AMD to demonstrate GPU physics using Havok at GDC?

[Eastcoasthandle]

Bit-tech has reports that AMD might announce the use of gpu physics using Havok during the Game Developers Conference. Furthermore, Terry Makedon provided a tweeter feed announcing ATI's GPU physics strategy which may include a demonstration next week at GDC on 26 March. You can read more about this in the links already provided.

Edit:

Yup it is. Havok is indeed our partner of choice. Go check out the session if you are around, should be educational.

twitter

[perkam]

Havok also seems to be Intel's partner of choice

Perkam

[Zaskar]

its just a shame that the physics controlled objects cant be anything of importance in multiplayer games as each system will place them a bit differently.

[jagass]

Thanks for the info...

[w0mbat]

Haha, Nvidia just got owned.

[RaZz!]

its just a shame that the physics controlled objects cant be anything of importance in multiplayer games as each system will place them a bit differently.

it did work quite well in hl2:dm which is havok based, didn't it?

[generics_user]

nvidia just screwed their own API by not giving it to AMD

havoc is present in much more games than physx is and has a larger userbase

smart move by intel to allow havoc acceleration on amd cards as it'll kill off physx

[Sumanji]

This would be a good dose of AMD pwnage, if it is a) true, and b) effective.

[FischOderAal]

nvidia just screwed their own API by not giving it to AMD

Yeah. Last thing we need is a physics-war... I sure hope this Havok-GPU-Physic will use OpenCL

Offtopic: Tachjen Auch hier?

[Calmatory]

AMD refused to support PhysX since it is more or less proprietary software unlike Havok, which is open as in free for all. Same with CUDA vs. OpenCL, Nvidia wants to keep it closed, AMD believes in Open Source.

[Zaskar]

it did work quite well in hl2:dm which is havok based, didn't it?

Not really, all the rag doll bodies and stuff like that are client side so they can be positioned by the clients physics processing, and because of that their not placed the same as on other peoples screens. It was for that reason that one of the TFC mods couldn't have the spy class fake death, due to the ragdoll client controlled death physics.

[Manicdan]

but things like barrels being thrown around by grenades is really important due to damage being done. i would think that could be done by the server just fine.

[Calmatory]

One can't really put that stuff server-side either, then the server would need accelerated physics to calculate the position.

[ILikeCosmosS]

+1 NVIDA got owned

[Hornet331]

Havok also seems to be Intel's partner of choice

Perkam

Intel owns havock.

[Telperion]

Not really, all the rag doll bodies and stuff like that are client side so they can be positioned by the clients physics processing, and because of that their not placed the same as on other peoples screens. It was for that reason that one of the TFC mods couldn't have the spy class fake death, due to the ragdoll client controlled death physics.

As someone who extensively played HL2DM, and racked up many kills with saw blades, barrels, grenade cooking, etc, I can safely say that physics played a significant role in multiplayer.

[DarkerMark]

As someone who extensively played HL2DM, and racked up many kills with saw blades, barrels, grenade cooking, etc, I can safely say that physics played a significant role in multiplayer.

I second that and anyway when mapping for hl2dm (and I suppose any other scource game) you can define for any prop if its physics are calculated server or client side.

[ColonelCain]

Physx in multiplayer is important, but what Zaskar is saying is that it cannot be used much with environments getting much more elaborate (more things to destroy).

The real question, as others have said, will be whether or not it is effective.

[[XC] Lead Head]

If you want to see physics based multiplayer, play Garry's Mod.

[570091D]

i really don't like the havok physics engine, i guess it's better than nothing, but it feels like all objects under the control of havo have the weight of a cardboard box. to me it feels very cartoonish, and i understand the point of a phisics engine is to increase the realism of a game.

but how is this news? didn't ati/amd announce havoc support a while ago? why has it taken so long to adapt to stream?

nvidia offered ati use of physix and cuda at the same time but ati declined. that would have given everyone a single physics and gpgpu standard. it doesn't matter now, we have openCL, let's hope everyone starts using that.

[trinibwoy]

AMD refused to support PhysX since it is more or less proprietary software unlike Havok, which is open as in free for all. Same with CUDA vs. OpenCL, Nvidia wants to keep it closed, AMD believes in Open Source.

Geez people can't possibly be this misinformed. Havok is free and open for all? Havok is just as proprietary and closed as PhysX my misguided friend.

[Helmore]

i really don't like the havok physics engine, i guess it's better than nothing, but it feels like all objects under the control of havo have the weight of a cardboard box. to me it feels very cartoonish, and i understand the point of a phisics engine is to increase the realism of a game.

but how is this news? didn't ati/amd announce havoc support a while ago? why has it taken so long to adapt to stream?

nvidia offered ati use of physix and cuda at the same time but ati declined. that would have given everyone a single physics and gpgpu standard. it doesn't matter now, we have openGL, let's hope everyone starts using that.

The problem with CUDA is that it is designed to work on a G80 type design specifically and that is not what AMD is using. This would make it very hard for AMD to get reasonable performance out of their cards when used with CUDA. One other problem, you are always dependent on some other company on which you have pretty much no influence for improvements to CUDA. That would mean that when NVIDIA comes with CUDA 3.0, AMD will only be able to start implementing it after NVIDIA has introduced and AMD will be lagging behind on CUDA feature support. This won't be the case with a more open standard like OpenCL.
Now I'm hoping for this to be running on OpenCL, that would be really awesome.

[B.E.E.F.]

Geez people can't possibly be this misinformed. Havok is free and open for all? Havok is just as proprietary and closed as PhysX my misguided friend.

However with Havok, if its just GPU based (without all that proprietary junk) you can use it on any GPU, rather than the ones licensed by nVidia.

Lame move by nVidia.

[trinibwoy]

Sigh, there is currently no standard for GPU compute. OpenCL still isn't ready to go. So what lame move are you referring to exactly? If AMD and Havok are cooperating on physics middleware it's either OpenCl based or proprietary to AMD. And if it's OpenCL based it will be highly optimized for AMD's architecture. So again, just as proprietary as physx. Opencl isn't like x86 where all the hardware on the market is basically the same. It's still possible to target a specific vendor's hardware.

Helmore: OpenCL is CUDA in a shiny new package. There's nothing in it that makes it more suitable for AMD hardware. When coding in OpenCL there will still be very architecture specific optimizations that will not transfer well to all hardware out there.

[Helmore]

Sigh, there is currently no standard for GPU compute. OpenCL still isn't ready to go. So what lame move are you referring to exactly? If AMD and Havok are cooperating on physics middleware it's either OpenCl based or proprietary to AMD. And if it's OpenCL based it will be highly optimized for AMD's architecture. So again, just as proprietary as physx. Opencl isn't like x86 where all the hardware on the market is basically the same. It's still possible to target a specific vendor's hardware.

You should not forget that Havok is now owned by Intel and I don't think Intel would like to see such hardware specific moves for hardware that is not its own. Especially when Intel would like to promote Larrabee as an excellent physics processor.