Performance and Physx

I have been reading a lot in the press latley about these Physx drivers Nvidia had released as beta, which as we have covered before were not given to futuremark so they are not valid as benchmark drivers.
My question is. Who cares and why? 3dm and Vantage are synthetic benchmarks and its all down to how the card handles actual games in the real world at the end of the day. As these drivers do increase performance in the tests i have seen why would anyone not want to use them ? Im guessing ATI are working on a version ? Also i cant see how these drivers can be kept out of things even if futuremark say they are not official (im guessing Nvidia will submit a set soon). The amount of performance increase is there and reviews will have to at least include then for comparison wouldnt they ?

Your thoughts and any info please.
Mactronix

The problem is, they decrease fps in actual games. Most games are gpu limited, not cpu limited. It's better to have physics calculations done on cpu that is mostly idle instead of the gpu that is already working high load. When you're running a game, it doesn't matter if that quad core is working at 10% or 30% load, it won't impact frame rates. If you keep that already underused cpu idle while tax the already busy gpu, frame rates will decrease.

In other words, Gpu based physics calculations only improve synthetic benchmarks, while tanking real world performance.

this graph is misleading......becoz it dint show fps when physX is off......upon turning on physX, ur fps will take a hit of upto 40%...

So am i getting this right then ?
The Physx reduces the load on the CPU. But for a top end system that needs the card running at its best it causes a reduction in FPS at high res, because obviously the card is doing more. So that leads me to thinking that at lower resolutions or on a machine with a highend GPU and a lower than ideal spec CPU this would help reduce any CPU restrictions in the system. (bottlenecks)

So long story short Nvidia are billing these drivers as something they are not ? or is it the industry that has been hoodwincked into doing it for them ? I have seen reviews like the firingsquad ones that show increases in the performance and i have seen magazine reviews that are getting upset about the increase and are sugesting Nvidia are trying to cheat in the FPS wars again, 3dm Vantage etc.

Again your thoughts or any info apreciated.

Mactronix

Err, such as?

It works

Ah! I really fouled up that one.

I forgot Nvidia physx is artifically made multithreaded to the extreme in order to run on gpu. In that state, it would be like running graphics, which is extremely multithreaded in nature, on cpu.

The light cpu load added applies only to original, unaltered physics caluculation, like those used in Crysis, based on engine developed in house. To be honest, blowing up a shed in Crysis and see a hundred differently shaped individual pieces falling looks better than bumping a single solid box in UT3. It's hard to notice that UT3 even use physics calculations at all if you're not careful.

Besides, the fps hit physx takes on gpu performance can be significant. While cpu load with generic physics calculations does not affect fps.
http://www.guru3d.com/article/physx-by-nvidia-review/4

Didn't nvidia's old architecture prevent it from utilizing DX10.1? Wasn't that why they bribed Microsoft to push the optimizations from DX10 to DX10.1?

In that respect M$ really ballsed up, by allowing the DX10 spec to be what it was and not what it should have been they not only shot themselves in the foot by making Vista undesirable Performance/FPS wise to gamers. They also held up graphics development being passed on to the end user. In their haste to rake in the bucks they released new OS with half arsed graphical support and Nvidia were laughing all the way to the bank.
If they had waited and left things as they were meant to be who knows where we could be now. I know personally that having seen the tests with DX10.1, if it had been that spec that was the Vista release performance i would have bought in to it.

Mactronix

8-12 fps more? And you do realize that like most games, Crysis use generic physics engine instead of Physx, right?

I havnt heard of such increases from this driver, tho Ive heard its good. Oh, and yea, it doesnt have anything to do with the physx either, as well as it being run in Crysis on physx. Daggers right.

Unfortunately. I wish it came for free, but thats not the case

That it is, but the average joe doesnt undertand all that, and thinks its just another part of the game, and no loss. Part of it can be blamed on hype. I noticed the average jill knows better tho heheh. OK, not so average

People get swallowed up in hype all the time. R600, Crysis, Atom, now PhysX. We'll soon see if PhysX lives long after official release.

PhysX runs in 3 mode:
Standard - one GPU renders both Graphics + PhysX (not ideal as you'll need a lot of GPU horsepower).
SLI mode - have two GPUs render both Graphics + PhysX.
Multi-GPU mode - GPU1 renders Graphics and GPU2 renders PhysX.(no SLI board needed.

so i think to get the best performance the last option is best if you really do want PhysX on your computer.

Obviously in game physics is a good thing but I'm not convinced that Nvidia's physx implementation is going to become the standard. I tried the drivers and some of the demos and was seriously underwhelmed. The demo with the water looks like goop. I've seen good in game water (Bioshock dx10) without physx, I've clear cut forests in Crysis and I still recall being impressed with destructible environments in Company of Hero's almost 2 years ago.

While recently deciding on a GPU upgrade physx and Cuda almost convinced to buy a GTX280. In the end I don't yet see a reason for me to need Cuda because by all reports Badaboom sucks. As I previously stated physx is underwhelming at this point. In the end Nvidia's attempt to sell a second rate card on the basis of added features wasn't convincing.

Having said that Nvidia's strategy may pan out, but in the hardware race you don't buy hardware today gambling for future feature support.

I'd be very happy if in the future my old 8800gtx could sit one slot down from my4870x2 and handle physics.

Why should I not want to be impressed? Nvidia's selling point is that a GPU can do better physics than a CPU. What I wanted to see was the added power of a GPU show me something better than what I've seen before. Otherwise their purchase of Ageia and pushing physics support on future titles that may only work with their hardware seems like a lot of waste.

erm actually you may have misunderstood!its meant to do physicx BETTER/FASTER and NOT!! NOT!! do MORE. so you shouldnt see MORE of what you have already seen before,

What are you on about? Where you quoted me I used the "better" twice and never used the word "more". Either way I have no clue what distinction you're trying to make.

The point is once upon a time dedicated graphics became desirable because it brought "more" graphical detail which was "better". Now Nvidia picking up where Ageia left off needs to show us "more" and "better" physics to make a case for Physx. In case anyone does ever offend you by using the word "more" in the words of Nvidia:

http://www.nvidia.com/object/io_1218533603421.html

My question here is: Are they developing with the help of *real* physicists? I mean, physics can get as nasty as your assumptions go of what "real physics" is. Independant of how hardware/software handles the issue, u can have a simplistic way of looking a fall (ex. g.) or a relativistic-too-much-realistic fall wich will only add more junk and have a close result in practical terms. That being said, maybe that's why HL2 "physics" engine is still so good into my eyes also, cause it was polished from the assumptions and up. Having omfg-hyper-complex formulas just to show off is kinda retarded even to point it out ("visual complexity"?). The real brain work goes on the design table of the engine itself, not the hardware that can take care of that engine. nVidia is playing a dangerous game here and focusing on hardware and not the engine itself is going to hurt bad.

Also, how much "thread-ability" does paper physics formulas can get? Are they using relativistic or plain old Newtonian? Well, it shouldn't matter. Physics in games are a joke so far. Maybe for, let's say, AutoCad, SolidWorks or Pixar Studios it is indeed important to have this kind of flexibility, but then again, they don't have to render in real time, just have accurate results.

My 2 pesos.

Cheers!