simpler:

1. Assumption from ATi/ nVidia: GPUs can process physics.
2. Assumption from forum members: CPU can process physics.
3. Information from various press releases and nature: Physics is an easily parallelizable task.
4. Information from personal experience: Graphics is an easily parallelizable task.

Given that Physics and Graphics can be processes by a GPU and both are easily parallelizable tasks, it would appear that the processing is very similar, if not the same.

Conclusion:
Since we assume that both CPUs and GPUs can run physics and physics and graphics processing are very similar: a CPU can run graphics. For this reason, we should be able to eliminate GPUs, since we have multi-core CPUs, unless either or both of assumptions #1 or #2 are wrong

 

[Met-AL]

I applaud the OP for an excellent post. I have tried saying the same thing in less words, but some people do not want to understand. I hope that your effort is not in vain.

I shall sum up the OP's post again, to make it simpler:

1. Assumption from ATi/ nVidia: GPUs can process physics.
2. Assumption from forum members: CPU can process physics.
3. Information from various press releases and nature: Physics is an easily parallelizable task.
4. Information from personal experience: Graphics is an easily parallelizable task.

Given that Physics and Graphics can be processes by a GPU and both are easily parallelizable tasks, it would appear that the processing is very similar, if not the same.

Conclusion:
Since we assume that both CPUs and GPUs can run physics and physics and graphics processing are very similar: a CPU can run graphics. For this reason, we should be able to eliminate GPUs, since we have multi-core CPUs, unless either or both of assumptions #1 or #2 are wrong

Huh?

You are incorrect in saying that a "CPU simply doesn't have the required power to do physics any better than what we have now". You mention, literally in the next sentence, dual core CPUs, and then go on to talk about how they will be 80% faster (per core). You have contradicted yourself. Newer CPUs (dual core or otherwise) will be able to do physics better than they can now.
.

Ok, he said current games have around 2,000 objects...PPU's can handle 30,000 objects. What is a 80 percent improvement of 2,000?....3,600 objects which is 12 percent of 30,000 which is a 6 percent improvement. But 3,600 objects is no where near 30,000, hence the statement a CPU can't do any better.

 

[ToastMaster]

Since we assume that both CPUs and GPUs can run physics and physics and graphics processing are very similar: a CPU can run graphics. For this reason, we should be able to eliminate GPUs, since we have multi-core CPUs, unless either or both of assumptions #1 or #2 are wrong

The whole idea behind the PPU, partly, is that it attempts to form a faster link between the processing of physics calculations, and then relating those calculations to the GPU so it can process what occurs on-screen accordingly. A CPU does this also, but it takes longer, and a PPU can handle a lot more. Neither actually processes graphics however.

 

[Jason711]

its wonderful that a ppu can provide 30,000 objects, etc. we are going to need that much more power from our video cards to make it come together though. hope it all happens sooner rather than later.

 

[hughJ]

I think most of the math in this thread is a giant waste of time. People are throwing around glib statements about hardware architecture and then adding some random numbers behind them, and trying to arrive at some sort of factual conclusion.

 

[jimmyb]

Conclusion:
Since we assume that both CPUs and GPUs can run physics and physics and graphics processing are very similar: a CPU can run graphics. For this reason, we should be able to eliminate GPUs, since we have multi-core CPUs, unless either or both of assumptions #1 or #2 are wrong

Interesting point. Some people believe that the CPU and GPU will eventually converge into one piece of hardware, particularly with the generalization of GPUs and the parallelization of CPUs.

Ok, he said current games have around 2,000 objects...PPU's can handle 30,000 objects. What is a 80 percent improvement of 2,000?....3,600 objects which is 12 percent of 30,000 which is a 6 percent improvement. But 3,600 objects is no where near 30,000, hence the statement a CPU can't do any better.

3600 is larger than 2000, hence the statement "A CPU simply doesn't have the required power to do physics any better than what we have now" being incorrect (assuming the numbers to be accurate).

 

[firewolf]

No ones arguing the fact that it can improve physics they're saying they don't want to add another $200 to the price of a computer. They're already expensive enough, making these standard would force us all to buy them (which is why Aegia exists). Having crazy physics isn't even that great. So what if bullets break walls apart, I'd rather spend that $200 on a better video card and play with a higher res that might let me see some of the detail in the wall.

 

[Zepher]

Why can't we get a GPU to replace the CPU if it is faster?

 

[HOCP4ME]

I stopped reading. What wall?

Show me this wall.

Even if a dev wanted to make every wall in his dream game breakable into individual bricks, there is no way GPU's (for graphics) can handle it.

Take a look at how gameplay physics have not improved much since HL:2.

There's a lot of technicality I don't understand, however I have reached the conclusion that pretty much any specific, focused piece of hardware (such as a GPU or PPU) will do a better job than for something which is more versatile (such as a CPU). But the question remains do we need a PPU? Do we need all that specific processing power for todays physics? I have yet to see something which has made me want to buy a PPU or wish they'd implement more physics to further the gameplay experience. Similarly, what happens when not doing something with your desktop which requires a PPU? Unlike a CPU or GPU (which are pretty much always being used on the desktop)...

No, we don't need a PPU for today's physics. But we will need a PPU for tomorrow's physics, at least if we want them to be much better than what we have today.

A small correction. Today's CPU's have more than one pipeline and do not process information sequentially unless they have to.. The original Pentium had 2 pipelines dubbed U and V. Hyperthreading just took advantage that there were multiple pipelines. Granted though it's nowhere near the scale of an application specific processor.

Otherwise very good.

Okay, I'll fix that. Thanks for telling me.

The main thing that graphics chips had+have over CPUs is how well they perform texture filtering. Even back when 3d accelerators only had a single pipeline, only working on 1 pixel at a time, they were far faster than CPUs due to texture filtering.

Most games right now are bottlenecked by your graphics card. Newer, faster and more parallel CPUs are obviously not being utilized as well as they could be. We've got dual core CPUs out now that aren't taken advantage of very well, and we've got quad core CPUs just around the corner. Will they be as efficient as specialized hardware? Of course not. But it's definitely a step forward to use what you already have, before trying to force something new to come along.

"Hardware acceleration" is definitely a giant buzzword right now, so I don't think anyone around here are is implying that CPU core = PPU in terms of performance. But wanting your extra cores to be properly used for physics and other tasks in games isn't too much to ask before crying out for specialized hardware.
Yesterday 08:47 PM

Well, if we have PPU's that can do 30,000 objects, is it really worth the trouble to get an extra 1,600 from the CPU?

You are correct in saying that a PPU is vastly more effective at running physics simulations than a CPU.

You are incorrect in saying that a "CPU simply doesn't have the required power to do physics any better than what we have now". You mention, literally in the next sentence, dual core CPUs, and then go on to talk about how they will be 80% faster (per core). You have contradicted yourself. Newer CPUs (dual core or otherwise) will be able to do physics better than they can now.

Furthermore, any game that is GPU limited (i.e. most new games on common hardware configurations) has spare CPU cycles, that could potentially be used on physics.

I guess I wasn't clear on that. I meant to say "single-core CPUs". I will fix that.

As for your last statement, once again, 1,600 objects aren't worth getting if we can just get 30,000 from the PPU.

I applaud the OP for an excellent post. I have tried saying the same thing in less words, but some people do not want to understand. I hope that your effort is not in vain.

I shall sum up the OP's post again, to make it simpler:
Assumption from ATi/ nVidia: GPUs can process physics.
Assumption from forum members: CPU can process physics.
Information from various press releases and nature: Physics is an easily parallelizable task.
Information from personal experience: Graphics is an easily parallelizable task.
Given that Physics and Graphics can be processes by a GPU and both are easily parallelizable tasks, it would appear that the processing is very similar, if not the same.

Conclusion:
Since we assume that both CPUs and GPUs can run physics and physics and graphics processing are very similar: a CPU can run graphics. For this reason, we should be able to eliminate GPUs, since we have multi-core CPUs, unless either or both of assumptions #1 or #2 are wrong

Thanks for the summarization.

The whole idea behind the PPU, partly, is that it attempts to form a faster link between the processing of physics calculations, and then relating those calculations to the GPU so it can process what occurs on-screen accordingly. A CPU does this also, but it takes longer, and a PPU can handle a lot more. Neither actually processes graphics however.

I'm not quite sure what your point is, but whether you meant to or not, you just restated the reason why CPUs cannot do physics. You said "neither actualy processes graphics". But if CPUs can do physics, then they should also be able to do graphics. And since they don't do graphics, we can say that they don't do physics either. (BTW, just so everyone knows, I am talking about physics beyond the very limited physics we have right now.)

its wonderful that a ppu can provide 30,000 objects, etc. we are going to need that much more power from our video cards to make it come together though. hope it all happens sooner rather than later.

That's a good point. Perhaps the reason why developers haven't released any games that make good use of the PPU is because it would place too much of a burden on the graphics card. Hopefully the DX10 cards will solve this problem.

I think most of the math in this thread is a giant waste of time. People are throwing around glib statements about hardware architecture and then adding some random numbers behind them, and trying to arrive at some sort of factual conclusion.

Worst argument I've ever heard. What I was doing is taking assumptions made be the general public (i.e. CPUs can do physics) and proving or disproving them using facts (i.e ~2,000 objects per CPU core, a CPU cannot do graphics...) and logical argument models that ARE ALWAYS CORRECT AS LONG AS THE PREMISES ARE CORRECT. You CANNOT dispute a valid mathematical argument as long as the premises are correct. My premises were 1. A CPU cannot do graphics, and 2. A GPU can do physics. I would like to see you try to tell me that those premises are wrong.

3600 is larger than 2000, hence the statement "A CPU simply doesn't have the required power to do physics any better than what we have now" being incorrect (assuming the numbers to be accurate).

The numbers are estimates. And I guess a better way to put it would be "A CPU simply doesn't have the required power to do physics noticeably better than what we have now when compared to a PPU."

No ones arguing the fact that it can improve physics they're saying they don't want to add another $200 to the price of a computer. They're already expensive enough, making these standard would force us all to buy them (which is why Aegia exists). Having crazy physics isn't even that great. So what if bullets break walls apart, I'd rather spend that $200 on a better video card and play with a higher res that might let me see some of the detail in the wall.

Put in my perspective, that means you have to chose between:

1. Pretty colors

2. More realistic gameplay

Putting graphics in the best possible light, it becomes:

1. Games that look more real, give you sharper textures, better lighting, and smoother edges that make you feel more immersed in the game.

2. More realistic gameplay

I would still choose option two.

Why can't we get a GPU to replace the CPU if it is faster?

Well, a GPU cannot do everything, while a CPU can, albeit at a relatively slow speed. A GPU can do physics, graphics, and possibly a few other things. But it cannot do system management, multitasking, background services, AI, hard drive matinence, and CPU-intensive tasks (i.e.Photoshop, Dreamweaver, Premier...). That is why the CPU is relatively slow at everything it does. It is more a general processor for everything than a chip specially designed to do one type of processing as fast as possible.

Interesting point. Some people believe that the CPU and GPU will eventually converge into one piece of hardware, particularly with the generalization of GPUs and the parallelization of CPUs.

That could happen. When we reach 64-cores (Intel's goal for 2010), we might be able to do reasonable graphics processing on the CPU if we added a bank of video memory to the mobo. But two or four cores don't even come close.

But it appears that Moore's Law is back again in full swing. Except now it's the number of cores that doubles every two years. In fact, if Intel reaches its goal, we will be going at almost twice the pace of Moore's Law.

 

[Hvatum]

Face it, the PPU doesn't have a future if the GPU can do it all. In fact, I predict all of the phsyics processing will eventually be unloaded unto the GPU. But this extra load on the graphics card might be a problem, so we'll need to parallelize some of the graphics work. I suppose some of it could be unloaded onto an extra processing unit, let's call it a Parallel Processing Unit (PPU). In fact it might be best to just free up an entire card for Graphics work and completely dedicate the GPU to physics calculation. Then the PPU could do all graphics work. Here's my idea for a redistribution of the workload:

1. GPU: Should be primarly used for physics calculations.
2. CPU: Should be used to do sound related work (EAX 5.0 should be purely CPU based).
3. X-Fi 2: Primarly used for basic system functions, IDE, Kernel etc.
4. PPU: All the graphics work should be unloaded here.
5. IDE: The IDE channels should be put to a good use, keyboards should be connected through IDE. This could be made easy by putting external IDE connectors where the USB connectors currently are.
6. USB: USB should become the new internal connector for hard drives.
7. Firewire: Displays should connect through Firewire, dump DVI. This would improve display security and help stop content pirates, just in time for new HD-DVDs.

 

[BinarySynapse]

How about this.

CPU does coordination, basic AI, system management.
GPU process graphics
PPU process physics
IDE dies
USB is fine for the minimal task of operating a keyboard and mouse
DVI is fine for monitors since I don't know of many pirates who say "I ripped this movie from my DVI port."

 

[Hvatum]

How about this.

CPU does coordination, basic AI, system management.
GPU process graphics
PPU process physics
IDE dies
USB is fine for the minimal task of operating a keyboard and mouse
DVI is fine for monitors since I don't know of many pirates who say "I ripped this movie from my DVI port."

But that makes sense.

I like my plan better!

 

[jimmyb]

I guess I wasn't clear on that. I meant to say "single-core CPUs". I will fix that.

Expect single core cpus to get faster with time. Physics on a CPU will likewise get faster and better with more time too.

The numbers are estimates. And I guess a better way to put it would be "A CPU simply doesn't have the required power to do physics noticeably better than what we have now when compared to a PPU."

That statement I agree with. Your original statement about CPUs having "hit a wall" and not going to get any faster/better is patently untrue though. There is a huge difference, and if you are trying to write a post that clears up the facts about a topic, you better make sure you don't say anything wrong.

I wouldn't dare to say that CPUs are better at doing physics (on a pure performance metric) than PPUs. I don't believe that at all. Just to make that clear...

Having made that clear, I'm not convinced that PPUs are necessary. There is a lot of performance to be pulled from a CPU right now (given that most games are considerably GPU limited), and multicore CPUs have yet to be used in a heavy physics scenario. Note that I don't think they will be faster than PPUs, just sufficient for developer's requirements.

 

[SuperGee]

Expect single core cpus to get faster with time. Physics on a CPU will likewise get faster and better with more time too.

And what will PPU stay on that aging 130nm design with only 128Mb to adress.
I think of PPU2 PPU3 wich compete with the next gen and the gen there after CPU's
If Ageia doesn't make it. But the MArket excepted mo Physics.
GpGPU, for HavokFx and later DirectX Physics, get more powerfull and Full and efficent supporting Physics to compete further against CPU.
But do they compete no.
Physics can be with more powerfull generation of Physics hardware accelerators do Much more Physics in a higher detail and scale.
So they will more compete against other hardware sccelerators. Like PPU vs GPGPU

Having made that clear, I'm not convinced that PPUs are necessary.

Does it have to be necceserly to be a succes? Could it be if sucsesfull for years be optional supported. I think it depends how Class AAA "killer PPU apps draw gamers to it.
That it is a very long road beyond sucses wenn such hardware part get a requierment in games.

There is a lot of performance to be pulled from a CPU right now (given that most games are considerably GPU limited), and multicore CPUs have yet to be used in a heavy physics scenario. Note that I don't think they will be faster than PPUs, just sufficient for developer's requirements.

That games are more GPU limited is because the most class AAA titles and the whole industry about games is more focused to cool Graphics. this come with a cost.
Bad to medicore AI sometimes there are exceptions. Simple Gameplay. Not much game world interaction.

What can they do with Dual core and quad core.
Well a Game consist on many task. rendering and Physics and AI are just three heavier ones. Till now the most heavy one was rendering. The other task get less reserved CPU cycles from Dev's.

DC and QC cores can be used for any task. But even heavy task can be splitt in more threads. Rendering. So a other core can be used for Physics but also for more eyecandy as a SMP render module like Quake4.
CPU get more spare time of rendering and reservers power 5 to 20% for more Physics.
Well a PPU and GPU can enhanced that with more detail and at larger scale.
the Dev's deside how they gonna split the Task or threads over the Cores. Depends where there focus is on.

Also depends how many gamers apriciate mo Physics then CPU can scale for and are willing to pay for mo Physics dedicated hardware.

It's nice that CPU can handle more Physics in time. but that commonly expected to be reached in time. It's even better for game emersion and fun and advance or more interactive gameplay if you go beyond that level with dedicated Physics hardware PPU or GPU.

I am lokin forward to Boost my PPU2 with a quad core

PPU is no Substitute to CPU but a add-on wich add Physics power to whatever CPU you have. For More Physics in games.

 

[GreenMonkey]

Maybe we don't care about 30,000 flying objects in games. Maybe some of us would be happy with 1,000 objects. I know I would. So what if a PPU is 30x better at it. I'd rather they utilize a spare CPU core - that's free.

 

[drizzt81]

I think you misunderstood the topic: We are not discussion whether you want to have 30k objects or 2, it's about whether a CPU can replace a PPU and currently the answer is no. I would love any solution that would finally fix such annoyances as getting stuck inside of cabinets/ couches/ tables and on chairs in CS:S. It's be really nice if bullet spray exhibited real behavior too. All things that can easily increase the realism and my value in a game without adding 30000 more vertices.

 

[SEALTeamSix]

//snip... MArket Physics...Full ...Physics ... Much ... Physics ...Graphics. ... Simple Gameplay. ... Dual core ...Game ... Physics ...Dev's ... Task .... Cores. ... Physics... Boost ...Substitute ... Physics ...More Physics....snip//

Pssst... yOu nEEd to Keep yoUr caT(s) aWAy fRom yOuR shiFT kEy(s) wHen yOU Type.



-SEAL

 

[Lazn_Work]

2. There are technical reasons why a CPU cannot do physics while a GPU can. It's all about pipelines. A typical GPU has between 24 and 48 pipelines. This allows it to do many calculations all at once or in parallel. A CPU, however, only has one pipeline per core (two if it's hyperthreaded). This means that a CPU must do instructions one by one in a sequential manner. Physics calculations require massively parallel processing to perform well. A GPU and PPU provide this, but a CPU doesn't. In fact, to do the amount of physics a 16-pipeline GPU can do, you would probably need a 16-core processor.

I have several issues with your post, but I will start with this one.

Starting with the original Pentium CPU's had multiple pipelines (it had two).

Also most Physics is Vector math, and this is handled by the MMX/SSE units, the Core 2 from Intel can handle 6 128bit Vector calculations per clock cycle. And with a clock speed much higher than that of a PPU it is actually quite a beast of a Vector machine.

So a 24 pipeline GPU 700MHZ is equal a Core 2 at 2.8ghz in pure IPC.. But of course that proves nothing as both the GPU and CPU have other jobs to do..

 

[HOCP4ME]

Face it, the PPU doesn't have a future if the GPU can do it all. In fact, I predict all of the phsyics processing will eventually be unloaded unto the GPU. But this extra load on the graphics card might be a problem, so we'll need to parallelize some of the graphics work. I suppose some of it could be unloaded onto an extra processing unit, let's call it a Parallel Processing Unit (PPU). In fact it might be best to just free up an entire card for Graphics work and completely dedicate the GPU to physics calculation. Then the PPU could do all graphics work. Here's my idea for a redistribution of the workload:

1. GPU: Should be primarly used for physics calculations.
2. CPU: Should be used to do sound related work (EAX 5.0 should be purely CPU based).
3. X-Fi 2: Primarly used for basic system functions, IDE, Kernel etc.
4. PPU: All the graphics work should be unloaded here.
5. IDE: The IDE channels should be put to a good use, keyboards should be connected through IDE. This could be made easy by putting external IDE connectors where the USB connectors currently are.
6. USB: USB should become the new internal connector for hard drives.
7. Firewire: Displays should connect through Firewire, dump DVI. This would improve display security and help stop content pirates, just in time for new HD-DVDs.

Lol, that's funny. That's another thing. If we move physics to the GPU, what's going to do the graphics?

But if we move physics to a seperate GPU, that would essentially be a PPU and is fine with me.

The best option, though, would be to give the consumer the option of both. If you can't afford a PPU, you give up some graphics performance by using part of your GPU for physics. If you want the best performance, you buy a PPU. It pleases everybody.

Expect single core cpus to get faster with time. Physics on a CPU will likewise get faster and better with more time too.

Quote:Originally Posted by HOCP4ME
The numbers are estimates. And I guess a better way to put it would be "A CPU simply doesn't have the required power to do physics noticeably better than what we have now when compared to a PPU."

That statement I agree with. Your original statement about CPUs having "hit a wall" and not going to get any faster/better is patently untrue though. There is a huge difference, and if you are trying to write a post that clears up the facts about a topic, you better make sure you don't say anything wrong.

I wouldn't dare to say that CPUs are better at doing physics (on a pure performance metric) than PPUs. I don't believe that at all. Just to make that clear...

Having made that clear, I'm not convinced that PPUs are necessary. There is a lot of performance to be pulled from a CPU right now (given that most games are considerably GPU limited), and multicore CPUs have yet to be used in a heavy physics scenario. Note that I don't think they will be faster than PPUs, just sufficient for developer's requirements.

When CPUs get fast enough that we no longer need a GPU, we can talk. That may happen someday, but for now, it isn't happening. You admit that a CPU has less physics potential than a PPU. What you don't realize is just how much less. There is no point in trying to squeeze out the last few drops of performance from our CPUs when we have technology that can do much better.

Maybe we don't care about 30,000 flying objects in games. Maybe some of us would be happy with 1,000 objects. I know I would. So what if a PPU is 30x better at it. I'd rather they utilize a spare CPU core - that's free.

Turn you game down the 640x480, DX7, no AA, no AF, low detail.

Maybe we don't need two million pixels. Mabye some of us would be happy with only three hundred thousand pixels. So what if it can be done 7x better?

Oh, and your extra core was not free. You paid for it because you wanted to upgrade to dual-core. If you were willing to pay for that upgrade (which does almost nothing for gaming), why aren't you willing to pay for the PPU upgrade (which, when utillized, will do a lot for gaming).

I have several issues with your post, but I will start with this one.

Starting with the original Pentium CPU's had multiple pipelines (it had two).

Also most Physics is Vector math, and this is handled by the MMX/SSE units, the Core 2 from Intel can handle 6 128bit Vector calculations per clock cycle. And with a clock speed much higher than that of a PPU it is actually quite a beast of a Vector machine.

So a 24 pipeline GPU 700MHZ is equal a Core 2 at 2.8ghz in pure IPC.. But of course that proves nothing as both the GPU and CPU have other jobs to do..

Sorry about the technical mistakes. I will fix those later. But the general argument remains true.

 

[jimmyb]

When CPUs get fast enough that we no longer need a GPU, we can talk. That may happen someday, but for now, it isn't happening. You admit that a CPU has less physics potential than a PPU. What you don't realize is just how much less. There is no point in trying to squeeze out the last few drops of performance from our CPUs when we have technology that can do much better.

Seeing Cell Factor run quite well without a PPU was enough to convince me that CPUs will do just fine handling physics for the next little while. I don't doubt the power of a PPU; I doubt the necessity currently. Until CPUs start bottlenecking games, there is little reason to move load off them.

 

[SuperGee]

Maybe we don't care about 30,000 flying objects in games. Maybe some of us would be happy with 1,000 objects. I know I would. So what if a PPU is 30x better at it. I'd rather they utilize a spare CPU core - that's free.

Seeing Cell Factor run quite well without a PPU was enough to convince me that CPUs will do just fine handling physics for the next little while. I don't doubt the power of a PPU; I doubt the necessity currently. Until CPUs start bottlenecking games, there is little reason to move load off them.

Physx aint all about a lot of boxes. That just one Feature.
cloth's fluids Destructable enviorment
Also i think the Box count is way under 1000. because a PPU can do 30K of boxes if it only have to do boxes. But real games use a Mix of some Physics features where some can be heavier then all the boxes you need like 2000. Like cloth fluid. even at low scale and detail it eats PPU power and resources. 128MB limit.

CPU is a very narrow bottleneck for CF:CT with fluid and cloth on.
Fear use CPU badly it so focused on GPU GFX
Quak4r have a balanced CPU GPU use but is low on Physics.

 

[MrWizard6600]

well...

GPU pipes doing physics, get backlogged and have to be reprocessed... pipes arnt meant for physics.

and umm. HOCP4ME, do you know sum1 named Terra?

 

[drizzt81]

well...

GPU pipes doing physics, get backlogged and have to be reprocessed... pipes arnt meant for physics.

and umm. HOCP4ME, do you know sum1 named Terra?

everyone on here knows terra. Unlike the evil T, this thread's OP is making a rather solid argument.

 

[DarkSideA8]

Why should we have dedicated physics processing? Because we can.

The OP started the thread with logic. Many of the contra arguments sound like this: Why do we need to have airconditioning or stereos in our cars? You can just as easily cool yourself down by rolling down the windows (known in the industry as 2-65 a/c) and more cheaply entertain yourself by listening to your children sing in the back seat. What a bargain. The flaw is in the argument.

If you remember playing BF 1942, you might recall killing yourself by driving a jeep into a fence or a tank into a tree. Having driven jeeps and tanks in real life, I can tell you that fences and trees do not fare so well when run into/over by real life military equipment. Having a dedicated physics processor would allow the fence to shatter and the tree to be knocked down. Instead, currently, such objects are purely static. Boring.

If you played Duke Nukem, you might remember seeing that wall with the peculiar shape telling you that if you hit it with the rocket launcher a hole would appear. Cool at the time. Now? Yawn. Shooting the walls left little markers which would fade overtime (self healing walls - wish I had those back in my dorm-days...)

As it stands, any object with a collision mesh (like a wooden fence) will block a simulated bullet. However, if you've ever been in combat, wooden fences provide good concealment, but not good cover.

The potential is reason enough for why we should do it. If you recall Remedy's Matrix-like demo (from Max Payne - Remedy being the first outfit to actually send simulated bullets down range) a PPU would allow for the actual destruction of the walls in response to different caliber bullets. Otherwise all we're doing is painting the walls ala Duke Nukem.

As has been said: the CPU should calc AI and general world information, the GPU paint it for us and calculate the lighting (as it does better and better each year) and we should have a dedicated PPU (solely because we can) to make things realistically breakable. Each year that would get better and better and the content available to the end user (i.e. us gamers - become more and more awesome (because lets face it we all need better reasons to avoid real life! ...and frankly its more fun to play "Desert Combat" than to be actually shot at in the desert.).

 

[Lazn_Work]

Why should we have dedicated physics processing? Because we can.

The OP started the thread with logic. Many of the contra arguments sound like this: Why do we need to have airconditioning or stereos in our cars? You can just as easily cool yourself down by rolling down the windows (known in the industry as 2-65 a/c) and more cheaply entertain yourself by listening to your children sing in the back seat. What a bargain. The flaw is in the argument.

If you remember playing BF 1942, you might recall killing yourself by driving a jeep into a fence or a tank into a tree. Having driven jeeps and tanks in real life, I can tell you that fences and trees do not fare so well when run into/over by real life military equipment. Having a dedicated physics processor would allow the fence to shatter and the tree to be knocked down. Instead, currently, such objects are purely static. Boring.

If you played Duke Nukem, you might remember seeing that wall with the peculiar shape telling you that if you hit it with the rocket launcher a hole would appear. Cool at the time. Now? Yawn. Shooting the walls left little markers which would fade overtime (self healing walls - wish I had those back in my dorm-days...)

As it stands, any object with a collision mesh (like a wooden fence) will block a simulated bullet. However, if you've ever been in combat, wooden fences provide good concealment, but not good cover.

The potential is reason enough for why we should do it. If you recall Remedy's Matrix-like demo (from Max Payne - Remedy being the first outfit to actually send simulated bullets down range) a PPU would allow for the actual destruction of the walls in response to different caliber bullets. Otherwise all we're doing is painting the walls ala Duke Nukem.

As has been said: the CPU should calc AI and general world information, the GPU paint it for us and calculate the lighting (as it does better and better each year) and we should have a dedicated PPU (solely because we can) to make things realistically breakable. Each year that would get better and better and the content available to the end user (i.e. us gamers - become more and more awesome (because lets face it we all need better reasons to avoid real life! ...and frankly its more fun to play "Desert Combat" than to be actually shot at in the desert.).

Don't forget an AIPU: http://www.hardforum.com/showthread.php?t=1094855 eventually we won't need a CPU at all. yay.

 

[Dan_D]

Why should we have dedicated physics processing? Because we can.

The OP started the thread with logic. Many of the contra arguments sound like this: Why do we need to have airconditioning or stereos in our cars? You can just as easily cool yourself down by rolling down the windows (known in the industry as 2-65 a/c) and more cheaply entertain yourself by listening to your children sing in the back seat. What a bargain. The flaw is in the argument.

If you remember playing BF 1942, you might recall killing yourself by driving a jeep into a fence or a tank into a tree. Having driven jeeps and tanks in real life, I can tell you that fences and trees do not fare so well when run into/over by real life military equipment. Having a dedicated physics processor would allow the fence to shatter and the tree to be knocked down. Instead, currently, such objects are purely static. Boring.

If you played Duke Nukem, you might remember seeing that wall with the peculiar shape telling you that if you hit it with the rocket launcher a hole would appear. Cool at the time. Now? Yawn. Shooting the walls left little markers which would fade overtime (self healing walls - wish I had those back in my dorm-days...)

As it stands, any object with a collision mesh (like a wooden fence) will block a simulated bullet. However, if you've ever been in combat, wooden fences provide good concealment, but not good cover.

The potential is reason enough for why we should do it. If you recall Remedy's Matrix-like demo (from Max Payne - Remedy being the first outfit to actually send simulated bullets down range) a PPU would allow for the actual destruction of the walls in response to different caliber bullets. Otherwise all we're doing is painting the walls ala Duke Nukem.

As has been said: the CPU should calc AI and general world information, the GPU paint it for us and calculate the lighting (as it does better and better each year) and we should have a dedicated PPU (solely because we can) to make things realistically breakable. Each year that would get better and better and the content available to the end user (i.e. us gamers - become more and more awesome (because lets face it we all need better reasons to avoid real life! ...and frankly its more fun to play "Desert Combat" than to be actually shot at in the desert.).

Well said. I agree completely. The PPU can in fact open worlds of gameplay to us.

The game should be able to tell the difference between bullets hitting cinder blocks and hitting wooden fences. It would be a whole new way to play the game if you get to chasing someone, they run into a building, and you happen to have a way to knock that building down around them and on top of them.

You could change the face of the maps you play on with wanton destruction.

The thought of playing games like that gives me a chubby.

 

[Lord Cecil]

Not to be rude, but almost no one here, including both the 14 year old OP and myself, know much of anything on this matter. I'm here to point out that a current gpu can not replace the cpu for physics. A ppu is not involved in my argument.

By the OP's logic a gpu can already do physics and can already do physics better than the cpu. Thus it makes no logical sense as to why there isn't a single game on the market that uses the gpu for physics. Clearly, something is missing from the logical equation.

The something would be called engineering. CPUs are general purpose number crunchers for general purposes while GPUs are for a very specific purpose. The difference here is that cpus are made to account for branching. Lots and lots of branching. This is something physics requires a lot of with dozens of if statements run per object on those 30,000 boxes. Luckily the cpu is designed for such things and can handle this quite well.

GPUs on the otherhand don't really consider branching. Graphics doesn't need it. You setup some initial matrix transformations, turn on some shaders, and then churn through thousands of polygons and vertices without a single conditional statement.

Thus the statement "If a CPU can do physics, then it can do graphics." is false and entire argument falls apart.

Good day.