Hardware AMD Talks GPU Gaming Physics

We discuss gaming physics with AMD's Manju Hegde, former co-founder of Ageia and Nvidia CUDA VP.

http://www.bit-tech.net/hardware/graphics/2011/02/17/amd-manju-hegde-gaming-physics/1

 

Interesting read. I hope both companies step up the plate and embrace the OpenCL platform very soon. Otherwise it will be the past all over again like for example the BluRay HD-DVD war. All the money spend on nothing that suplied the consumer (or gamer) with anything cool. And it looks it is a long way to go if AMD is going to use Bullit and Nvidia sticks with CUDA. Must be a nightmare also for game developers.

 

We definately need an open source code for Physics, until any new gen card can effictively render... well I wouldn't spend the time coding something only half or less of my consumer base can utilize. Physics is one area where PC gamers should easily trump our console counterparts in user experience. Open it up so more devs will jump on board!

 

I can't help thinking that the main reason why any of these engines are not ultilised fully in terms of gaming is because consoles can't handle the complexity of the type of physics PC gaming GPU's are capable of - the the combined sales of consoles if far greater than the PC - what we're asking for is developers to invest more money developing more features for their smallest market.

In short the extra cost required to make a multi platform game and then invest in extra coding/effects for the PC (which is probably the smallest share of games sales Vs (PS3+Xbox sales) is preventing profit driven corporations to anything other than stick with the basics that's easy to transfer to any system.

OPENCL seems to be a common platform for the future, although any hardware thats opencl capable will also be able to adopt bullet or CUDA - and then it will be up to NVIDIAs 'The Way It's Meant to Be Played' Vs AMD's 'Can You Dodge The Bullet' teams to continue investing in future development - with opencl being used when netiehr company wants to invest.

But, until next gen consoles are out - physics effects will probably remain at the basic/novelty level regardless of any new platform developments.

 

Nice article, good read. +1 for open source physics for PC games.

 

I'm no expert in the mechanics of any Physics coding but shouldn't ATi (AMD) concentrate on getting Anti-Alaising working more effectively on their platform. AA appears to be the only thing holding back the 6 series from being a clean sweep recommendation. If AA is part of the Physics coding then apologies for my lack of knowledge. I just group my components together based on their performance without fully understanding how they actually work.

 

Fundamentally the reason we don't get fancy hardware accelerated physics is that most games are console ports and consoles don't support it, until that changes HW physics, like DX10 and 11 will just be an afterthought for most games.

As for bullet, well it has potential, it's used in software form in a few games, but so far there's been nothing hw accelerated in a game. It's a long way behind physx which is in a huge % of games in software and a few in hw. Reading between the lines I take it that "let the developers adjust it" is short hand for "our team is too small to do it ourselves so you'll have to do it if you want feature X".

Also as I understand it AMD are mostly talk. They say they want to support bullet but actually provide the bullet team with very little support - hw bullet was being developed on nvidia I suspect mostly because their openCL driver and sdk support is better. Not sure today but AMD was only releasing the odd opencl driver, it had to be downloaded separately and was only compatible with one version of the graphics drivers (i.e. most of the time not the latest). Then AMD have a history of hw physics trash talk and no action - they were singing the praises of hw havok from 2006 onwards till about 2009 when it became obvious nothing was ever going to happen.

 

sounds like AMD's making sure people hear about they are finally putting into developers. let's hope they actually achieve something this time round. (DbD's recall of history is pretty accurate)

developers *need* support. nVidia was there to provide support, AMD only just started to do so recently, this is why an open standard is not present. all organization is in it to make money, nVidia developed CUDA before OpenCL was finished, so it makes sense for nVidia to continue pushing it, until someone tells them not to. it's the same as Adobe Flash vs HTML5, former is older and is from an organization wants to make money, while latter is a new open standard that currently needs a lot of support.

if the PC gaming alliance actually pull their finger out and push towards an open standard that all companies have to stick to, there will not be such API problem.

 

With Sony and M$ more concerned with selling motion detectors rather than next gen boxes, we could be waiting some time for those.

I suspect, that gpu's moving to 28nm will be a driver to shortening the wait, as in faster, cheaper cooler. Then we'll see better games on PC.

 

Games have slacked off in quality as developers continue making cheap console ports and DX9 games.
Ageia was a scam from the start. $129 card just to do a few physics tricks, then Nvidia took the bait, bought the company and have been trying to recoup their money every since but a few particle effects and some wavy capes or flags is all we got.

AMD once again is putting gamers first and has the right idea with OpneCL Bullet Physics. I hope they follow through and really push this cause I feel many developers will use this Open software to really push DX11 games without having to pay Nvidia for it's bad acquisition of Ageia.

 

What? AMD put gamers first? Are you kidding? AMD, just like Nvidia, are about making money. If anything Nvidia are doing more for PC gaming because they actually support developers.

 

until gaming consoles can handle physics calculations, i don't want any company investing time and money into it because pc games will never get the acceleration. amd needs to spend money on developing their video drivers, because their video drivers are ALWAYS crappy. if you buy a brand new ati/amd video card, within the year you get significant performance improvements just by driver updates alone. when you gain as much as 15+FPS just because of a driver, you know their software developers haven't been doing a good job.

in linux, there are times where cards like the HD46xx can outperform the HD 58xx. thats a problem.

also just so everyone knows, theres actually a massive handful of games that use physx, just not gpu accelerated physx. to me, this is stupid because you still have to install physx for those games, so if the games REQUIRE the libraries, shouldn't they still be capable of gpu-acceleration?
as i see it, the main reason physx never took off is because nvidia got into it a little too late and their terrible idea of preventing ati cards from running at the same time restricted SOOO many users. i've used a hack where i could use my ati card with an nvidia card for physx, but i shouldn't have to do that.

 

why doesn't M$ integrate a physx solution into Directx then up to gpu manufacturers to support it or not?

 

If you mean buying out studios/game devs so games will be more 'optimised' for Nvidia cards so Nvidia can use it as propoganda and show how fast their own cards run the game so much better then AMDs then yeah. I suppose you could call that a form of support.

other then that what are Nvidia doing to 'support' game devs? its called competition. im not trying to paint Nvidia as the big bad wolf but weve all heard of the tactics and schemes that Nvidia have used in the past to throw dirt on ATi/AMD

 

Exactly.

 

This is not my usual AMD PR marketing rant, so please don't read it in that tone.

What's more puzzling on why they're commenting on this now, is the massive uncertainty with AMD today. A company running without any permanent management really should not be making comments about which department may do what with products in developement. AMD really is hanging in there and some feel rumoured talks of takeover won't come to fruition due it's huge debt, higher than most chip markers. Any potential new owners would need to also convince their middle east majority shareholders to agree to any purchase.

With only 20% of the market, many big tech names would not buy also because the second someone like Toshiba buys them, Sony Dell HP Acer HP etc would cancel any business with AMD as they wouldnt want to be helping their own competition.

It might sound like one of the AMD departments cry for funding publically since the internal management is in such turmoil.

 

The main problem is that Nvidia have locked out a large chunk of PC gamers, restricting the attractiveness of PhysX based Physics to game developers and AMD can't decide whether to push software based Physics on their multicore CPU's or GPU accelerated Physics or both.

Intel probably have the best of the 3 with Havok based Physics on the CPU, which provides enough for subtle Physics effects a la source engine and doesn't worry about 'eye-candy' physics that don't add to the gameplay in any way.

Over 88% of Steam users now have a CPU with 2 cores or more, which makes CPU based Physics the more viable option (especially with the architectural differences between AMD & Nvidia GPU's).

As for AMD's work with game developers, i would guess it is harder and harder to get PC optimisations implemented on various console ports without throwing a lot of money at it or resorting to some of Nvidia's 'tactics' of the past. Anyone who thinks that they don't work closely with game developers now needs to read some of the various blogs regarding their work on tessellation with Aliens v Predator, their work with DICE on the Frostbite engine (including Frostbite 2) and Bad Company 2 etc.

 

I get sick of the argument 'extra physics doesn't add to the gameplay experience'. Well, if people are only worried about the 'gameplay' experience, let's go back to Quake or Doom or something similar since graphical quality clearly doesn't matter.

IMHO, the whole point of a GPU is to provide eye candy which enriches the gaming experience. So, if a GPU can offer to render and calculate (for example) bits of masonry chipping of in a gunfight, I'm all for it.

Not a dig at you frontline, just a mini-rant!

 

No problem! I guess everyone has a different definition of what the best balance is between great gameplay and visuals. Personally, the most fun over the longest period i've had gaming is on source engine games, where the DX9 visuals aren't the best by a long way, but they aren't poor and the ragdoll physics and debris in something like Day of Defeat Source is more than adequate for a good multiplayer experience with up to 32 players on a server at once.

Don't get me wrong, i love great looking games, but some games seem to have become over-obssessed with the visual impact to the detriment of the overall game experience.

 

When the day comes that we can run games on a PC in much the same way as people already do on a console, insert disk and play with installation optional for those of that prefer to; When consoles and PC's start using common libraries specifically adapted and optimised for alternate platforms and equally capable of rendering useful and practical Physics and not just eye candy we will have gone a long way to something truly useful.

All the time we have proprietary systems out there, no matter the marketing and hype, developers will have problems to solve and extra work.

As for Havok, that is dead in the water, Intel has it and Intel is known for it's onboard graphics being a really poor offering compared to AMD and Nvidia.

Physx should have remained independent and not sold themselves short.