News Carmack shuns dedicated PPU cards

http://www.bit-tech.net/news/2007/07/19/carmack_shuns_dedicated_ppu_cards/1

John Carmack has spoken out on the subject of dedicated physics cards and it's not good news for Ageia.

 

It's a tough thing to support; everyone has a CPU and GPU so programmers can easily code for those. But adding an optional PPU means it's not going to get used. Ageia should be contacting motherboard manufacturers. Get a lower power onboard PPU in the works.

If more and more people had one, programmers would code for them. Then once loads of engines/games support the physics processor, bang out higher end PCI & PCIe models. Or just leave the current one for sale. Whatever. Real gamers won't use an onboard gfx card, and in the future, they wouldn't use an onboard PPU. But by making it a less optional component, more games would take advantage of it.

I mean why not? If 50% of gamers (obviously workstation mobos won't have a PPU) or whatever have some sort of PPU (even a low power one) game devs will realize, "Hey, we should harness this power, make our games look better, run smoother/at higher FPS etc."

Hmmm, bit of a rant there but I'm sure there are some massive flaws with my plan. And if there aren't, I'll be expecting a cheque from Ageia soon enough.

 

I've just done another review on a Physx title, GRAW 2 PC, and I have to say I'm still not convinved by Ageia. Going to see them tomorrow though, so maybe they can persuade me over to support them, but as it is what I've seen in Cellfactor and other games is either not jaw dropping, or its been handled badly.

 

Then again, why not just code for QuadCores as they're going to be mainstream soon, the same amount (if not more) of processing power is already filtering throughout the market, why should developpers bother writing extra code for something which is already there and might perform better then an integrated/add-on PPU?

I don't see why any motherboard manufacturer would add a PPU anyways, most are already extremely tight in board space, so adding another chip and integrating it into it all would make them lose room for more demanded features that enthusiasts demand.

And really, when you think about it, if you need an additional processor o calculate physics/make them more realistic, how are you going to work that into the game for those who don't have one but still wish to play online with those that do? Because if you make the physics accelerated/calculated by the PPU, then those who don't have one are left out. It'd be kinda unsync'ed in multiplayer if one person's physics are much greater/more accurate then the other guys'. Graphical quality doesn't matter in MP, just affects your framerate, but when gameplay altering effects come into the picture, then you need a global available-to-everyone solution, and if it's dependant on you having a PPU or not.. then it's not hat global anymore is it?

 

Knew there was something I didn't think about! That actually makes much more sense than my idea.

 

I actually can't see me buying a PPU card in the near future but maybe UE3 might change that. Oh, wait... no, I don't even have enough room for another PCI card. P5B-Es are very limited in space if you have a G80 in it.

[offtopic]
What does John do in front of an EA sign?
[/offtopic]

 

The problem with Physx hardware is that their software is so good as distributing physics load for multiple cores (this in essence is what a Physx card is, multiple dedicated cores I guess not unlike Cell SPUs) it makes there hardware a tough sell to the people with heavyweight machines who'll have all the cores.

Alot of the games that have made good use of multicore CPUs like GRAW/R6V/CF/CoH etc are using the Ageia API, strangely not all of them benefit from the hardware despite using the API.

Though Carmack also said that the these new multi core consoles are shite and is now doing engines for them

 

Damn the extra pci cards!

 

what about the GPU solution.. how hard is that being pushed? I had high hopes that one of my 7800's would become my physics card.

 

I generally bring this up when people start talking about PhysX.

Folks should read this toms hardware article.

In particular, look around page 7 -- according to TH, when they managed to enable fabric and fluids simulations without the PPU running in CellFactor, and approached the pertinent parts of the game environment, the framerates would plummet and the CPU usage on an AMD Athlon FX-60 would max-out for both cores. With the PPU going, framerates were kept playable.

In other words, for *certain* types of intensive physics, a quad-core processor WILL NOT be the answer. A general-purpose CPU is not optimized for physics the way a hardware physics card is. If you get 2-3 FPS when doing hardcore physics on a dual-core CPU, you'll probably get 6 FPS with a new quad-core...but you'll get playable framerates with a PPU in the right situation.

Of course, the argument is still open as to whether you're interested in the types of physics that require a PPU, and whether that is worth the money to you for a dedicated PPU, and whether game authors should bother with the PPU at all... my only point is to bring up the fact that a PPU is *not* just a regular old CPU with lots of cores. It is specialized hardware (much like a GPU...) with special capabilities not available on a CPU without paying a huge performance penalty.

 

I'm not saying you're wrong, but it's worth noting that CellFactor was designed for the PPU and not for the CPU. It's a bit like Ageia's RealityMark "benchmark" in that respect - of course it's going to perform the best on a system with an Ageia PPU in it because that's what it is designed to do.

Another example of this is the ATI physics presentation I saw at Computex 2006 - they compared a Core 2 Extreme X6800 (then unreleased) to an X1900 XTX with code that was clearly optimised for the ATI card. It simply looked awful and laggy on the Core 2 Extreme because the code clearly wasn't written in a way that it could manage efficiently. I'm sure, with optimisation the same code would run just fine on a dual- or quad-core processor.

Look at Alan Wake for example (search YouTube for Alan Wake IDF demo), vortices are often considered one of the hardest physics calculations to do and Remedy has managed to do just that on a quad-core processor inside the game's engine - no fancy demos needed here.

The moral of this story is that you can write physics code in many ways and if you write it to work well on a CPU, you can bet it'll run damn well on a CPU and not very well on a GPU/PPU. The same can be said for the reverse, too.

 

The optimisation argument is one you can have till the cows come home unfortunately. The problem is that with the power of todays systems, programmers don't have to spend the time ensureing that their codes runs as good as possible, because there is just such a glut of computing power in a modern CPU.

If programmers ensured their code was well optimised then we would not have the issue of needing a PPU to do physics as we could do it well on any current CPU.

 

My 2 cents.

If the hardware is part of a graphics card, that is invariably what it will get dedicated to. The end result will be the gamer trying to figure out if the slider goes to the graphics side or the physics side.

The problem that the PPU should be solving is addressing the specialized advanced math of physics. Mostly, I've just seen kinematics's out of the card though. It would be nice to see some advanced calculus and diff-eq so things like deformation of solids, fluid flows, and energy physics could creep into games.
I forsee the multicore CPUs essentially becoming schedulers that feed the other parts of the system and keep the OS happy too because once you have dedicated hardware, every clock cycle that hardware is idle is wasted power. Just keeping a GPU fed now presents interesting challenges.

But there is plenty of potential for the PPU to be another irrelevant idea. When was the last time you bought a soundcard for its 3D positional sound capability?

The biggest argument I have heard against the PU has to deal with the inability to optimize the 'rendering' of the physics. Currently, they need to be run for a very large portion of the game world, particularly things that are not visible. Graphics cards have benefited from reducing overdraw, but there does not seem to be much on the bale to do the same for physics. There is a lot that can be said on this subject, I wonder what would happen if we had the 'community' of demo and game programmers of the earlier days of the PC.

 

i think its a bit sad really, its a fantastic idea, however its kind of a circle problem

No one wants to program for the PPU because no-one has it
And no-one has it because there are hardly any games that make use of it

its a bit of a shame, because if people use the Ageia physix api, they could run on PPU/CPU, if people have the PPU it will be offloaded, if not it still runs fine
That way more and more games start to support PPU's, more and more people get it, and then they can start programming more towards it

However, havok phisix have remained as popular as ever, and ageia isn't making much headway

However, if i wanted to choose between GPU or PPU physix, i would go with the PPU every single time
Lower power consumption, 1 slot, cheaper

 

That was a picture taken of him at the 2006 E3.
http://en.wikipedia.org/wiki/Image:John_Carmack_E3_06.jpg

 

It's not really about optimisation.. If that was what I implied, it wasn't meant to be that way. If the code is written for a certain use (i.e. CPU, GPU or PPU), it'll obviously perform the best when the code is run on that hardware.

 

I think that there will eventually be a market for PPUs, as at one point there was no market for GPUs. It will just take time, and may finish off Ageia in the process.

 

I was and remain convinced that the price was the biggest limitation - given a choice between 6-10 games or a card that might improve a couple what will most people do?

On reflection I think they couldn't have found a worse hardware market to launch into either, with CPU and GPU prices being very high a lot of people will have decided that the core parts were already costing them enough without adding another expensive and unknown component.

I think if they'd launched at around £50 or so with a few great demos/games/apps to show off they might have seen mass takeup much as happenned with voodoo 1 cards back in the day, but at £150 with one poor demo .. well we know how that went.

 

I totally agree with you there...granted, TomsHardware was a "neutral" party in the review I linked, but you're right that CellFactor was produced by/for Ageia to demo the PhysX chip. So, as you said, there's no telling what they might have done to ensure that it ran well on PhysX and poorly on a CPU.

Just based on similar arguments of GPU doing graphics better than a CPU, I'm willing to believe a PPU will have similar benefits for doing physics...but at this point, there's no real way to know that (and therefore no real reason to go buy a product) for most people. Until somebody completely independent writes some software to test on a PPU versus a quad-core CPU, we won't really know. Even Alan Wake, as you mentioned, runs great on a Quad Core -- but I'm left wondering how much more (or less?) could they have done if they used a PPU instead (perhaps paired with single, dual, and quad-core CPU's in different permutations, too).

Bla bla bla -- I guess my point is that, at this point , consumers don't really know if a PPU does anything special or not. In this situation, it's terribly difficult to get a product off the ground.

 

100% agree here. What if they had one killer game to give the card a rocket-like start AND made the card cheap enough to actually turn some heads? I might have considered one back when first rumors occured. But after I read several tests and saw the pricing...

I don't really see them sell more than a handful cards as long as they're expensive and (IMHO) useless.