Physics Processing Unit

[PeteRoy]

It sounds like another way to make money, I don't believe we really need this, modern CPU and modern GPU are already very high powered and they don't need another processing unit in the computer.

 

[RaiderJ]

The more transistors a chip has, the more it costs to produce. That increase in cost is also non-linear - thusly a chip with twice as many transistors will cost more than twice as much to produce (assuming all else is equal). A chip that does graphics, processing, physics, etc in all one chip does would do so at a cost that is greater that the cost of producing those chips separately. This is why you see add-in cards (and for flexibility reasons).

Processors are designed in such a way to maximize their effectiveness at a certain task - therefore no one design could ever be the "jack of all trades". If you think about it, all a motherboard does is act as an information bus for all the different processors in a computer. Instead of on-die interconnects, you have PCI(e) slots, ATA ports, PS/2 ports, USB, Firewire, etc. The motherboard allows flexibility in that only those processors that are deemed neccessary in a system can be added in, thus reducing cost. Certain elements have been incorporated "on-board" because they were deemed important enough to become permanent features. That is why we now have onboard ethernet, USB, ATA/SATA, etc.

So, Mr. Controversial, while you may feel that a dual-core (or triple, etc) chip is the answer to everything, it is not. A physics board is not a step backward, but a step forward. Specialized processors are on every component of your computer, and you will continue to see new ones added.

 

[mhillary]

Originally posted by: Geforcer
Sounds cool, it's just too bad the games have to be designed to use the technology. I guess they have to start somewhere with it

sounds cool to me too. I think they will start something new like a new RPG game

 

[DAPUNISHER]

Originally posted by: PeteRoy
It sounds like another way to make money, I don't believe we really need this, modern CPU and modern GPU are already very high powered and they don't need another processing unit in the computer.

You didn't try the rocket demo eh? Give it a go tell me how high powered your system already is This type of tech is necessary for the realism we all want in games.

 

[nellienelson1]

mr_c says that the cpu can handle it when they become multi cored. yet in the aricle it says that the cpu can process 00's and the ppu 0000's. so yes the cpus will speed up but they will need a hell of a lot (massively) parrallel processors to catch up. the ppu is massively parrallel, cpu's arent by design. the cpu needs to be able to do so many different things, whilst ppu, needs to know where the 'bone' is. its current velocity, the force applied. thats it. it then calculates the new velocity sees what that bone does to others and moves on, this started off great in my head and has now bailed. a cpu cant do this as it will have to at the same time think about the inputs from the user, whilst sending the data of what is happening in the game to the ppu and then to the gpu, that cant be done on the scale of the ppu in the cpu without a whole new specifically designed ppu core, you cant just use another amd64 core bolted on the side as it doesnt have the nesscesary design in it in herently. (class play spot the spelling and grammar errors, there are loads!!!)

have some of that to suck on and then flame as you see fit.

 

[nellienelson1]

just wonder what fps ppl get on exploding building with th rocket programme that was linked to i get 3.85 at peak on my laptop!!

 

[shinotenshi]

I think this is a great idea, far more important 3d graphics. while most here are just viewing in terms of frame rates, i think they miss the point. 3d graphics and sound are about presentation, the topping on yout all beef pattie. This however is different, this is a true paradigm shift, more than just mere presentation or superficial, this is down to bone gameplay and mechanics change. this isn't evolution this is revolution. becuase the scope of physics can be broaden to encompass everything, must like how einsteins's theory differ from newtons. just think about half life 2 on the moon was differ from hl2 on earth or in a gas gaint or underwater. I have no doubt that this will have massive developer support, because all the truly great developers, id, bioware,blizzard et al, see themselves not a peddling games but works of art, and no artist no matter how rich or risk averse has ever meant an idea that would expand his artistic possiblies that he didnt like.

1a. the idea that a cpu could even rever replace this is too silly for words, completely ignores reality, physics and has no merit.

 

[VirtualLarry]

Originally posted by: DAPUNISHER
Better and more versatile hardware is coming all the time for audio&video encoding&decoding, 2D/3D graphics processing, now physics, maybe specialized AI processors next? Couldn't this type of system architecture eventually make "central processing unit" no longer an operable term? The Borg computing model?

It's a bit of a circular cycle, actually, as before the "integrated microprocessor", there actually were discrete processing sub-elements, inplemented seperately, and wired all together onto one logic board. For example, the integer ALU, the control unit/sequencer ROM, interface bus, etc. were seperate on older mini-computers.

We are witnessing the return of multiple discrete processing units all wired together. Perhaps they will be eventually integrated into one, but then the cycle will likely begin anew again. In fact, many microprocessors were designed specifically to support "co-processors" on the bus (I'm sure you know this, DP, but making an example for readers), such as the Motorola 68000 series, the MIPS series, and a number of others. Strangely absent in that group is the x86 CPU lineup, which defines signals for multi-master-processor (SMP) systems, but not one single master processor, and multiple slave co-processor elements. (There used to be some support for a dedicated 8087 FPU co-processor, but AFAIK those were all removed when the '87 went on-die starting with the 486DX CPUs. But that wasn't an open-ended general purpose co-processor bus arrangement like that others I mentioned.)

 

[VirtualLarry]

Originally posted by: MrControversial
Like Punisher said, if we keep adding add-on cards, we wouldn't need a CPU. This is so counter productive. We want less components. Would you rather have a smaller CPU with a larger motherboard and larger case, or would you rather have a larger multi-cored CPU with memory/audio/video on chip

Actually, I would want the best price/performance mix of dedicated hardware and general-purpose CPU power. Intel's big push to support "software modems", back in the Pentium MMX days, was designed to load down your main CPU so much doing trivial tasks that were ill-suited to it (and better suited for dedicated hardware), that in order to perform ordinary software duties, a user would be forced to upgrade their CPU to one faster (and costlier) than they would otherwise need.

I mean, which is cheaper, $30 worth of discrete hardware for a modem controller + DSP, or 40% of the CPU power of a $200 main CPU? Hmm. The math seems easy for me to do. It's quite the irony that many people during that era chose that sort of solution, believing that spending $60 for a good-quality hardware modem was "too expensive". Yet, their systems suffered mightily because of it.

 

[VirtualLarry]

Originally posted by: MrControversial
My point is that this PPU concept seems to be going against the grain of more powerful CPU's handling more tasks and eliminating specialized hardware. However, since these CPUs haven't come out fast enough, this bolten on card seems like a stop gap. I'll be very suprised if PPU cards become the standard and this tech isn't put on a graphics card or delegated to a powerful mutli-core CPU. Stop gaps like these exists so developers get what they want NOW instead of waiting for hardware to mature to the point where the CPU can handle it with ease.

Your confusion seems to stem from a belief that past history of developments in the CPU technology sector, are absolutely indicators of future performance, and that effective performance for certain tasks nearly doubles every few years or so. But in light of current process-size, heat-output, and related issues, pressuring both mainstream desktop CPU makers to both move to multi-core on the desktop, should tell you something - that the past is the past, and the future of desktop CPU evolution is not going to be the same as it was in the past, leading up to this point.

It's kind of like this - imagine that single-core CPUs are like a single-driver sports-car. They've managed to make them go faster and faster up until this point, due to improvements in wheels, engines, transmissions, etc. But finally, they've hit the redline, and can't make the transmission any more efficient, and can't run the engines that they have any faster, due to the current physical limits of the materials that they are using. If they did, they would just explode.

So they take a different tact - multi-core. Instead of a faster sports-car, now they are evolving the personal-computing minivan. It won't go any faster, in terms of absolute speeds, than the sports car will, and in many cases won't go even as fast. But for those tasks that involve potentially multiple trips using the sports car, they could be done in a single trip using the larger (multiple) carrying capacity of the minivan, for passengers or cargo. IOW, the time that it takes to get from point A to point B using either method of transportation will not be reduced, but the effective carrying capacity is increased.

So they have pushed the latency of current CPU architectures to the limit, and now they are looking to increase the bandwidth (effective overall work/clock done) instead.

If the Phys-X marketing material can be believed, then what you seemingly fail to understand, MrControversial, is that they are promising an actual order-of-magnitude improvement in terms of the ability of this chip to compute physics operations. Think of it as a personal Learjet for transport. Sure, eventually, it might be possible for a car to catch up to a Learjet in terms of speed in theory, but in terms of real-world usage today, it provides a clear advantage in lowering transportation time, for those that are willing to pay the additional price for it.

Sure, eventually, when all cars will be able to fly and break the sound barrier, then perhaps the Learjets of today will be obsoleted. But that future is not guaranteed, and there is no immediate timeframe for their development. Hopefully that all made sense.

Edit: Sorry, I guess I had already replied to this thread in the past.

 

[Jeff7181]

Originally posted by: DAPUNISHER

Originally posted by: PeteRoy
It sounds like another way to make money, I don't believe we really need this, modern CPU and modern GPU are already very high powered and they don't need another processing unit in the computer.

You didn't try the rocket demo eh? Give it a go tell me how high powered your system already is This type of tech is necessary for the realism we all want in games.

Yeah, that's a really cool demo. The physics still need to be tweaked... all the demo's look like it's all happening under water. But yeah... that building explosion is VERY demanding. In some of the simpler demo's I get over 1000 FPS. But in that one I get around 3.

 

[Gannon]

I think people are missing the business sense from the game developers perpsective. Game sales for the PC haven't been amazing compared to consoles.

PPU will fail plain and simple. They were talking about it needing 128MB of RAM, thats simply way too much, imagine adding another 128MB of ram to a video card PLUS the cost of the PPU. it will never fly, PC game developers who think this is a good idea and who already have a tough time with predicting market video card upgrades should think twice about another add in card that will fragment the market... AGAIN, theres no way Joe consumer is going to be able to keep up with TWO addin cards.

It's no wonder console gaming is taking over completely, 3D video cards fragmented the gaming market into players that could have good graphics and players who could have just bearable ones. If you do that with physics that will screw-up the whole market up even more, now they won't have to worry just about graphics, but whether the user has the latest "PPU". Bad idea unless they integrated it onto video cards, video cards are already prohibitivey expensive as it is.

Ideally all gaming related functions should be done on a single card to minmize confusion and stop market fragmentation.

Just think about this: Majority of gamers will have entry level 3D cards or mid range cards, but not PPU's. Only the hardcore would have PPU's so games would have a huge divide in their ability of sustaining and computing numerous objects on the screen. Not only that, it would make Multiplayer gaming a nightmare as the number of objects on a screen increase their position data has to be transmitted to everyone on the server. Frankly the latency and bandwidth is not there yet on the broadband side of things either.

 

[biostud]

The good thing about having a PPU is that "if" it actually offloads lots of calculations from the CPU, then you wouldn't need to change CPU as often. Many games are atm CPU limited so all cards that can offload the CPU would be a benefit IMHO. But in the start it will probably be very enthusiast market. But when you're games start to run a bit choppy, then instead of doing a mboard + CPU upgrade, you will instead invest in a PPU card. I would also suspect that the PPU product cycle would be a lot slower than GPU, so the investment might be of better value.

 

[MobiusPizza]

Good point, that might be true.
It all comes down to the hand of software and game developers' of how much support from them there is

If more software take advantage of it, it will be a success for PSUs

 

[biostud]

Originally posted by: AnnihilatorX
Good point, that might be true.
It all comes down to the hand of software and game developers' of how much support from them there is

If more software take advantage of it, it will be a success for PSUs

I guess PSUs have a good chance of becoming a succes

 

[biostud]

nm

 

[PerfeK]

They have to be able to scale PPU's well for them to catch on. You can't have objects acting one way for one player and another way for another player in a multiplayer environment. They can probably improve aesthetic details like hair, clothes, swaying branches and grass but objects like balls, desks and walls have to be the same across the board. Perhaps systems without PPU's will just be forced to use the CPU to process those things. That way everyone will have access to the settings but with a larger performance hit on non PPU systems.

 

[Reapsy00]

Do they say anywhere how much it is likely to cost? 128MB DDR3 sounds kind of expensive. Still if it means I get to see little bits of brains and bone splattering all over the place I'm all for it

 

[biostud]

Originally posted by: Reapsy00
Do they say anywhere how much it is likely to cost? 128MB DDR3 sounds kind of expensive. Still if it means I get to see little bits of brains and bone splattering all over the place I'm all for it

Apperently there will be different version priced like videocards.

 

[MobiusPizza]

Originally posted by: PerfeK
They have to be able to scale PPU's well for them to catch on. You can't have objects acting one way for one player and another way for another player in a multiplayer environment. They can probably improve aesthetic details like hair, clothes, swaying branches and grass but objects like balls, desks and walls have to be the same across the board. Perhaps systems without PPU's will just be forced to use the CPU to process those things. That way everyone will have access to the settings but with a larger performance hit on non PPU systems.

The calculation done by the PPU is not a random process and so there should not be any discrepencies between two PPU calculating the same interaction..

As said already CPU at all are not optimised for Physics calculation. It can process several hundred rigid bodies at a time; But in PPU's term it's no match at all. If a game is designed such that there are 1000 objects to interact; There's no way a player without PPU would be able to play...

 

[biostud]

News about PhysX

 

[Lonyo]

Bad news IMO. I don't want to have to pay that much for it.
Unless it brings a LOT to gaming, I don't think it will be an initial hit until costs come down or we see some integration like on the motherboard or something.

 

[biostud]

Originally posted by: Lonyo
Bad news IMO. I don't want to have to pay that much for it.
Unless it brings a LOT to gaming, I don't think it will be an initial hit until costs come down or we see some integration like on the motherboard or something.

I would think software support would be the keyword. I like that it has it own card as you can carry it over to the next computer, just like a soundcard.

 

[Soffty]

I'v heard this is going to be the next big thing in gaming.

 

[biostud]

a new overview