Ageia RealityMark™ (physics benchmark)

[Pottsey]

Talk about a great idea

http://www.ageia.com/physx/rm.html

Don’t get me wrong I am all for physics benchmarks and at least its based on a real game engine. Even though it’s not made by Ageia they should not have released it them self’s. That’s like Nvidia getting a 3rd party to make a benchmark then releasing it under Nvidia’s name. No one would take it as a fair benchmark and certainly not ATI even if it was.

The benchmark is basically a pre-scripted gameplay sequence from CellFactor and it does run on none PPU systems. I have not ran it my self yet and I have no idea how fair or not it is. But I dont see it going down well.

EDIT:

Side by side of hardware V software.

 

[Pottsey]

I think it’s meant to demonstrate that the CPU alone is not powerful enough to handle 1000’s of object with cloth and liquid at the same time. Which is true anyone who thinks dual core CPU’s alone are enough to do high end physics are just tricking them self’s or in denial.

You need a GPU doing physics or PPU for high end physics.

 

[Pottsey]

There are mirrors at hardocp in there news. I cannot access the site at work though, so cannot post links.



“Alan Wake game physics runs on 1 core of a multicore (4 in this case) CPU”
I wasn’t to impressed with Alan Wake physics. There where good but nothing special. The video I watched had no high end physics like cloth or 3d liquids. The amount of physical objects on the screen where no more then today’s game.

Saying that the Alan Wake engine as a hole does look impressive just the Physics are good not excellent.




“And if Alan Wake physics can be run on a single core, and very impressively I may add”
Once people get used to coding for hardware physics on both the Ageia PPU and GPU’s we should see physics far in advance of what is in Alan Wake.

 

[Pottsey]

“ITs kinda like NVidia Releasing a benchmark that runs terrible on ATi tech to make you buy their product!
What a load of rubbish really!”
Not really as the benchmark software is not slowing down when useing software/CPU to make the hardware look better. You really do get those slowdowns on the CPU with cloth and high end physics. It match’s the games.

Its more like Nvidia realising a benchmark with Hardware T&L back when ATI didn’t have Hardware T&L.

Remember Ageia didn’t make the benchmark and the engine is from a real game. So you really do get those results outside benchmarking.

 

[Pottsey]

“odd how cell factor seems to run far faster than what that demo shows in systems without the ppu.”
Are you sure your not confusing the results from the first demo, which ran with cut down effects on the CPU? The first demo turned off effects like liquid and cloth without the PPU. Later demos and this benchmark make the CPU render the same as the PPU with cloth and liquids. Well I think liquids are still cut out from the CPU but cloth is working now and it’s the cloth that really hurts the CPU.





“The cellfactor demo runs quite fast on my non physx pc. I wonder if it'll magically run slow in this benchmark. ”
I bet you ran with cloth and liquid turned off when you tried the Cellfactor demo which would explain the slow down. I have never seen a none physx pc run the demo quite fast with full effects.

 

[Pottsey]

”Unsurprisingly its just ridiculously crippled so it only runs well with Ageia's hardware ”
Have you ever thought for a second that perhaps its not ridiculously crippled and that’s how it is in real life? Games get the same results and there is no evidence its crippled.

Its seems a lot of people are in denial that hardware can be way faster the software on the CPU. Its not surprising it only runs well with hardware CPUs have always been super slow at liquid and cloth.

Why are so many people surprised hardware is faster then software?




“I wonder if the graphics are in software too, like the cpu tests in 3d mark.“
Graphics are fully in hardware as soon as you turn off the high end physics like cloth the FPS jump up. The more you turn down the physics the higher the FPS go.

The benchmark is based off a map in Cellfactor. You can test it your self with different effects turn on and off.

 

[Pottsey]

“I fail to see why software running on the PPU should be hugely faster than software running on a second core.”
CPU only run 1 or 2 threads at once meaning the CPU must do instructions in a sequential manner while PPU’s and GPU run in parallel letting them run lots of threads at once. Phsyics needs lot of threads running this is why CPU’s are not the best choice for physics.

CPU’s just cannot match the physics power on a PPU or GPU. There is no way 1 CPU core is going to beat dedicated hardware at its job.




“The possibility that they've designed this cloth/liquid effect software to run poorly on anything but their own hardware seems much more likely. I see no reason at all why advanced physics effects can't be done on a second core, there's plenty of CPU power available there not being used.”
Look around for fully destructible 3d cloth in games on the CPU only. It not done as it’s not possible at playable speeds. The few games to use none destructible cloth that’s simplified over what the PPU does, suffer slow downs on the CPU. Backing up that Cellfactor is no different from other games. This also proves that cloth/liquid are not made to run poorly on anything apart from Ageia hardware. Unless you think Ageia bribed every developer in the world to make liquid and cloth run slow on the CPU. The same goes for 3d liquid you dont get it at useable speeds on CPU's.

 

[Pottsey]

“Average frame rate numbers for each run are then computed and displayed in a chart at the end. Oddly, though, this chart displays a "default" hardware-accelerated result of 35.23 FPS if RealityMark is run on a system without a PhysX card installed”
That makes no since at all. Every system I have seen gets less then 3fps on average without a PhysX card installed. How on earth did they get 35.23fps did they disable the cloth? Can you give a link to that quote? How does that quote prove the benchmark is heavily biased? If its due to the cloth being turned off all it proves is the CPU is not good enough for full cloth. Also you didnt compare the PPU with cloth to the CPU without cloth? Could that be because as the PPU is still faster?






“So, £180 to see moving cloth? Me don't think so. If agei is in the future then i'm more than happy to stay in the present.”
Its more then just cloth it’s also liquids and a 30 to 60% performance boost. Look at COH that gets a large performance boost and new effects with a PPU. BOS gets nice chemical and other liquid effects as well as a speed boost.

 

[Pottsey]

”Pottsey, if you have a PhysX card, run my little test Here and tell me what your lowest fps is while holding down space. I get about 30fps minimum with out a PPU.”
Sure when I get home from work. Did you have FSAA or AF on? What settings?






“From another site:
"You may wish to look at this.
To summarize: Going to www.ageia.com and downloading the software package increases game performance, even if you DO NOT have the card itself.”
I am presently running with full particles, and full (not recommended without a PhysX card) physics, and I am fully stable. I have not gotten a huge fps jump, but my game is now much, much prettier and a bit more stable.
The Ageia API is for both software and hardware its not surprising new versions give a speed increase. Lots of games only use the API for software and some games don’t even work unless its installed even witout the PPU. Still the PPU is 30% faster then a dual core CPU in that game.

http://images.anandtech.com/graphs/physx performance update city _090506100924/12955.png

Sure you don’t need a PPU to run max settings but you get a nice speed increase.
EDIT Just remembed without the PPU the liquid effects are missing. Need to double check that though. If liquids are on and missing then he is not running on full without the PPU like he thinks he is.

 

[Pottsey]

“Oddly, though, this chart displays a "default" hardware-accelerated result of 35.23 FPS if RealityMark is run on a system without a PhysX card installed”
After reading the original at techreport my brain just went click and I under stand them now. I don’t know why I couldn’t understand it yesterday, perhaps last nights long sleep helped.

Anyway 35.23 refers to the average speed people get with a PPU. So the default hardware-accelerated result of 35.23 FPS is for none PPU owners to compare to there score. For some reason I was reading 35fps as the score the none hardware people where getting. Bit of mistake really. Sorry about that.

Still back to your point. I don’t see why that is biased. All they are doing is giving people with software a way to compare there score to hardware. If they don’t post the average hardware score wouldn’t the software score be meaningless?

 

[Pottsey]

“How many games actually suppourt the card? Even with GRAW quite a few people prefered the game without the physics card.!”
9ish out now, with they say 100 in development although to be honest I expected about 15 to 20 to be out by now.

 

[Pottsey]

With first cloth program I got 100fps on loading the program and 52 with spacebar hold down. FPS where jumping around a lot it was more 97 to 103 and 48 to 55.




“Ok, I have to turn cloth effects off, but £180 for some cloth effects? No thanks ”
Your missing the point of the card. It’s not just cloth effect, its cloth, liquid and a performance boost.