Poll: What would the price of Physx card have to be

[Acanthus]

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: Avalon

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: inveterate
Guys,, why is it that the cpu can only handle 30-40 physics objects while this thing can do like "30000"

is the card actually 1000 faster than a normal cpu? why not just make a software that can run on a cheapo computer and have that thing supply the physics data?

I TELL U answer to both.. No it's not faster,, and 2 they're assholes who just want me to buy more stuff. like seriously, first it's graphics "accelerator cards" then its SOUND cards,, Then RAM,, wtf was that about,, now PHYSICS accelerators... soon they're gonna make DVD accelerator cards that's only function is to make the tray come out faster...

I'm really pissed off at this. they're just dividing up these computing tasks so they can sell more stuff at once. and u know the sadest part is that i'm gonna have to buy it.. GRRRRRR. Same deal with dual core, instead of making faster processors, they just glued 2 togther so they can sell u 2 processors at one time. Less development cost, excuse to make more money.. I hate this,,, Now i'm gonna go save money to buy this physx crap...!!!!

This is the single, least educated thing i have read about computing in a very long time.

This is right up there with daily comments like " i need to get more memory for storage " and "my computer is only 4 years old and its still fast because i upgraded it 2 years ago"

It is very educated. It breaks down the nonsense that corporate assho*es want you to believe into the raw components of their chicanery. Yet because there is a monopoly on production, one can't make their own computer parts, we're forced to oblige. You're the one that is too quick to judge what I've said. Granted I was typing with aim speech so it probably appears crude.

So you honestly believe that we havent reached a clockspeed ceiling, and both of the corporate assholes, along with the entire microengineering community, are "fooling us" into buying parrallel processors.

You do realize that a CPU cant do Video, Physics, or sound anywhere near the complexity levels that these dedicated cards do, right?

Easy explanation: 2.7ghz cpu does instructions in order, 1+1+1+1+1+1+1+1 etc...

GPUs, and PPUs do them IN PARRALLEL.

Which means:
GPU: (with 24 pipelines)
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1

Which one do you think is going to finish 1st?

You have no idea what youre talking about, and you look foolish to those who do.

<== EE student, like many on these boards.

The reason CPUs cant be clocked up more is because the MATERIALS SCIENCE OF HUMANITY HASNT COME UP WITH ANYTHING BETTER THAN SILICON YET.

WOWOWO! I didn't say I have any degrees, nor did I say I knew the science or math. All Ive said is that what the corporate machine is doing is done in the biggest portion to get our money. I also didn't say there is anything wrong with it except that I'm forced to dangle on their chain. It makes no difference in my perspective how much math/science you know. They could've integrated everything together, but no they're going to sell us multiple things to get more money.

You're wrapped up in the fact that you think you are smarter than everyone else, and you feel compelled to prove it. Even though it doesn't apply to the point Im making

If they sold everything integrated to your motherboard (assuming it was possible), your motherboard would cost $1000-$1500.

Do you think its even possible to have things like a GPU and PPU, both with very different memory subsystems, and multiple controllers (read: a sh!t ton of traces) as well as system memory and CPU/s? youre talking a 30 layer motherboard, if not more.

Not to mention the lack of elegance, you could no longer upgrade.

It hasnt been done because its not cost effective, this method would indeed cost substantially more.

You are too into the technicalities, while it isn't what I was argueing about

Umm you're bitching because it isn't done this way, because you think it'd be cheaper and/or faster... and i'm listing the reasons why that isn't true.

How can you not get technical? We are talking about computer architecture.

 

[Avalon]

Originally posted by: Acanthus

Originally posted by: Avalon

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: inveterate
Guys,, why is it that the cpu can only handle 30-40 physics objects while this thing can do like "30000"

is the card actually 1000 faster than a normal cpu? why not just make a software that can run on a cheapo computer and have that thing supply the physics data?

I TELL U answer to both.. No it's not faster,, and 2 they're assholes who just want me to buy more stuff. like seriously, first it's graphics "accelerator cards" then its SOUND cards,, Then RAM,, wtf was that about,, now PHYSICS accelerators... soon they're gonna make DVD accelerator cards that's only function is to make the tray come out faster...

I'm really pissed off at this. they're just dividing up these computing tasks so they can sell more stuff at once. and u know the sadest part is that i'm gonna have to buy it.. GRRRRRR. Same deal with dual core, instead of making faster processors, they just glued 2 togther so they can sell u 2 processors at one time. Less development cost, excuse to make more money.. I hate this,,, Now i'm gonna go save money to buy this physx crap...!!!!

This is the single, least educated thing i have read about computing in a very long time.

This is right up there with daily comments like " i need to get more memory for storage " and "my computer is only 4 years old and its still fast because i upgraded it 2 years ago"

It is very educated. It breaks down the nonsense that corporate assho*es want you to believe into the raw components of their chicanery. Yet because there is a monopoly on production, one can't make their own computer parts, we're forced to oblige. You're the one that is too quick to judge what I've said. Granted I was typing with aim speech so it probably appears crude.

So you honestly believe that we havent reached a clockspeed ceiling, and both of the corporate assholes, along with the entire microengineering community, are "fooling us" into buying parrallel processors.

You do realize that a CPU cant do Video, Physics, or sound anywhere near the complexity levels that these dedicated cards do, right?

Easy explanation: 2.7ghz cpu does instructions in order, 1+1+1+1+1+1+1+1 etc...

GPUs, and PPUs do them IN PARRALLEL.

Which means:
GPU: (with 24 pipelines)
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1

Which one do you think is going to finish 1st?

You have no idea what youre talking about, and you look foolish to those who do.

<== EE student, like many on these boards.

The reason CPUs cant be clocked up more is because the MATERIALS SCIENCE OF HUMANITY HASNT COME UP WITH ANYTHING BETTER THAN SILICON YET.

WOWOWO! I didn't say I have any degrees, nor did I say I knew the science or math. All Ive said is that what the corporate machine is doing is done in the biggest portion to get our money. I also didn't say there is anything wrong with it except that I'm forced to dangle on their chain. It makes no difference in my perspective how much math/science you know. They could've integrated everything together, but no they're going to sell us multiple things to get more money.

You're wrapped up in the fact that you think you are smarter than everyone else, and you feel compelled to prove it. Even though it doesn't apply to the point Im making

If they sold everything integrated to your motherboard (assuming it was possible), your motherboard would cost $1000-$1500.

Do you think its even possible to have things like a GPU and PPU, both with very different memory subsystems, and multiple controllers (read: a sh!t ton of traces) as well as system memory and CPU/s? youre talking a 30 layer motherboard, if not more.

Not to mention the lack of elegance, you could no longer upgrade.

It hasnt been done because its not cost effective, this method would indeed cost substantially more.

I definitely agree. It would be near impossible to do, and certainly not practical at all. What would bug me the most is the fact that if you wanted to upgrade your system, you'd have to buy another entire system. I like my system divided into seperate components, it just makes sense.

 

[tfcmasta97]

I'd pay like $40 to see ****** fly everywhere. i dont think i need to see 10000 boxes fly.... like 100 smoothly would do fine.

 

[inveterate]

Originally posted by: Acanthus

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: Avalon

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: inveterate
Guys,, why is it that the cpu can only handle 30-40 physics objects while this thing can do like "30000"

is the card actually 1000 faster than a normal cpu? why not just make a software that can run on a cheapo computer and have that thing supply the physics data?

I TELL U answer to both.. No it's not faster,, and 2 they're assholes who just want me to buy more stuff. like seriously, first it's graphics "accelerator cards" then its SOUND cards,, Then RAM,, wtf was that about,, now PHYSICS accelerators... soon they're gonna make DVD accelerator cards that's only function is to make the tray come out faster...

I'm really pissed off at this. they're just dividing up these computing tasks so they can sell more stuff at once. and u know the sadest part is that i'm gonna have to buy it.. GRRRRRR. Same deal with dual core, instead of making faster processors, they just glued 2 togther so they can sell u 2 processors at one time. Less development cost, excuse to make more money.. I hate this,,, Now i'm gonna go save money to buy this physx crap...!!!!

This is the single, least educated thing i have read about computing in a very long time.

This is right up there with daily comments like " i need to get more memory for storage " and "my computer is only 4 years old and its still fast because i upgraded it 2 years ago"

It is very educated. It breaks down the nonsense that corporate assho*es want you to believe into the raw components of their chicanery. Yet because there is a monopoly on production, one can't make their own computer parts, we're forced to oblige. You're the one that is too quick to judge what I've said. Granted I was typing with aim speech so it probably appears crude.

So you honestly believe that we havent reached a clockspeed ceiling, and both of the corporate assholes, along with the entire microengineering community, are "fooling us" into buying parrallel processors.

You do realize that a CPU cant do Video, Physics, or sound anywhere near the complexity levels that these dedicated cards do, right?

Easy explanation: 2.7ghz cpu does instructions in order, 1+1+1+1+1+1+1+1 etc...

GPUs, and PPUs do them IN PARRALLEL.

Which means:
GPU: (with 24 pipelines)
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1

Which one do you think is going to finish 1st?

You have no idea what youre talking about, and you look foolish to those who do.

<== EE student, like many on these boards.

The reason CPUs cant be clocked up more is because the MATERIALS SCIENCE OF HUMANITY HASNT COME UP WITH ANYTHING BETTER THAN SILICON YET.

WOWOWO! I didn't say I have any degrees, nor did I say I knew the science or math. All Ive said is that what the corporate machine is doing is done in the biggest portion to get our money. I also didn't say there is anything wrong with it except that I'm forced to dangle on their chain. It makes no difference in my perspective how much math/science you know. They could've integrated everything together, but no they're going to sell us multiple things to get more money.

You're wrapped up in the fact that you think you are smarter than everyone else, and you feel compelled to prove it. Even though it doesn't apply to the point Im making

If they sold everything integrated to your motherboard (assuming it was possible), your motherboard would cost $1000-$1500.

Do you think its even possible to have things like a GPU and PPU, both with very different memory subsystems, and multiple controllers (read: a sh!t ton of traces) as well as system memory and CPU/s? youre talking a 30 layer motherboard, if not more.

Not to mention the lack of elegance, you could no longer upgrade.

It hasnt been done because its not cost effective, this method would indeed cost substantially more.

You are too into the technicalities, while it isn't what I was argueing about

Umm you're bitching because it isn't done this way, because you think it'd be cheaper and/or faster... and i'm listing the reasons why that isn't true.

How can you not get technical? We are talking about computer architecture.

Whose "we're", you are talking about computer architecture, and I?m talking about corporate greed. I've already said 3 times that your arguements are irrelevant. And I agree and believe everything you said except for your attacks on me because you like being a smart as*s hat

 

[Acanthus]

Originally posted by: tfcmasta97
I'd pay like $40 to see ****** fly everywhere. i dont think i need to see 10000 boxes fly.... like 100 smoothly would do fine.

cellfactor demo, ill pm you my paypal address.

 

[Acanthus]

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: Avalon

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: inveterate
Guys,, why is it that the cpu can only handle 30-40 physics objects while this thing can do like "30000"

is the card actually 1000 faster than a normal cpu? why not just make a software that can run on a cheapo computer and have that thing supply the physics data?

I TELL U answer to both.. No it's not faster,, and 2 they're assholes who just want me to buy more stuff. like seriously, first it's graphics "accelerator cards" then its SOUND cards,, Then RAM,, wtf was that about,, now PHYSICS accelerators... soon they're gonna make DVD accelerator cards that's only function is to make the tray come out faster...

I'm really pissed off at this. they're just dividing up these computing tasks so they can sell more stuff at once. and u know the sadest part is that i'm gonna have to buy it.. GRRRRRR. Same deal with dual core, instead of making faster processors, they just glued 2 togther so they can sell u 2 processors at one time. Less development cost, excuse to make more money.. I hate this,,, Now i'm gonna go save money to buy this physx crap...!!!!

This is the single, least educated thing i have read about computing in a very long time.

This is right up there with daily comments like " i need to get more memory for storage " and "my computer is only 4 years old and its still fast because i upgraded it 2 years ago"

It is very educated. It breaks down the nonsense that corporate assho*es want you to believe into the raw components of their chicanery. Yet because there is a monopoly on production, one can't make their own computer parts, we're forced to oblige. You're the one that is too quick to judge what I've said. Granted I was typing with aim speech so it probably appears crude.

So you honestly believe that we havent reached a clockspeed ceiling, and both of the corporate assholes, along with the entire microengineering community, are "fooling us" into buying parrallel processors.

You do realize that a CPU cant do Video, Physics, or sound anywhere near the complexity levels that these dedicated cards do, right?

Easy explanation: 2.7ghz cpu does instructions in order, 1+1+1+1+1+1+1+1 etc...

GPUs, and PPUs do them IN PARRALLEL.

Which means:
GPU: (with 24 pipelines)
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1

Which one do you think is going to finish 1st?

You have no idea what youre talking about, and you look foolish to those who do.

<== EE student, like many on these boards.

The reason CPUs cant be clocked up more is because the MATERIALS SCIENCE OF HUMANITY HASNT COME UP WITH ANYTHING BETTER THAN SILICON YET.

WOWOWO! I didn't say I have any degrees, nor did I say I knew the science or math. All Ive said is that what the corporate machine is doing is done in the biggest portion to get our money. I also didn't say there is anything wrong with it except that I'm forced to dangle on their chain. It makes no difference in my perspective how much math/science you know. They could've integrated everything together, but no they're going to sell us multiple things to get more money.

You're wrapped up in the fact that you think you are smarter than everyone else, and you feel compelled to prove it. Even though it doesn't apply to the point Im making

If they sold everything integrated to your motherboard (assuming it was possible), your motherboard would cost $1000-$1500.

Do you think its even possible to have things like a GPU and PPU, both with very different memory subsystems, and multiple controllers (read: a sh!t ton of traces) as well as system memory and CPU/s? youre talking a 30 layer motherboard, if not more.

Not to mention the lack of elegance, you could no longer upgrade.

It hasnt been done because its not cost effective, this method would indeed cost substantially more.

You are too into the technicalities, while it isn't what I was argueing about

Umm you're bitching because it isn't done this way, because you think it'd be cheaper and/or faster... and i'm listing the reasons why that isn't true.

How can you not get technical? We are talking about computer architecture.

Whose "we're", you are talking about computer architecture, and I?m talking about corporate greed. I've already said 3 times that your arguements are irrelevant. And I agree and believe everything you said except for your attacks on me because you like being a smart as*s hat

So how is it corporate greed if they are doing it the right way?

 

[akugami]

inveterate, I think you're under the mistaken impression that Microsoft, Intel, Apple, AMD, ATI, nVidia, etc are nonprofit organizations. I work in the business world (not tech related) and I can tell you that there are a lot of costs involved in marketing a new product. Cost of manufactoring aside, new technology like Aegia's PhysX card also costs a lot to research and develope. Not to mention convincing games companies to support the new tech.

Ask yourself this, if it was your company would you try to sell it for as much as the market is willing to pay? If so, you're just as greedy as the corporations you're accusing of greed. These companies have to answer to shareholders (owner or some other type of boss if it's not a public company). It is their job to make as much money as possible. Not to give you value for your money. If they could package feces and make money off of it, they'd do it.

A specialized processor like a graphics accelerator are faster than general purpose processors like the central processing unit (cpu) in a normal computer. There is no question even if game AI isn't factored in, a processor built specifically to handle physics will be faster than running it through a general processor like a cpu. The question you then have to ask yourself is if it's worth it to you to buy these add-ons. No one is putting a gun up to your head to buy a computer much less expensive add-ons like GPU's and PPU's.

 

[inveterate]

Originally posted by: akugami
inveterate, I think you're under the mistaken impression that Microsoft, Intel, Apple, AMD, ATI, nVidia, etc are nonprofit organizations. I work in the business world (not tech related) and I can tell you that there are a lot of costs involved in marketing a new product. Cost of manufactoring aside, new technology like Aegia's PhysX card also costs a lot to research and develope. Not to mention convincing games companies to support the new tech.

Ask yourself this, if it was your company would you try to sell it for as much as the market is willing to pay? If so, you're just as greedy as the corporations you're accusing of greed. These companies have to answer to shareholders (owner or some other type of boss if it's not a public company). It is their job to make as much money as possible. Not to give you value for your money. If they could package feces and make money off of it, they'd do it.

A specialized processor like a graphics accelerator are faster than general purpose processors like the central processing unit (cpu) in a normal computer. There is no question even if game AI isn't factored in, a processor built specifically to handle physics will be faster than running it through a general processor like a cpu. The question you then have to ask yourself is if it's worth it to you to buy these add-ons. No one is putting a gun up to your head to buy a computer much less expensive add-ons like GPU's and PPU's.

I'm glad this is relevant. and i appreciate ur input. however i disagree with the gun to my head part. Being an addict gamer, they're essentially telling me buy our stuff or don't play the games

 

[DeathReborn]

Originally posted by: Avalon

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: inveterate

Originally posted by: Acanthus

Originally posted by: inveterate
snip

snip

snip

snip

snip

If they sold everything integrated to your motherboard (assuming it was possible), your motherboard would cost $1000-$1500.

If they integrated everything into the Motherboard I wouldn't touch it. That's more interference than the wife when trying to watch football (soccer). In a way they already have someting close to that and it's a Laptop. I'm sure if this takes off Intel will wedge one into a Laptop mobo and add $200 to the price.

There are going to be games that require GPU/PPU based physics as they are more capable at it than the current CPU's.

 

[Munky]

Originally posted by: Acanthus
Easy explanation: 2.7ghz cpu does instructions in order, 1+1+1+1+1+1+1+1 etc...

GPUs, and PPUs do them IN PARRALLEL.

Which means:
GPU: (with 24 pipelines)
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1

Which one do you think is going to finish 1st?

That's not entirely true. A cpu is first of all pipelined, which means the various stages of the pipeline can work on multiple things simultaneously, such as executing one instruction while decoding the next one, while fetching a predicted branch, etc. Now, where the rubber meets the road, the cpu's are still a lot less parallel than a gpu, but they do exhibit some instruction level parallelism. For example if the A64 has 2 of its FPU's on 2 different ports, then it can execute a FLOP on each port at once, so you already have 2 instructions per cycle. In a dual core cpu you have even more potential parallelism, but the code has to take advantege of it.

The benefits of specialized hardware are much greater speed and efficiency in executing the instructions it was designed for, but that also means it has a more limited scope of abilities than a general purpose cpu. In fact the gpu itself had a very limited range of applications until it was given shader capability. I'm curious if the physx ppu will be programmable, and to what extent. If the ppu is nothing more than a processor with a specific set of pre-defined instructions, then it may actually limit the scope of the physics that the developers must use in order to take advantage of the added speed.

 

[Acanthus]

Originally posted by: munky

Originally posted by: Acanthus
Easy explanation: 2.7ghz cpu does instructions in order, 1+1+1+1+1+1+1+1 etc...

GPUs, and PPUs do them IN PARRALLEL.

Which means:
GPU: (with 24 pipelines)
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1

Which one do you think is going to finish 1st?

That's not entirely true. A cpu is first of all pipelined, which means the various stages of the pipeline can work on multiple things simultaneously, such as executing one instruction while decoding the next one, while fetching a predicted branch, etc. Now, where the rubber meets the road, the cpu's are still a lot less parallel than a gpu, but they do exhibit some instruction level parallelism. For example if the A64 has 2 of its FPU's on 2 different ports, then it can execute a FLOP on each port at once, so you already have 2 instructions per cycle. In a dual core cpu you have even more potential parallelism, but the code has to take advantege of it.

The benefits of specialized hardware are much greater speed and efficiency in executing the instructions it was designed for, but that also means it has a more limited scope of abilities than a general purpose cpu. In fact the gpu itself had a very limited range of applications until it was given shader capability. I'm curious if the physx ppu will be programmable, and to what extent. If the ppu is nothing more than a processor with a specific set of pre-defined instructions, then it may actually limit the scope of the physics that the developers must use in order to take advantage of the added speed.

I simplified the idea tremendously.

 

[Acanthus]

Originally posted by: munky

Originally posted by: Acanthus
Easy explanation: 2.7ghz cpu does instructions in order, 1+1+1+1+1+1+1+1 etc...

GPUs, and PPUs do them IN PARRALLEL.

Which means:
GPU: (with 24 pipelines)
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1

Which one do you think is going to finish 1st?

That's not entirely true. A cpu is first of all pipelined, which means the various stages of the pipeline can work on multiple things simultaneously, such as executing one instruction while decoding the next one, while fetching a predicted branch, etc. Now, where the rubber meets the road, the cpu's are still a lot less parallel than a gpu, but they do exhibit some instruction level parallelism. For example if the A64 has 2 of its FPU's on 2 different ports, then it can execute a FLOP on each port at once, so you already have 2 instructions per cycle. In a dual core cpu you have even more potential parallelism, but the code has to take advantege of it.

The benefits of specialized hardware are much greater speed and efficiency in executing the instructions it was designed for, but that also means it has a more limited scope of abilities than a general purpose cpu. In fact the gpu itself had a very limited range of applications until it was given shader capability. I'm curious if the physx ppu will be programmable, and to what extent. If the ppu is nothing more than a processor with a specific set of pre-defined instructions, then it may actually limit the scope of the physics that the developers must use in order to take advantage of the added speed.

Only on processors that support SMT (an intel cpu with hyperthreading, the Xbox GPU) can you have more than one process in the pipeline itself at once.

As for programmability, Ageia claims that custom algorithms can be written for engines and that through driver updates they can increase functionality. They also claim the life cycle of a ppu to be much longer than a GPU.

 

[43st]

Originally posted by: Acanthus

Originally posted by: munky

Originally posted by: Acanthus
Easy explanation: 2.7ghz cpu does instructions in order, 1+1+1+1+1+1+1+1 etc...

GPUs, and PPUs do them IN PARRALLEL.

Which means:
GPU: (with 24 pipelines)
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1

Which one do you think is going to finish 1st?

That's not entirely true. A cpu is first of all pipelined, which means the various stages of the pipeline can work on multiple things simultaneously, such as executing one instruction while decoding the next one, while fetching a predicted branch, etc. Now, where the rubber meets the road, the cpu's are still a lot less parallel than a gpu, but they do exhibit some instruction level parallelism. For example if the A64 has 2 of its FPU's on 2 different ports, then it can execute a FLOP on each port at once, so you already have 2 instructions per cycle. In a dual core cpu you have even more potential parallelism, but the code has to take advantege of it.

The benefits of specialized hardware are much greater speed and efficiency in executing the instructions it was designed for, but that also means it has a more limited scope of abilities than a general purpose cpu. In fact the gpu itself had a very limited range of applications until it was given shader capability. I'm curious if the physx ppu will be programmable, and to what extent. If the ppu is nothing more than a processor with a specific set of pre-defined instructions, then it may actually limit the scope of the physics that the developers must use in order to take advantage of the added speed.

Only on processors that support SMT (an intel cpu with hyperthreading, the Xbox GPU) can you have more than one process in the pipeline itself at once.

As for programmability, Ageia claims that custom algorithms can be written for engines and that through driver updates they can increase functionality. They also claim the life cycle of a ppu to be much longer than a GPU.

What about multiplayer? How do the PPU chips work in a networked environment?

 

[BenSkywalker]

I also have no doubts that ATI and Nvidia will come up with their own versions built into the GPU's.

Whichever one does it first will likely die off pretty quickly. Every transistor being used for physics is a transistor not being used for graphics, every bit of bandwidth is the same thing. Utilize enough of the GPU to offload physics and you will get your @ss handed to you in vid stress tests versus a PPU+GPU(perhaps even a CPU+GPU at high enough settings). Don't get me wrong, offering the option for lower tier systems may be a good idea, but when the bench charts come out do you want to explain to people that you get clobbered because you are doing additional work? Didn't work for nV in the FX days(FP32 v FP24) and it won't work with physics either.

 

[Munky]

What if the cpu was given an additional 3rd "core" if you will, specialized for vector processing, and it was used for physics? I'm not talking about a SSE/3DNOW unit, but something along the lines of the vector processors found on the PS2, which has 5 FMAC and 2 FDIV units. If it was made to work at modern cpu clockspeeds, it would give the cpu a huge boost in float calculations performance. Maybe the PC and the cpu have not evolved to that stage yet, but with consoles we've already seen this concept being used with success.

 

[PingSpike]

Originally posted by: munky

Originally posted by: fishbits

Right... but that can't be used for game play, only visual effects. Say that "seed" gives one player a perfect hiding place and not the others? I think it's fairly certain, due to bandwidth and host/client constraints, that PhysiX will have zero multiplayer support.

Visual effects only... seems to me gamers have proven not only willing but eager for hardware that gives them more in that department. Guys who just plopped $300+ on a GPU are willing to do it again in 6-12 months for incremental advances in textures, frame rates. Give them a $100-200 hardware option to add many new facets to the visual experience and one that should be good for say 3 years, and I see many gamers adopting once they've laid eyes on it. That's placing zero value on all the other things that could be enabled. After the uber-gamers get it, the power gamers then the regular gamers will get it too as word spreads, competition ramps up and prices come down.

I don't think the idea of seeding a physical event is getting across though. Think of it as one of those elaborate toppling dominoes setups. Can do some pretty intricate and large things, and all you have to do is get it started. The bandwidth would only be knocking over the first domino, physics processing handles the inevitable outcome in real-time, at the same pace for everyone. The seed, setup, whatever would be the same on everyone's system and map-based, so the outcome would be the same. Alternately it could be a single random numeric variable generated at the time of the event and sent along with the packet. How does an area of effect weapon have the bandwidth to send out every possible coordinate it affects? It doesn't. You just have to send the starting point of the blast, and let each machine take care of the rest. Of course there's lag, but that applies for moving and firing a weapon too. The "start event" packet is sent, and each player's machine handles the rest. This setup however would require every system to have physics processing, which will be the case soon enough in one form or another. Guess having lived through some people saying "math-coprocessors/ sound cards/ 4MB of ram/ $200 graphics cards/ CD Roms will never catch on" has given me a different perspective.

I see what you're saying, but that brings up a number of other issues, such as:
1. What happens when you have a slow machine and a fast machine running the same triggered event (remember the events have to be synchronized)
2. What if one machine has a PPU and one doesnt
3. How would a dedicated PPU make a noticeable difference when it's a pre-defined event and can be run as a shader on the gpu, given that it only affects the visual aspect of the game

1. You could get away with this by making the effects not synced...of course they would have to be visual and not involved in the game (tiny rocks and stuff) but if they're non interactive the effect is less then impressive.
2. Of course it will have to play to the least common demominator during the transition period...just like when 3d cards first came out. There were software renderers. However if the software renders look just as good, or even just close...the value of the card is going to diminish. There's a lot of extra processing power available with dual cores these days too...so the performance hit might not even be signifigant.
3. Prerended events do not need to be processed. If the outcome is to be the same for everyone's machines...then you already know the outcome right? Which would probably make the event a trivial action of reading memory. Seems like it would sort of make the physics card useless.

The multiplayer aspect is a tough nut to crack. The united states doesn't have fat pipes going to everyone's house...there isn't really bandwidth to spare. You could process the physics at the server and then try to sync it...but with the precision we're talking about that just plain isn't going to be possible. Plus, none of the clients would even need the card. You can process it on the clients, but then it won't be synced. And the bottom line is for the objects to be interactive, they MUST be synced in some capacity.

So basically you're just going to see some nice visual effects in multiplayer at the most...unless there is some magic way to do it that I'm not aware of, the internet link is the lowest common denominator that games play too...and our processing power is already way beyond what it can reasonably be expected to push out to the clients. Everything that can be done locally is.

 

[Acanthus]

Originally posted by: Thera

Originally posted by: Acanthus

Originally posted by: munky

Originally posted by: Acanthus
Easy explanation: 2.7ghz cpu does instructions in order, 1+1+1+1+1+1+1+1 etc...

GPUs, and PPUs do them IN PARRALLEL.

Which means:
GPU: (with 24 pipelines)
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1
1+1+1+1

Which one do you think is going to finish 1st?

That's not entirely true. A cpu is first of all pipelined, which means the various stages of the pipeline can work on multiple things simultaneously, such as executing one instruction while decoding the next one, while fetching a predicted branch, etc. Now, where the rubber meets the road, the cpu's are still a lot less parallel than a gpu, but they do exhibit some instruction level parallelism. For example if the A64 has 2 of its FPU's on 2 different ports, then it can execute a FLOP on each port at once, so you already have 2 instructions per cycle. In a dual core cpu you have even more potential parallelism, but the code has to take advantege of it.

The benefits of specialized hardware are much greater speed and efficiency in executing the instructions it was designed for, but that also means it has a more limited scope of abilities than a general purpose cpu. In fact the gpu itself had a very limited range of applications until it was given shader capability. I'm curious if the physx ppu will be programmable, and to what extent. If the ppu is nothing more than a processor with a specific set of pre-defined instructions, then it may actually limit the scope of the physics that the developers must use in order to take advantage of the added speed.

Only on processors that support SMT (an intel cpu with hyperthreading, the Xbox GPU) can you have more than one process in the pipeline itself at once.

As for programmability, Ageia claims that custom algorithms can be written for engines and that through driver updates they can increase functionality. They also claim the life cycle of a ppu to be much longer than a GPU.

What about multiplayer? How do the PPU chips work in a networked environment?

Fidelity is one of the things Ageia touts as a feature, 2 different PCs will yeild the same results in a networked environment.

 

[Acanthus]

Originally posted by: munky
What if the cpu was given an additional 3rd "core" if you will, specialized for vector processing, and it was used for physics? I'm not talking about a SSE/3DNOW unit, but something along the lines of the vector processors found on the PS2, which has 5 FMAC and 2 FDIV units. If it was made to work at modern cpu clockspeeds, it would give the cpu a huge boost in float calculations performance. Maybe the PC and the cpu have not evolved to that stage yet, but with consoles we've already seen this concept being used with success.

With consoles you have nothing even approaching the level of physics in those demos.

The physx core is a 16 core processor iirc...

Not to mention there is a LOT more memory bandwidth available to the card vs a cpu.

 

[HomeAppraiser]

$150

 

[xtknight]

For the love of ******, don't use so many quotes!

 

[Keysplayr]

These cards are probably going to be priced as much as the corporation thinks the enthusiasts market will bear, due to the enthusiasts niche it is intended for.

But remember guys, GPU's weren't always GPU's. The original GeForce was "I think" the first Geometry Processing Unit, designed to take "some" duties from the CPU. Or allow for a slower CPU without sacrificing performance.

What makes you think, (any of you) that it's not possible or cost effective to incorporate a physics processing unit into the core of lets say G90, or R700? Believe it, ATI and Nvidia are working on this somehow. Either that, or one of the companies will buy up AGEIA.

If a X1900XTX or 7900GTX costs 500 bucks (just for example), who here would not pay an extra 100.00 if it was equipped with an integrated physics processor? I am seeing people in here say they would pay up to 200.00 bucks for a standalone physics board.

I think a lot of you are looking at the AGEIA chip and maybe thinking that this is the only way to accomplish this. Completely untrue. All those who are keeping up with this stuff knows that HAVOC is working on software to utilize idle GPU transistors for just this reason.

In conclusion, I would say NOT to jump on the AGEIA card, at least not right out of the gate. Wait for the competitors to arrive. Whoever they may be. HAVOC, NVIDIA, ATI etc. etc.

My $0.85

Keys

 

[Philippine Mango]

Originally posted by: Thera
Need a poll option for "Don't want one".

I think I'll see what ATI, Nvidia, and dual core bring to the table before spending anything on this. You can already run the demos on regular hardware, no sense in spending the cash if a dual core can take it.

Noobish comments like these sadden me...

 

[Philippine Mango]

Originally posted by: Arkane13131
100$ sounds reasonable. i mean why would it cost 300$ as thats the same price as a 3800+ x-2 dual core

Because a dual core is worlds behind in physics processing speed than the physics card, THATS WHY.

 

[Keysplayr]

Originally posted by: Philippine Mango

Originally posted by: Thera
Need a poll option for "Don't want one".

I think I'll see what ATI, Nvidia, and dual core bring to the table before spending anything on this. You can already run the demos on regular hardware, no sense in spending the cash if a dual core can take it.

Noobish comments like these sadden me...

Energy wasted on comments like this sadden me when you could have explained yourself with even a "little" extra effort.

 

[Philippine Mango]

Originally posted by: Avalon

Originally posted by: inveterate
Guys,, why is it that the cpu can only handle 30-40 physics objects while this thing can do like "30000"

is the card actually 1000 faster than a normal cpu? why not just make a software that can run on a cheapo computer and have that thing supply the physics data?

I TELL U answer to both.. No it's not faster,, and 2 they're assholes who just want me to buy more stuff. like seriously, first it's graphics "accelerator cards" then its SOUND cards,, Then RAM,, wtf was that about,, now PHYSICS accelerators... soon they're gonna make DVD accelerator cards that's only function is to make the tray come out faster...

I'm really pissed off at this. they're just dividing up these computing tasks so they can sell more stuff at once. and u know the sadest part is that i'm gonna have to buy it.. GRRRRRR. Same deal with dual core, instead of making faster processors, they just glued 2 togther so they can sell u 2 processors at one time. Less development cost, excuse to make more money.. I hate this,,, Now i'm gonna go save money to buy this physx crap...!!!!

That was the best laugh I had today.

QFT, thank you for the great laugh kid!