• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

The real nuke in Crysis. Do we need a PPU? (Not an Ageia thread)

FOR EVERYONES ATTENTION

As in another PPU thread around here, I recently studied PPU on agieas physx card on agieas own benchmarks......

It was a good chunk of FPS faster in CPU mode (~15/20fps), apart from cloth trearing was a couple (1 or 2 fps faster) in hardware mode, my cpu is 3.1ghz 2mb L2 cache core 2.

NOW 2/3 years ago agieas hardware was faster at ppu calc compared to the old processors, nowadays its not, roughly the same, intels next cpu will again be faster still.

Simply PPU's need to get a hell of a lot faster 2/3 year old tech while ok now is going out of date.

in software mode my CPU usage was 55%/65% on the processors, in hardware mode it was ~ 15%, so yes it WAS taking the workload off the cpu however it wasnt doing it any faster, infact it was indeed slower,

Pottsey has still NOT commented on the fact the PPU was slower in hardware mode, why not ?

To me the tech is now old, in crysis I think physx would actually cripple under the work load to be honest, a quadcore would be ideal, more cache than mine and can be clocked faster, plus more cpu cores.

Im not getting rid of my agiea pci ppu, as im an avid UT3 addict, but next gen game, current PPU tech does not have a place, - TOO SLOW

Now dont quite its internal 2tb malarky at me, in the real world its falling behind current CPU's, END OF, as seen in real life by me, thanks.


It wasn’t the best video. The problem is the nuke blast is very tiny, doesn’t affect building or trees. When you make it effect tress and buildings and/or change its size then it turns to a slide show.
So the big question is can the CPU problem be fixed and can 100% utilization spread on all four cores do the effects at decent speeds?

In current MUCH LESS COMPLICATED PPU enabled comparisions as I have just said above, the PPU was SLOWER than the cpu, so putting it on current PPU's would be pointless
 
Last edited:
Yea it does I think, it needs gpu more the cpu. Im not sure why thats a problem, Im interested to know if it uses 4 cores or not but it will use 2 cores ok.

People are disappointed it doesnt use the cpu more for physics effects?


I guess I will mess about on the editor and see if it can max out cores. Maybe the full game will have such a level
 
The answers are all over the web for you and in the CEO's interviews.

The Game does not seem to support Quads fully and he has admited it will run the same on a Dual as a Quad last week.

The Game was written to be CPU Dependant, so if anyone is happy it aint thats not good as thats why its probably running crap as CPU usage should be high (something in code broken ?).

Even if a Game is GPU Dependant, it all depends on what it actually uses the CPU for (ever run Prime95 or Orthos and you can still surf the web and do other things instantly, but other apps that use 100% slow your PC right down and you lag to open a new window).

F.E.A.R is a GPU Dependant game, I bet if you Alt-Control-Del out of game the CPU use on a Single Core or Dual Core will be higher than some of get in Crysis.
 
errm, isn't the entire fact all round that it just doesn't matter, crysis being one of the few games with editors where you can just make a silly unplayable useless level with lots of twisters and nukes (as long as you make nukes bigger than they are designed to be) then you can have a level that needs more cpu/ppu power. but, so what, the game won't have that problem, any maps you make that you want to "play" and not just demo won't have the problem, no games have the problem.

who wants to play a game where every level, in the most advanced game about, has so many twisters that you can't actually move around and gameplay is crap, or have nukes so big that they can't be used without killing you. seriously, what would be the point.
 
“Pottsey has still NOT commented on the fact the PPU was slower in hardware mode, why not ?”
I did ask we didn’t bring Ageia into this thread. But as you’re insisting. I did comment before, what I said was I have no idea about your odd tech demo results but as games work I wouldn’t worry about it. Below explains why it might be slower.




“In current MUCH LESS COMPLICATED PPU enabled comparisions as I have just said above, the PPU was SLOWER than the cpu, so putting it on current PPU's would be pointless”
You have a few odd results and your acting like everything’s like that. Try it in a game not a tech demo that’s not even made for testing speed. Try UT3/GRAW2 its faster with a PPU then with the CPU.

Two things you not accounting for. First in the demo you put 100% of all CPU cores on rendering the physics. In a game that’s not possible you need the CPU to do AI, sound, network and everything else. The results change massivly when you go from 4 cores on physics to 1 core or less. Next you used one demo to show off a physics effect that’s not designed for speed comparisons. Its nothing at all like a real game.

Didn’t you say your self the benchmarks and games where faster it was just the cloth effect demo that was slower?.
But that’s not a benchmark or like a game.



“It was a good chunk of FPS faster in CPU mode (~15/20fps), apart from cloth trearing was a couple (1 or 2 fps faster) in hardware mode,”
So when there was no high end physics just a still cloth the CPU was faster but when the tearing and physics started the hardware was faster? That and as above in a game you cannot put 4 full cores onto physics the CPU has to do other stuff.




"Now dont quite its internal 2tb malarky at me, in the real world its falling behind current CPU's, END OF, as seen in real life by me, thanks."
Is UT 3 slower? Are your games slower as well? Whats slower apart from the one demo I explained. I am sure yous said your benchmark program was far faster last time.
 
Last edited:
The Agiea SDK ins't optimized for CPUs coz they want people to buy the PPU. I'm getting 24fps with 1250 ridgid bodies on screen. Would be a LOT higher if it didnt only use 40% of 2 cores and 0% on the other 2
 
Last edited:
“The Agiea SDK ins't optimized for CPUs coz they want people to buy the PPU.”
That’s not true, 99% if the business and money is from the API running on the CPU. They would be extremely stupid to not optimized for the CPU as it would hurt them more. Its clear the SDKis optimized by how well it performs and by how many people use it for the CPU only.

Your forgetting they want people without the PPU to use the SDK and API.
 
DX9.

The big problem with having loads of physics objects isnt the CPU, its making them look good with shadows and stuff. Even if the PPU could handle 10,000 objects, whats the poit if they look crap.
 
Whilst I don't think PhysX cards are the way forward right now, I do have to laugh at those who seem to think multi-core CPUs will negate the need for any such physics processing at any future point.

We used to think the same thing about graphics, that a dedicated processor would never be required, yet look at us now. 3D rendering is a specialist task for which a dedicated processor is far more efficient than a general purpose one. As the quantity and quality of things like physics calculations increase in future games, we may well arrive at a point where a specialist, dedicated processor is required as the load is simply too great for the general purpose CPU. What form such a processor would take is obviously debatable.
 
Whilst I don't think PhysX cards are the way forward right now, I do have to laugh at those who seem to think multi-core CPUs will negate the need for any such physics processing at any future point.

We used to think the same thing about graphics, that a dedicated processor would never be required, yet look at us now. 3D rendering is a specialist task for which a dedicated processor is far more efficient than a general purpose one. As the quantity and quality of things like physics calculations increase in future games, we may well arrive at a point where a specialist, dedicated processor is required as the load is simply too great for the general purpose CPU. What form such a processor would take is obviously debatable.

Problem is, physics aren't as demanding or interesting to develop as graphics, graphics are huge selling points for the entire indurtry, physyics are not.
 
That’s not true, 99% if the business and money is from the API running on the CPU. They would be extremely stupid to not optimized for the CPU as it would hurt them more. Its clear the SDKis optimized by how well it performs and by how many people use it for the CPU only.

Your forgetting they want people without the PPU to use the SDK and API.

You saved me writing a long reply, if its optimized for both, the cpu is still faster
 
The problem with PPUs were:

(A) No game made use of the physics features that couldn't be done in real time on the CPU
(B) No game made heavy enough use of primitive physics to tax most CPUs

Trust me if a game put in 10,000+ primitive physics objects and made use of cloth physics and fluid dynamics then no CPU currently out would be able to run the game in realtime without a PPU.

Enabling the nuke in Crysis with proper physics and a large radius would not run properly even if it was properly supported on a quad core.

Just because a core/CPU isn't being 100% used, all the time doesn't mean its not being used properly.
 
Problem is, physics aren't as demanding or interesting to develop as graphics, graphics are huge selling points for the entire indurtry, physyics are not.
Not at their current level of complexity perhaps, but if and when they evolve it may come to that. Remember, simple computer games had no need of 3D accelerator cards until the games became complex enough to warrant them. Current sound cards offload a lot of processing from the CPU. Who's to say the same might not happen to physics and other advanced environmental modelling in games?
 
The problem with PPUs were:

(A) No game made use of the physics features that couldn't be done in real time on the CPU
(B) No game made heavy enough use of primitive physics to tax most CPUs

Trust me if a game put in 10,000+ primitive physics objects and made use of cloth physics and fluid dynamics then no CPU currently out would be able to run the game in realtime without a PPU.

Enabling the nuke in Crysis with proper physics and a large radius would not run properly even if it was properly supported on a quad core.

Just because a core/CPU isn't being 100% used, all the time doesn't mean its not being used properly.

I agree that a PPU would be needed for 10000 objects and fluids and cloth. The point is that the CURRENT PPU wont do any better than a good duel core CPU.
 
I agree that a PPU would be needed for 10000 objects and fluids and cloth. The point is that the CURRENT PPU wont do any better than a good duel core CPU.

The current ageia PPU can handle thousands of primitive physics objects and cloth and fluid effects fine while even a quad core won't...
 
i was wreading in a mag this month that theres been things happning in the world of rendering and that an old type of rendering previously unusable because there wasnt enugh cpu power is now allmost possible with the use of multicore cpus.
I cant remember the exact name for it but basicly from what i understood the cpu becomes the graphical workhorse and by all accounts is supposed to be very effective, what im trying to say is that maybe cpu's wont take up the rold of PPU's maybe they will take up the role of CPU/GPU and then bigger more advanced physix systems/cards can be implamentad as a standerd to replace the graphics card also carrying the graphics ram?.
years down the road i know but isnt it a possability?
 
damian you're probably talking about raytracing. It won't happen for quite a while and even then there will probably be dedicated raytracing cards if it ever takes off. For the next few years any results from raytracing won't be able to outdo the visual quality of current poly based rendering methods while maintaining the same performance.
 
Back
Top Bottom