Just thought of this..physics insight

[amddiesel]

With all of the hype about physics and graphics cards doing these calculations to reduce strain on the cpu..Why dont the game makers dump all the physics calculations onto the second core of a dual core processor..Thats the way games are being threaded in the future and we really dont see much benefit yet from dual cores just yet in gaming... Does my long daydreaming thought make any logical sense? No flames ..lol :twisted:

 

[prozac26]

Thats the way games are being threaded in the future and we really dont see much benefit yet from dual cores just yet in gaming

That's because dual cores were available for the customer for only a year, so they havent developed games to take advantage of customer dual cores. Vista will have much better dual core support, and so will games coming out for Vista (even earlier).

 

[Maxis17]

Yes, you make perfect sense, especially considering that that is exactly what Crytek is doing with Crysis - physics on one core, the rest on the other for dual-core owners. Because of this, Crytek recommends a good dual core CPU if you want to play Crysis smoothly. Tx...

 

[Track]

Yes, you make perfect sense, especially considering that that is exactly what Crytek is doing with Crysis - physics on one core, the rest on the other for dual-core owners. Because of this, Crytek recommends a good dual core CPU if you want to play Crysis smoothly. Tx...

I personally dont like this very much.
I mean, if ur saying that the GFX cards do almost everything in games, ur being fair, because then i can spend all my money on a good GFX card and neglect the CPU, wich would save me a lot of money. But now that CPUs are becoming more important in games, im going to have to spend twice as much and who has enough money to get both a top of the line GFX card and a top of the line CPU, and especially where i live?

 

[TheGreatGrapeApe]

With all of the hype about physics and graphics cards doing these calculations to reduce strain on the cpu..Why dont the game makers dump all the physics calculations onto the second core of a dual core processor..

It makes some sense until you see the numbers for the VPU and PPU based phyics, which can run at high multiples (10-20X) faster than a CPU, and therefore the second CPU can't do nearly the same amount of shiny eye-candy physics. The second core would be perfect for the truely game dependant phsyics that is a minor load in most games, then the VPU/PPU could do the cool effects stuff.

Games like Oblivion already use the second core for some basic phsyics and AI stuff, but to do the things like found in GRAW or that will be in UT2K7 / Crysis would bring them to a screeching halt compared to the add-in solutions.

 

[prozac26]

Games like Oblivion already use the second core for some basic phsyics and AI stuff

Dude, I didn't know that.

 

[TheGreatGrapeApe]

Yep, the surprising thing is that the difference is minor because the load doesn't seem to be very heavy. I was surprised by the minimal difference of the second core, especially considering the relative complexity of the AI, and some 'good' (not great) physics.

 

[ivoryjohn]

With all of the hype about physics and graphics cards doing these calculations to reduce strain on the cpu..Why dont the game makers dump all the physics calculations onto the second core of a dual core processor..Thats the way games are being threaded in the future and we really dont see much benefit yet from dual cores just yet in gaming... Does my long daydreaming thought make any logical sense? No flames ..lol :twisted:

Because the second core of a dual core is only ONE SINGLE additional core, while the GPU's and Ageias PPU offer MANY MANY available cores for physics processing. IBM's cell processor (not x86) can dump physics on the additional cores (up to 8 of them depending on yield).

A dual core can't compete with a handful of shader cores.

 

[viper84]

Ivoryjohn is closest to reason why games aren't threaded that way. The reason why there are specialist physics processors and the suggestion that GPU's could be used is because they are all massively parrelled compared to a relatively non-parrell CPU.

 

[sailer]

As the others said, some games are starting to use the second core and more will do that as time goes by. At the same time, as GreatGrapeSaid, the differences so far are few.

A bigger difference in processing speed would probably be gotten by making dual core gpu's. But isn't that part of the idea with SLI/Crossfire anyway? Then come the cooling problems. There is no free lunch, so they say. As I remember, VooDoo did the multiple core thing years ago, going as far as having four cores on a card if I remember correctly. Didn't work all that well then, but VooDoo was dying as a company anyway so it may just not have had a chance to work.

 

[4745454b]

I would like to point out one other problem. I heard someone suggest something like this before, an thought it was cool until I remembered some things. First, GPUs have way more transistors and processing power then a CPU, quite a bit more if you count only one core. You then have the memory bandwith issue. Current video cards use super clocked DDR3/4, resulting in bandwith of >50gbps. What does dual channel DDR2 provide? Nothing even close.... Last, you now have to deal with higher latancies. Its going to take time to move data between the GPU and second core.

 

[korsen]

as far as i care, they should just leave physics processing to GPU's. The cpu will never be good enough to do such things unless you design one from the ground up to be that way, and Ageia's card can suck it when you put physics on a gfx card thrown in a pcie16 slot. And by the way, for those of you who didn't read the physics article, Agieas card has built in physics calculations that cannot be changed in the hardware. GPU's are open book and can be changed if someone finds a better calc. So GPU's 1, Ageia 0.

 

[sailer]

I agree that the physics should be left to the gpu's and that the Ageia cards, while nice in the idea, don't work as well as hoped. I think it would have been far better if the way Ageia worked was in software, rather than hardware. Ageia did little more than show that there were ways of adding more realism to games, but it didn't show the best way. For now, all we can do is wait.

 

[Gary_Busey]

The physics card was dead before it even had a chance to sell. Even though it's not graphics, it's still in competition with ATI and Nvidia. You think ATI and Nvidia are going to leave this physics area untouched? Nope, and with this peaking the publics interest, they're both going to offer new solutions to physics handling that will perform better. They're too big for Ageia to compete against them.

 

[sailer]

Agreed. Agiea had one chance to sell physics cards, and it didn't manage to do it right. Now Ageia will probably join VooDoo in the list of dead companies. Hopefully, Nvidia and ATI will do what Ageia didn't, package and sell the idea in a product that does make a real difference in graphics.

 

[prozac26]

Yep, the surprising thing is that the difference is minor because the load doesn't seem to be very heavy. I was surprised by the minimal difference of the second core, especially considering the relative complexity of the AI, and some 'good' (not great) physics.

Ah, I see.

 

[Gary_Busey]

I could see ATI and Nvidia integrating advanced physics into their cards, maybe even making a seperate line of cards with advanced physics and a budget line of regular cards.

 

[TheGreatGrapeApe]

Because the second core of a dual core is only ONE SINGLE additional core, while the GPU's and Ageias PPU offer MANY MANY available cores for physics processing. IBM's cell processor (not x86) can dump physics on the additional cores (up to 8 of them depending on yield).

Yeah, but it's not just the cores but the fact that the GPU and especially PPUs are better designed for physics math.

Remember those dual core processors are running at like 4-6 times the speed of the VPUs, so if it was just the cores they should be much closer.

 

[TheGreatGrapeApe]

You then have the memory bandwith issue. Current video cards use super clocked DDR3/4, resulting in bandwith of >50gbps. What does dual channel DDR2 provide? Nothing even close....

Yeah but that has less to do with the speed (which granted is usually about twice that of PCs), but the bit width of the memory. The PC memory is much 'thinner' than the graphics cards's 128/256/512bit memory, usually creating 4+X the bitwidth clock for clock.

Last, you now have to deal with higher latancies. Its going to take time to move data between the GPU and second core.

Except that the VPU already has a priority assignment in most architectures (of which Hyper Memory and TurboCache also take advantage). So througput and latency aren't hus but of course they'll be slower than across the package from core to core.

 

[TheGreatGrapeApe]

I could see ATI and Nvidia integrating advanced physics into their cards, maybe even making a seperate line of cards with advanced physics and a budget line of regular cards.

Actually we'll see. With the way DX10 cards are more programmable and can do more complex calculations, we'll se what's left missing from the equation, and really what's needed to bring then in line whether it's dedicated tansistors or something that has more global applications.

I think the ability to divide the VPU up as separate functioning units will greatly increase the possibilities, but in some areas they may still fall short of a dedicated PPU.

We'll probably have a much better picture of this by Xmas.