Bullet Physics

Neither does OpenCL.

But regardless of this, why do you think it's taking so long for functioning OpenCL physics implementations (and other GPGPU applications) to surface?
Because OpenCL is more limited at the moment, and because ATI's drivers are not production ready.

 

Reminds me never to buy a GTS250..

 

OK, loud and clear

 

sorry for my poor english, i'm not english, i apologize.

i want to clarify that i am an HAPPY ati user,

i had both nvidia and ati cards and i had never problems.

but from X1000 series we have seen many interesting features sponsored by ati and never well applied. Example?

- gpu hardware conversion of video: only a buggy ati converter tool that doesn't use GPU and NOW (last year) the cyberlink media expresso that use stream

- stream and opencl: i can't see games using this tecnology, may be only few titles (metro 2033...marginally), and i can't see popular tools, instead of cuda

- phisics, havok gpu, bullet phisics... just techdemo, they speaks about hardware phisics from x800 series, meanwhile nvidia bought ageia and give us a poorly BUT WORKING gpu phisix, of course NOT optimized to run on cpu (because the must sell gpu).

 

ok i know i said im gonna keep my mouth shut, but i had to reply.


well it does with all dx10 gpus, guess i should have said that earlier and it runs fine off Cpu too, that Bullet SDK ran at 60fps in all cases HW or SW.

and OpenCL wasn't possible before dx10 (unified) hw so that's why we didnt see anything sooner., yea it was in the works since early 2005?, but real dx10 gpus arrived end 2006 - and this was to soon for it anyway.



@Mineria

yep you're right its a $hit gpu, but at least it does its job well at low 1280x1024 reso atm (50Gtexel/s is allot).
And you know what's funny, i get the same or sometimes better fps at my reso then lets say 480gtx at 1920x1200; be it physx games or normal dx10 games and i saw that many times,.. not that i would want to brag, why would i its a low reso :loco: lol, but still i can enjoy them at full speed too. Although yea i need to get rid of it soon as its starting to show its age + its missing dx11c1::biggun:

 

Nope, on the ATI side only HD5000/6000 series fully support Opencl.

 

Wasn't what I meant, more that you seem to dislike comments in favor for nivdia making it seem you are bitter cause of them.
The GTS250 is mighty fine GPU for it's price.
I think all here agree on that PhysX is to limited, even the people you consider as fanboys, so comparing Bullet with Havok isn't a bad thing at all.
You think Bullet stays open source if AMD starts to ship it with their products?
Somehow I doubt..

 

you want to know why i dont like physx anymore?

i use to allot, but then i saw the real truth behind it..

1) optimization
2) crippling it on purpose
3) it could be allot better visually, but it can't becuase of #2
4) its not for all

like i already said it, optimization is the biggest problem. It should run above or at least 60fps in heavy physx scenes, not drops to low 30fps.. and then they go and say gpu is to weak,... of course they will so they could sell more gpu's. I bet it would run amazing if they recoded the whole thing, it would be like Bullet - that runs great on both cpu and gpu.. but they don't want that, because it would hurt their sales.

havok isn't openCL anyway - that's why i said its irrelevant and looks like intel thinks the same, otherwise they would have done something with it, not just leave it as it is... it looks like its efficient enough running on cpu.

well i doubt it, why would they make it open source and then pull the trigger and closed it like physx, that's like shooting in your own foot.

If nv really decides and port physx to openCL than great, but as long its a closed std its pointless, especially when they're crippling it on purpose, in this case or until then Bullet is much much better alternative.

 

HD4000 series support OpenCL too. You just need to download AMD Catalyst™ Accelerated Parallel Processing (APP) Technology Edition driver package instead of regular Catalyst.
I tested OpenCL on HD4870 and it worked. HD3000, even HD2000 series might be supported too.

 

Anyone used fraps to test performance on this, uninstalled it the other night.

Bullet will fail because its open source, there is no money to be made from open source stuff, especially on the PC these days.
Similar to how OpenGL is ignored in favour of DirectX, and also because AMD have nothing to gain from it financially, so they won't pay for it to get put into games.

 

and what's directX a closed std?

lol go troll over Bullet idc anymore.

 

Who said DirectX was a closed standard?
I said open source stuff doesn't always do well financially, tends to get pushed under by the companies willing to splash out the cash.

You can call it trolling, i call it a prediction, AMD said one game would be released before the end of 2010, but we have a week left of 2010, and the game hasn't even been announced, so until that game appears, i will be skeptical.

I do wonder if they will be DX11 only though.

Edit: This looks pretty good though, better than this Robot Demo, apparently uses 6 threads too, just grabbed the XBLA demo to see how it runs.
http://www.youtube.com/watch?v=TGUvbNYOb5E

 

In-house modified bullet physics from last year. This just shows AMD and Bullet guy can't even make any decent demo at all.

Indie or arcade like to use Open source due to license/royalty issue.
AAA developer like to use corp's middleware probably due to better support.

 

There is beta support for HD 4000 series only. I tried it on a 4870 and a 4670 and it only worked on the 4870.

Not to mention the known performance problems.
http://tech.icrontic.com/news/amd-limited-opencl-performance-for-radeon-hd-4000/
http://translate.google.de/translat...000-karten_-_kein_grund_fuer_treiber-updates/


According to ATI there will be no support for HD2000 and HD3000 as they lack required hardware.

 

Open source doesn't equal no money, there's lot of open source that isn't free, there's open source that charges support, there's just different bussiness money.
Open source guarantees the software will survive crappy bussiness choices, free learning, cross platform reach, etc.
Directx not only improved with time, it has microsoft behind, so there's no point to compare it with opengl just for being a graphic api.

 

Hahahahahahahahaha . Fail. This is an opinion, not a fact...

Maybe you should do your own research, and btw - a blog is not proof or evidence of anything.

This, from the same blog u used as "proof", :giggle2:

D3D11Bitonic Time: 0.059 FPS: 16.9869[/quote]
http://unigine.blogspot.com/2010/02/cuda-vs-opencl-vs-directcompute.html

 

What's your point exactly? If you don't know who Unigine are, you may want to look them up.

When optimized properly, there should eventually be little differences. Of course you can find differences right now, I've explained this previously:

The Scalable HeterOgeneous Computing (SHOC) Benchmark Suite

5.2 CUDA vs. OpenCL
When choosing to port a legacy scientific application to a
heterogeneous computing platform, the choice of program-
ming paradigm is critical. While CUDA enjoys an advan-
tage in popularity and elegance, it confines codes to NVIDIA
GPUs. OpenCL has tremendous potential for code portabil-
ity, but is rather verbose and may not meet the performance
requirements of all developers.
SHOC provides an easy way to compare the relative per-
formance of CUDA and OpenCL on a target platform, as all
kernels have received the same optimizations on each plat-
form. Figures 9, 10, and 11 detail CUDA and OpenCL per-
formance for the Tesla C1060 GPU, using a single node of
the Lens cluster. Performance is nearly identical for all Level Zero
benchmarks. However, when nontrivial device kernels
are used, OpenCL begins to trail CUDA performance signif-
icantly (especially in FFT). This demonstrates the immatu-
rity of the OpenCL compiler and has dramatic consequences
for potential application developers.

Note the explanation for the difference in speed.

 

Here, I'll help you out, considering that "Blog" is the only word on the site you understood.
http://en.wikipedia.org/wiki/Unigine

 

It's the end result that matters most...

 

Yes, it is. Which is why it's taking so long for OpenCL physics to become a reality for consumers.