Mirror's Edge PhysX Performance: PPU vs GPU vs CPU

[Rroff]

LOL

 

[Final8y]

Exactly! Physx as it stands is a joke

 

[Lightnix]

I dunno, our systems analysis and design tutor always told us the client doesn't know what they want! Or that might've been web design. Either way!

 

[Final8y]

I dunno, our systems analysis and design tutor always told us the client doesn't know what they want! Or that might've been web design. Either way!

Its not the same.
As what were are being told is not what we are getting.

 

[Stanners]

Burn all 3d acceleration!! i tell you nothing will come of it

 

[HangTime]

Realism wasn't the issue with those cloth effects (which were very nicely done in splinter cell - and it was quite an old version of splinter cell) the problem was you could only use a very limited number of those affects active in any one scene before performance plummeted from nice and smooth to single digit framerates. With hardware physics you can have say 20 of those effects running in a single scene and it will still be giving you smooth framerates.

Well, in a way realism was the issue since by your own admission, there is a constant tradeoff between realism and performance. You probably could have had 20 of those effects in a single scene in Splinter Cell if the level of realism had been reduced.

In my opinion you get severely diminishing returns with increased realism. Put say a dozen of those meshes in a room and it looks very impressive (hence I remembered it all these years later), it sticks in the memory and has that "wow, cool!" factor. But do we really want to have a whole level full of them all being processed in real-time considering that the player probably won't be able to see all that many of them at the same time? You talk about how we can make the whole world much more realistic, but in fast paced shooters, that's a lot of processing being done for only moderate benefit.

If you look at games like the Call of Duty series, what they excelled at was channelling the player down a specific path, thus allowing them to utilise a lot of impressive scripted scenes which had very high visibility (unavoidable in many cases), i.e. very little in the way of wasted development time or processing power. You focus on making the key elements of a scene look good and don't worry so much about the side issues and independence of the game world. I guess what I'm talking about kinda falls into the 80:20 rule.

Obviously in an ideal world we could have a completely 'organic' sandbox where every last detail is subject to a whole host of processing, but that isn't realistic with current hardware limitations (outside of small tech demos I mean - I'm talking about having huge game levels like we know and love with that level of detail).

 

[Rroff]

Yeah but increasingly with newer games you have to give the player greater freedom and your playing with a lot more different types of objects than traditionally... rather than write highly specific routines for handling each unique type of object its better to have a unified physics system that you can just plug objects into and they will be affected by a unified set of realworld rules... its a lot more performance intensive granted but in the end this is the only way to go.

You couldn't just take those one shot effects from splinter cell and apply them to a totally different object and it would behave correctly, you'd have to spend time implementing a specific way that object was handled.

Its not just all about visual niceities either - its getting to a point where software physics even at low accuracy just can't handle all the objects you want to put in a game and give good performance.

 

[HangTime]

I can see where you are coming from in terms of having a 'one size fits all' solution and the benefits that would bring. However I think part of the problem is that humans have so much inbred expectation of how certain types of object will/should behave that in some cases it's probably just as well they have 'bespoke' physics applied to them, rather than some crude, half-hearted generic system. For example ripple effects on water, I imagine that's probably fairly hard to accurately simulate, and that a 'hardcoded' ripple effect triggered under certain circumstances probably actually looks a lot better than most attempts at doing it 'properly'.

Obviously the more objects you decide you want to have physics applied to, then the greater the argument for a universal system becomes. I just don't think we've reached the point in gaming development yet where there is enough weight behind that type of requirement.

 

[Rroff]

I can see where you are coming from in terms of having a 'one size fits all' solution and the benefits that would bring. However I think part of the problem is that humans have so much inbred expectation of how certain types of object will/should behave that in some cases it's probably just as well they have 'bespoke' physics applied to them, rather than some crude, half-hearted generic system. For example ripple effects on water, I imagine that's probably fairly hard to accurately simulate, and that a 'hardcoded' ripple effect triggered under certain circumstances probably actually looks a lot better than most attempts at doing it 'properly'.

Obviously the more objects you decide you want to have physics applied to, then the greater the argument for a universal system becomes. I just don't think we've reached the point in gaming development yet where there is enough weight behind that type of requirement.

No we haven't reached that point yet... but development is quickly converging on that point and probably would do a lot faster if there was a viable cross platform hardware physics system. Certain niche games are already running into the limitations and it won't be that long before it becomes an issue for more mainstream titles.

 

[drunkenmaster]

Whats really funny is this, apparently Intel will be making the next Playstation GFX(maybe cpu aswell), intel, who like Havok . Ati almost certainly have both the xbox and the Nintendo, opps, more havok. So this massive market saturation, despite the cross platform-ness of the market and no PPU's available for Ageia in the next gen consoles.

Ageia is dead. Even the software API won't be worth it, if either Intel or AMD add in acceleration for Havok will make not using even the havok api pointless as all hardware available will be able to accelerate Havok, and not Ageia.

This comes today as rumours spread of Nvidia bringing in lots of x86 designers and wanting to build a CPU. My personal take on it is, they've just lost console sales, their desktop sales are going down a lot ATI were making up market share and Nvidia took a what hit of what, 15mil PS4 gpu's to be sold in the future, if not a lot more. Me thinks they are going to try and put together their own console so they aren't left out, but chances of a success in the current market and with by then maybe everyone favouring intel/amd software wise, almost non existant. They also don't have an x86 licence, which both AMD and Intel own who wouldn't remotely sell them a licence to build them.

Ageia is dead, it was dead before it started, they only wanted to make people interested enough to persaude someone to buy it then they can retire happy and leave whoever buys it to watch it die.