Carmack Says Dedicated Physics Processing Units Unnecessary

[apoppin]

http://www.bootdaily.com/index...6&limit=1&limitstart=1

What the QX6850 really does is help galvanize Intel as the preeminent leader in quad-core technology and should help bolster developer support as we see the industry move towards more CPU cores. id Software's leading guy, John Carmack, told us this regarding his feelings of CPU cores and physics:

?I am not a believer in dedicated PPUs. Multiple CPU cores will be much more useful in general, but when GPUs finally get reasonably fine grained context switching and scheduling, some tasks will work well there.?
-John Carmack

From that, we can see that CPU roles in gaming will become even more prevalent so it?s easy to see why Intel is shuffling a lot of resources at multi-core technology.

is Aegia out in the cold? it appears only Unreal3 seems to promise support for it.

Of course, you Ageia PPU guys have Ghost Recon2 ... but nvidia and AMD prefer to do the physics on the GPU and even Valve's Ch2 and Crysis all are using the GPU or CPU for physics - no support for the PPU at all!

 

[Stumps]

I can't see the dedicated PPUcard lasting too much longer...much like the dedicated 3D card back in the mid 90's died out.

Nvidia or ATI will go with using a seperate GPU for physics or intergrate a physics unit into GPU's in the future

OR....

as Carmack said, use multicore CPU's to do the job....

 

[SickBeast]

It would be cool if intel/AMD would purchase these guys and put a PPU as a CPU 'core'. The problem is that AMD probably can't afford to buy them, and intel has too much pride.

 

[apoppin]

i feel kinda bad for the guys that paid ... what $300? over a year ago for their Aegia PPU ... and don't ever appear to get any money's worth ... unless they LOVE one game

other then than, i really don't feel too bad ... it IS one less thing to buy

 

[BFG10K]

I think the PPU was DOA. Even in games that supported it you usually got a massive performance hit by enabling it.

Now take something like a Voodoo 1: when it came along games got much faster by enabling it and also looked better too.

3D acceleration began on a solid premise. The PPU did not. I'm not even sure what they were trying to achieve with it.

 

[NoStateofMind]

With more multi-core CPU's becoming the norm for everyday PC users, PPU is all but dead. Within a few years almost all PC's will be at least dual core with enthusiasts being a base quad core and up. That's a lot of processing power to use, physics would take advantage of those extra cores. Here's to the future of physics gaming! :beer:

 

[Bateluer]

They should have flooded the market with cheap 50 dollar PPU cards, would have gotten a lot more support.

 

[DAPUNISHER]

Originally posted by: Bateluer
They should have flooded the market with cheap 50 dollar PPU cards, would have gotten a lot more support.

Precisely, had it been the cost of a good heatsink, many more gamers would have bit. And It still seems a poor choice to have it on the PCI bus. With many boards the last couple years featuring an average of 3 PCI slots, and unused PCIe slots, asking us to reserve one for a card that is almost useless, well.....

 

[Genx87]

I really do think the PPU is a piece of hardware that was ahead of its time. This Physics on a GPU business sounds rosy but right now even the newest gen cards choke on the newest games. So cycles arent there to dedicate to the PPU.

But if you read what Nvidia and ATI are really saying, they want you to purchase a 2nd or 3rd video card to run Physics. At that point arent you right back where you started with the PPU?

The only thing I can see is mutil-core CPUs having extra cycles, or GPU's that are twiddling their thumbs for the CPU for physics on either being a reality.

 

[DAPUNISHER]

Originally posted by: Genx87But if you read what Nvidia and ATI are really saying, they want you to purchase a 2nd or 3rd video card to run Physics. At that point arent you right back where you started with the PPU?

Well, that depends on the implementation. The PPU does one thing, and one thing only. But if AMD and Nvidia do it so that the card I use for gaming, can then be demoted to PPU duty when I upgrade, that is a better alternative IMO.

 

[dreddfunk]

I think DAPUNISHER is correct, it really depends on implementation. It wouldn't be a bad thing to have the option to have your old GPU handle physics when you buy a new one.

Then again, with multiple heterogeneous cores a possibility in the not-too-distant future (10 years, possibly), I'm not sure where any of this is really going (PPU, GPU, CPU development).

 

[bigsnyder]

Originally posted by: Genx87
But if you read what Nvidia and ATI are really saying, they want you to purchase a 2nd or 3rd video card to run Physics. At that point arent you right back where you started with the PPU?

Yes, but at least you have the option of running some kind of dual/quad display or a crossfire/SLI setup if you have supporting cards. A dedicated PPU does not offer that flexibility.

Just a side note, Carmack might very well be right, but I don't take everything he says as gospel truth.

 

[Acanthus]

I still stand behind the PPU concept as sound, although its likely it will be integrated into other solutions before taking off as a standalone.

Ageia is extremely slow with releases and just doesnt have the "killer app" that really brings out the physics.

The main issue with physics imo is the multiplayer aspect, how the hell do you move 1000+ tumbling objects accurately on 2 different clients?

 

[aka1nas]

Originally posted by: bigsnyder

Originally posted by: Genx87
But if you read what Nvidia and ATI are really saying, they want you to purchase a 2nd or 3rd video card to run Physics. At that point arent you right back where you started with the PPU?

Yes, but at least you have the option of running some kind of dual/quad display or a crossfire/SLI setup if you have supporting cards. A dedicated PPU does not offer that flexibility.

Just a side note, Carmack might very well be right, but I don't take everything he says as gospel truth.

The GPU-based solution is still vaporware, so it's a little early to compare the merits of that approach.

My biggest concern with GPU-based physics is that Dx10 and newer GPU architectures are trying to move away from forcing the GPU to communicate constantly with the rest of the system for performance reasons. Having the GPU calculate physics and send it back to the CPU to update the game state is moving in the other direction. Thats why we getting this "effect" Physics BS.

 

[Ackmed]

He also said we'd have Toy Story graphics in 10 years after it came out. Well, Ive got the 10 yeah anniversary DVD.. and still waiting on the game.

 

[apoppin]

Originally posted by: Acanthus
I still stand behind the PPU concept as sound, although its likely it will be integrated into other solutions before taking off as a standalone.

Ageia is extremely slow with releases and just doesnt have the "killer app" that really brings out the physics.

The main issue with physics imo is the multiplayer aspect, how the hell do you move 1000+ tumbling objects accurately on 2 different clients?

you have one don't you?

the question is whether Aegia will survive until that killer ap ... i think nvidia and AMD insure they don't so you use their products ... and Carmack's point is that MultiCore CPU will ALSO make it moot.

i think Aegia should have "teamed-up" with one of the big players ... there are some advantages to a dedicated PPU ... but we may never see them.

 

[aka1nas]

Heterogeneous multi-core CPUs will eventually render the concept of the PPU pointless. Personally, I hope that the first on-die integrated GPUs coming next year will be able to be put to work on other tasks such as physics as that is the only thing that folks like us will find them useful for. Otherwise, it's going to be a while before multiple general-purpose cores will be able to out-perform a PPU as it's currently something like 40x faster when doing physics processing than current CPU cores.

 

[Extelleron]

I don't hate the idea of a dedicated PPU, but I just don't think it has any purpose when you look at what is going on in the CPU world.

The first problem with the PPU is a classic chicken-or-egg problem. Gamers will not buy PhysX until there are a large number of games that utilize it, meanwhile developers will not utilize PhysX in their games until a large number of gamers buy it. With the PhysX costing $250 (IIRC) initially, and $150~ now, getting gamers to adopt it will be hard. $150 is just too much to spend for something that isn't going to get much utilization. If Ageia made the PhysX around $50-60, then they'd get a good number of gamers willing to take a chance on it.

The second problem is the increasing power of processors and the move to massively multi-core designs. Dedicated PPUs like PhysX made sense in the world of single-core processors, where complex physics calculations were too much to add onto all the processes that the CPU had to execute. A dedicated processor to handle physics makes sense, because with only a single processing core, complex physics are just too much of a burden for the CPU to handle.

However, in the world of processors with 4, and soon more, cores, the PPU doesn't make much sense anymore. Now, you can dedicate a core, or even 2 (like Alan Wake will) to physics, and run the rest of the game on the other cores. There is no longer any real reason to off-load physics onto a dedicated processor. In the future, CPUs will have dozens of cores, and the benefit of a PPU will continue to diminish. There just isn't a place for the PPU when you can dedicate several, or eventually, dozens, of processing cores just to physics.

Personally I love the idea of PhysX, which is to bring complex physics into games. My "perfect game" would be a game where everything was destructible; if I fire a rocket into a wall, I want parts of the wall to be destroyed. If I have a tank, I want to be able to utterly destroy a building. IMO, full destructibility and accurate physics would completely change the gameplay and be simply amazing. I love good graphics as much as anyone, but graphics can't change gameplay like physics can. I just don't think that PhysX is the way to go.

 

[Acanthus]

Originally posted by: apoppin

Originally posted by: Acanthus
I still stand behind the PPU concept as sound, although its likely it will be integrated into other solutions before taking off as a standalone.

Ageia is extremely slow with releases and just doesnt have the "killer app" that really brings out the physics.

The main issue with physics imo is the multiplayer aspect, how the hell do you move 1000+ tumbling objects accurately on 2 different clients?

you have one don't you?

the question is whether Aegia will survive until that killer ap ... i think nvidia and AMD insure they don't so you use their products ... and Carmack's point is that MultiCore CPU will ALSO make it moot.

i think Aegia should have "teamed-up" with one of the big players ... there are some advantages to a dedicated PPU ... but we may never see them.

I dont have one, ive been waiting for a 2nd gen model for over a year now.

Who knows if they will go belly up before then.

It has a lot of hurdles to achieve success to say the least.

The news of DDR3 1600 could be the fatal blow, as current CPU-based physics models are limited by memory bandwidth.

System memory is approaching clockspeeds that can deliver reasonable performance for Graphics and Physics. All they need is a parallel controller system to match or even surpass the levels we are seeing today. (it would be expensive however, and require 4-8 dimms)

 

[Extelleron]

Originally posted by: Acanthus

Originally posted by: apoppin

Originally posted by: Acanthus
I still stand behind the PPU concept as sound, although its likely it will be integrated into other solutions before taking off as a standalone.

Ageia is extremely slow with releases and just doesnt have the "killer app" that really brings out the physics.

The main issue with physics imo is the multiplayer aspect, how the hell do you move 1000+ tumbling objects accurately on 2 different clients?

you have one don't you?

the question is whether Aegia will survive until that killer ap ... i think nvidia and AMD insure they don't so you use their products ... and Carmack's point is that MultiCore CPU will ALSO make it moot.

i think Aegia should have "teamed-up" with one of the big players ... there are some advantages to a dedicated PPU ... but we may never see them.

I dont have one, ive been waiting for a 2nd gen model for over a year now.

Who knows if they will go belly up before then.

It has a lot of hurdles to achieve success to say the least.

The news of DDR3 1600 could be the fatal blow, as current CPU-based physics models are limited by memory bandwidth.

System memory is approaching clockspeeds that can deliver reasonable performance for Graphics and Physics. All they need is a parallel controller system to match or even surpass the levels we are seeing today. (it would be expensive however, and require 4-8 dimms)

What makes you think physics on CPU is limited by memory bandwidth?

Graphics is memory bandwidth limited because you are rendering something on screen and the textures especially require a huge amount of bandwidth. Pure calculations, like physics, do not require any large amount of bandwidth.

You are right that one of the reasons we are not seeing graphics on CPUs is lack of memory bandwidth. We need considerably more bandwidth than DDR2-800 and a 64-Bit interface can provide before graphics really become possible on a CPU.

 

[sandorski]

I was thinking the Physics card was going to be big, but after the price and what not I think everyone lost any enthusiasm for it.

Thanks Carmack for sitting out the debate until the result had become obvious!! Way to predict the future!!!

 

[Acanthus]

Originally posted by: Extelleron

Originally posted by: Acanthus

Originally posted by: apoppin

Originally posted by: Acanthus
I still stand behind the PPU concept as sound, although its likely it will be integrated into other solutions before taking off as a standalone.

Ageia is extremely slow with releases and just doesnt have the "killer app" that really brings out the physics.

The main issue with physics imo is the multiplayer aspect, how the hell do you move 1000+ tumbling objects accurately on 2 different clients?

you have one don't you?

the question is whether Aegia will survive until that killer ap ... i think nvidia and AMD insure they don't so you use their products ... and Carmack's point is that MultiCore CPU will ALSO make it moot.

i think Aegia should have "teamed-up" with one of the big players ... there are some advantages to a dedicated PPU ... but we may never see them.

I dont have one, ive been waiting for a 2nd gen model for over a year now.

Who knows if they will go belly up before then.

It has a lot of hurdles to achieve success to say the least.

The news of DDR3 1600 could be the fatal blow, as current CPU-based physics models are limited by memory bandwidth.

System memory is approaching clockspeeds that can deliver reasonable performance for Graphics and Physics. All they need is a parallel controller system to match or even surpass the levels we are seeing today. (it would be expensive however, and require 4-8 dimms)

What makes you think physics on CPU is limited by memory bandwidth?

Graphics is memory bandwidth limited because you are rendering something on screen and the textures especially require a huge amount of bandwidth. Pure calculations, like physics, do not require any large amount of bandwidth.

You are right that one of the reasons we are not seeing graphics on CPUs is lack of memory bandwidth. We need considerably more bandwidth than DDR2-800 and a 64-Bit interface can provide before graphics really become possible on a CPU.

You need bandwidth because of the rate that reads and writes have to be done by the PPU.

Take 1000 objects, update their position and rotation hundreds of times a second...

You dont seem to be considering things that stay resident in memory for physics. Cloth, water, destructable rigid bodies, positional data, etc...

 

[Extelleron]

Originally posted by: Acanthus

Originally posted by: Extelleron

Originally posted by: Acanthus

Originally posted by: apoppin

Originally posted by: Acanthus
I still stand behind the PPU concept as sound, although its likely it will be integrated into other solutions before taking off as a standalone.

Ageia is extremely slow with releases and just doesnt have the "killer app" that really brings out the physics.

The main issue with physics imo is the multiplayer aspect, how the hell do you move 1000+ tumbling objects accurately on 2 different clients?

you have one don't you?

the question is whether Aegia will survive until that killer ap ... i think nvidia and AMD insure they don't so you use their products ... and Carmack's point is that MultiCore CPU will ALSO make it moot.

i think Aegia should have "teamed-up" with one of the big players ... there are some advantages to a dedicated PPU ... but we may never see them.

I dont have one, ive been waiting for a 2nd gen model for over a year now.

Who knows if they will go belly up before then.

It has a lot of hurdles to achieve success to say the least.

The news of DDR3 1600 could be the fatal blow, as current CPU-based physics models are limited by memory bandwidth.

System memory is approaching clockspeeds that can deliver reasonable performance for Graphics and Physics. All they need is a parallel controller system to match or even surpass the levels we are seeing today. (it would be expensive however, and require 4-8 dimms)

What makes you think physics on CPU is limited by memory bandwidth?

Graphics is memory bandwidth limited because you are rendering something on screen and the textures especially require a huge amount of bandwidth. Pure calculations, like physics, do not require any large amount of bandwidth.

You are right that one of the reasons we are not seeing graphics on CPUs is lack of memory bandwidth. We need considerably more bandwidth than DDR2-800 and a 64-Bit interface can provide before graphics really become possible on a CPU.

You need bandwidth because of the rate that reads and writes have to be done by the PPU.

Take 1000 objects, update their position and rotation hundreds of times a second...

You dont seem to be considering things that stay resident in memory for physics. Cloth, water, destructable rigid bodies, positional data, etc...

If physics is so bandwidth intensive, then why does the PhysX card only have 733MHz effective memory on a 128-bit bus, for 11.7GB/s of memory bandwidth, and why is it only on a PCI bus that provides very little bandwidth compared to PCI-E? The PhysX card provides less (theoretically) than DDR2-800 in a dual-channel setup, which would provide 12.8 GB/s. DDR2-1066 would provide 17GB/s.

And if that isn't enough, from X-bit labs:

The processor is equipped with the GDDR3 controller that communicates with the memory via the 128-bit bus. The memory frequency is 366 (733) MHz and the peak memory bus bandwidth equals 11.7GB/s. While top-of-the-line contemporary graphics cards boast over 50GB/s memory bus bandwidth, this number is relatively small, but AGEIA PhysX is very unlikely to suffer from insufficient memory bus bandwidth. Firstly, the PPU doesn?t need to transfer textures (which eats up a lot of memory). And secondly, AGEIA physics accelerator supports regular 32bit PCI interface with 133MB/s bandwidth, so it will turn into a real bottleneck much sooner than the local memory on the accelerator card.

 

[apoppin]

Originally posted by: sandorski
I was thinking the Physics card was going to be big, but after the price and what not I think everyone lost any enthusiasm for it.

Thanks Carmack for sitting out the debate until the result had become obvious!! Way to predict the future!!!

at least it turned into a much more interesting discussion here then i thought it would be

 

[imported_Equinox]

The biggest problem I have with a PPU is that I would have to choose between a dubious benefit (PPU) and a benefit that is guaranteed to improve my computing experience - a sound card (X-FI). There is no guess work involved as to which is more beneficial to me!