AGEIA VP responds to my further Queries about Slides

[Bohdy]

If I were a mod I would close all bar one Ageia topic, and ban everyone still arguing about them.

Unfortunately I'm not.

*posesses Vysez with psychic rays*

"Or am I?"

 

[Titanio]

Those Ageia guys seem pretty confused. If they just make their code run on the hardware without optimisations for each platform I don`t expect great performance on both XB AND PS3.

What makes you think they don't optimise for each platform?

I think Titanio already sent an e-mail to Andy Keane addressing those issues.

I didn't actually ask about that at all.

And yes, I think we perhaps need one thread.

 

[Shifty Geezer]

onetimeposter : If you're pestering Aegia under the guise of a TeamXBox correspondant or some other media respresentative, you'd better damned well be for real. They're a business who shouldn't be being pestered by non-entities for information. Non-entites should wait for the real media representatives to ask the questions and get the answers. I hate to think how inundated Aegia are if everyone who took the urge emailed them.

Regards the reply I don't see that it's any different to what we already know. No, there are no benchmarks, but the point

We do have engineering code that tests operation of the software, but these are not benchmarks.

means they do have FPS results for simulations that they have run. I wonder if they'd be willing to share this information and prove once and for all that XB360 runs the simulations as well as PS3? Another couple of points that are inconclusive...

1. ExtremeTech reported what they heard and saw. If an AGEIA person stated that the Xbox360 has technical limitations that prevent the Xbox360 from running algorithms in our SDK, then that statement was not correct.

No-one's saying the XB360 can't run the algorithms, only suggesting that it can't handle them well enough to include them in most game situations.

2. We did infer in a slide that the PS3 has more resources that could be dedicated to physics. This was an error since this statement needs real data as backup.

What was the error? That the PS3 has more resources, or that they said that on a slide with recorded evidence to prove that?

Any conclusion on relative performance was built from specifications released publicly by the console vendors. As I stated earlier, the performance of physics in a game is dependent on so many factors and resources.

The specifications include architectural designs. Did this offending slide just look at public FLOP figures and say PS3 has twice as many, or do the guys at Aegia understand that peak FLOPS is a pretty meaningless figure and instead looked at the architectures of the processors and see the Cell design is more accomodating of their engine?


The response is once again inclusive.

The questions that would prove one way or another would be...
1) When you ran the Novodex test cases on the different processors, which performed best and by how much (in terms of frames per second)
2) Outside of a game environment, if the two machines were dedicated to running physics simulations, which platform would you expect to handle this better.
3) What is the architecture of the PPU and how does it compare with the XeCPU and Cell?

Of course Aegia will be smart and not respond to this with conclusive answers. And rightly so. At the end of the day
1) If Cell's better at XeCPU at physics, so what? It's been designed to be and everyone with any sense is expecting it to be. XeCPU can achieve as much as Cell in every field and in less transistors, that's a damning inditement on the capabilities of IBM, Sony and Toshiba in designing a processor.
2) Even if Cell is better at physics, will it really have a big impact on the games?
3) Even if Cell is better at physics, what is it not as good at than XeCPU and how will that ipact the games
4) Why the hell are people trying to prove one machine better/as able as the other? This is supposed to be a technical forum discussing technical solutions to problems. There's been very little consideration of technical approach to physics simulations. Just a bunch of people trying to prove/discredit statements and each others creditability in communicating responses. This discussion is worthless regards the ideals of the forum when it's just fanguys trying to descredit each other POVs.

PS : Quoting with Bold messes up formatting. This isn't my fault.

 

[Jawed]

Shifty, plenty of people have jumped to the conclusion that certain Novodex features are unavailable in XB360. Fluid simulation comes up time and time again.

What on earth makes you think they have FPS results for the SDK on different platforms? A regression test is not a benchmark.

Having said that, I can't believe that they have no benchmarks at all. On that point I think their obfuscating their position, overly defensively. You don't write performance-sensitive middleware without having measurements of performance.

This would be like saying the graphics cards companies don't use benchmarks in their driver validation and driver-feature prioritisation.

Jawed

 

[Titanio]

means they do have FPS results for simulations that they have run. I wonder if they'd be willing to share this information and prove once and for all that XB360 runs the simulations as well as PS3? Another couple of points that are inconclusive...

They've clearly said they won't, and won't ever in the future. They claim PS3 and X360 are not their products to benchmark. They even question the usefulness of benchmarks for physics in games (seemingly ignoring the usual "all else being equal" qualifier that can be applied to these things). Yet they bench PhysX vs Intel chips with their own demos as "proof" of something with regard to the PhysX. So I think there is clearly some "special treatment" being applied to the consoles, probably for political/diplomatic reasons i.e. so as not to ruffle the feathers of a partner or partners.

The specifications include architectural designs. Did this offending slide just look at public FLOP figures and say PS3 has twice as many, or do the guys at Aegia understand that peak FLOPS is a pretty meaningless figure and instead looked at the architectures of the processors and see the Cell design is more accomodating of their engine?

SenatorMonkey's report on the explanation of suitability of the various platforms for FD etc. suggests they've considered how algorithms will map to the hardware in a much more comprehensive fashion than simply looking at floating point performance.

 

[Shifty Geezer]

What on earth makes you think they have FPS results for the SDK on different platforms?

They've provide info that on a single Intel CPU they get, with a simulation, 4-6 fps, and on a dual-CPU + PPU they achieve 40 fps. They will have have simulations that they use as test cases, the demos for example, just to see their product actually runs. Let's say the soap demo with the car. They would have run that demo on single core CPU, PPU, XB360 and PS3, and seen for themselves whether it's a smooth frame rate or not, and seen the little counter in the corner measuring FPS that's oft included to see how it''s performing. It may not be an official test, but you can be sure the developers of the physics engine on XB360 and Cell would have tried different demos out just to see their engine working. Okay, I have no conclusive proof and I accept as such. But it's beyond my comprehension that thse guys will write a library of functions and algorithms and NOT use example code to test them, NOR use a uniform example program across all platforms to test portability, and fail to have figures for basic performance metrics across platforms.

Comparing it to something like ODE or the old Tokamak, there were never benchmarks to measure the engine's performance. There wasn't a 'this implementation of a physics solver rates at 7.3 wiggatrons performance'. You'd pick up the engine and implement it and see how fast or not it runs. That's inevitable - there are no benchmarks to measure simulation performance and compare engines AFAIK which is why its perfectly true of Aegia to say they have no benchmarks. But I can run an ODE or Tokamak application on two different computers and see for myself how much difference there is, which gives a rating for the systems, and Aegia will almost absolutely certainly have some such information from the results of their own development. How could this not be the case?

 

[Titanio]

They've provide info that on a single Intel CPU they get, with a simulation, 4-6 fps, and on a dual-CPU + PPU they achieve 40 fps. They will have have simulations that they use as test cases, the demos for example, just to see their product actually runs. Let's say the soap demo with the car. They would have run that demo on single core CPU, PPU, XB360 and PS3, and seen for themselves whether it's a smooth frame rate or not, and seen the little counter in the corner measuring FPS that's oft included to see how it''s performing. It may not be an official test, but you can be sure the developers of the physics engine on XB360 and Cell would have tried different demos out just to see their engine working. Okay, I have no conclusive proof and I accept as such. But it's beyond my comprehension that thse guys will write a library of functions and algorithms and NOT use example code to test them, NOR use a uniform example program across all platforms to test portability, and fail to have figures for basic performance metrics across platforms.

There are tests you can run to ensure the code is functioning properly that don't relate to performance or indicate performance. But yes, I think we all agree that it's unthinkable that AGEIA would not be testing performance on these systems, and wouldn't be trying out all those same demos either. The tests that don't indicate performance, regression testing etc. simply give the functions or whatever a workout to make sure they're throwing back correct results, but that doesn't matter a damn if your performance isn't there, optimisation also relies on performance feedback. Doubtless they're doing performance testing, they just want to completely sidestep any question of relative performance on these systems now to keep everyone happy and by saying "there is no data", that's the best way to do so. I think if they said "there is data", by not subsequently sharing it it'd look like they're trying to hide something, which could carry implications in light of their statements at these conferences.