First OpenCL demo on a GPU

[sriram87]

@solobackpacking No, Intel is working along with other developers like AMD and in OpenCL development, and doing well at it.

[solobackpacking]

There isn't a single OpenCL demo app or benchmark app for the Mac that can be downloaded and works.

I have a feeling Intel is trying their best to sabotage OpenCL.

[WXIIIR]

@solobackpacking OpenCL was first developed by Apple and the first manuals and applications of OpenCL were made for Mac, so your argument is complete crap. i probably shouldn't be saying this as i hate Apple and all that Apple stands for but they came up with a really great thing.

[solobackpacking]

@WXIIIR

You make no sense considering it is a reply to my comment.

Perhaps less paranoia and more comprehensive reading skills.

[poonwoofer]

Soon enough we're going to be simulating a big bang event, and thus the creation of a simulated universe.

[trecool9992]

@poonwoofer Just wait until quantum computing comes into full play.

[meming4]

I calculate transformations myself so Im pretty much using my cpu, the only time I use my gpu is to draw the pixels.

[AspirinaSaucer]

@meming4 you could calculate the transformations yourself ON THE GPU so you could do so many calculations at the same time. You should learn more about this instead of nonsense complain; if you do the calculations yourself or not it's still up to you, but you can use the hardware resources ie. for doing float multiplication in a lot of threads running on parallel.

[meming4]

@AspirinaSaucer how do do you do that? the reason I make my own matrices to multiply by vectors is because I can obtain the current positions of the vertexes rather then just the starting vertexes, This is highly useful for Lighting, Collision detection, Physics, shadowing, ect.

[AspirinaSaucer]

@meming4 I think you still can with opencl, just you can do a lot of parallel calculations, then with a few clock ticks you get a lot of big number operation results. Try reading the opencl overview in the khronos website.

[meming4]

@AspirinaSaucer Im using SDL, which comes with OpenGL, but its nearly impossible to get new versions of ogl. I calculate my own lighting and shadows. "why dont you just use shaders" because I cant. lol

[AspirinaSaucer]

@meming4 You can't? Why is that? If you're coding for ancient hardware then of course you can't do any of this, even on software. OpenCL runs on CPUs which don't have "shaders" either

[meming4]

@AspirinaSaucer lol, thats not the reason, the reason is that Im using the OpenGL that comes with SDL, but you its nearly impossible "upgrade" the this ogl. If I didn't want to calculate all this myself I could just make a windows app with ogl, then I could get the latest version, but, I need SDL for its event handling, networking, sound, ect. note that I didn't say it was impossible to upgrade the opengl with SDL, its just extremely hard.

[Bananadine]

I look forward to the day when programming has outgrown this nonsense

it's cool of course, but if a computer has a processor, then why can't we just use it

seriously--with OpenCL or whatever, you have to connect your program to a clonky specialized library in order to tell a processor how to process stuff, when you're already using some programming language whose sole purpose is to allow you to tell processors how to process stuff

that is ridiculous

[itsacorporatething]

@Bananadine

I totally agree, look at this published on the khronos opencl page: "In todays world of copy paste programming expertise, OpenCL will force you to learn the nuts and bolts of programming. Learn why it is important to learn about the hardware you are programming for. "

I think the industry is working on a system for scientific computing that wont distinguish between GPU and CPU. I think the idea is that you can write the code once, and then push it out to mismatched hardware.

[AspirinaSaucer]

@Bananadine Are you kidding? You always have to. Or what kind of software do you write? Of course it's a specialized driver, because it's for specialized processors. It makes sense if you think about it. If you don't need a specialized processor then just don't use OpenCL, but I can make good use of it just to do parallel multiplications and I think a standard is really cool instead of a specialized assembly language for each vendor.

[Bananadine]

@AspirinaSaucer A standard is way cooler than no standard! But if processors specialized for graphics are being used for general computing, then they are no longer just graphics processors and their specialization is somewhat inappropriate. Parallel computation is too important a concept to waste only on graphics--hence OpenCL, sure. But maybe it's even more important than that--too important to be hidden, in the long run, behind some weird optional library.

[AspirinaSaucer]

@Bananadine but hidden how? the thing is that before OCL and other parallel computing APIs, there wasn't any other way but OGL or DX for "hacking the gpu for general computing", or making your own drivers, but it would be really annoying to make every driver for every chip in the market. Now there is OpenCL, DirectCompute, CUDA, etc. but I think OCL is the one which has more advantages for me.

[Bananadine]

@AspirinaSaucer Yes yes, it's a good thing! ...as long as the industry doesn't stop there. I'm being negative about a good thing because I'm afraid (w/o evidence) that it will. If it's so useful to perform general computing on graphics processors that dudes generally want to do it, then ideally, every decent programming language would let you put ordinary threads, running programs written in that language, on any CPU or GPU (supposing there even are things called GPU's), without any fuss.

[AspirinaSaucer]

@Bananadine There are indeed things called GPU with a very different architecture that a CPU and that's why we need a new programming language. Old programming languages doesn't know anything about parallel computing so they provide no means to do these things or you will still need to use an external lib from a particular OS to do it. OCL would be like a language for taking advantages of both multicore CPUs and GPUs, it still may have a lot of future unless something better comes up.

[Bananadine]

@AspirinaSaucer Yeah, there are GPU's in present-day reality. I was talking about a hypothetical ideal.

Many old programming languages have threads, which are quite easy to use. Some old languages have other mechanisms for parallelism.  Parallel programming didn't appear with OpenCL and CUDA....

[AspirinaSaucer]

@Bananadine Yes, but the parallelization run on the GPU is very different to the ones running on the CPU, because the threads work in a way that they are faster if they do exactly the same operation in the same cycle, which would also be hard to do in other languages, because parallel computing hardware like this didn't exist before.

[Bananadine]

@AspirinaSaucer Ah. That's interesting. I cannot comment further because I'm out of my depth now.  :)

[troelsblum]

@AspirinaSaucer

Actually the Cray-1 super computer worked in a similar fashion. It was created in 1976

[RunAspirinaRun]

@troelsblum but most likely at that time a standard api wasn't necessary and now C hippies command (?) At that time they hardly had compilers probably to even think about it as something really needed--they barely thought parallel computing had any future.

[renkinjutsu01]

he does some pretty cool things by messing with the settings.

[ActionTux]

OpenCL is openGL?

[PowerExpress]

@ActionTux No, OpenCL is just a specification for parallel computation using different devices, among them graphics card. OpenGL is about graphics.

[DarkSoulX19]

damn,i have opencl in my video card,but the opengl is gone...

why?

i can't even run counter strike 1.6 in opengl mode.

and i can't install 10 directx.

and before i was able to intstall them...

my video card is MSI n9500gt md1g-OC/d2

[WalterSolano]

@DarkSoulX19 download the driver? D=

[DarkSoulX19]

@WalterSolano you think i haven't tried 

[WalterSolano]

@DarkSoulX19 no i don't think u have tried, if u unistall everything and reinstall it it will work

[sujin99999]

@DarkSoulX19 when u try to download it does it pop up a send errorr message about internet exploer?

[DarkSoulX19]

@sujin99999 the problem was from the new driver :\ it had errors with the 9series.. now everything is alright :)

[DarkS3phiroth17]

there are no possibility to be disappointed with a modern video card ATI/NVIDIA ..you laugh at what? you work for one of this company?wtf.. console doenst mean nothing here.

well...doesnt matter

[selo2010]

dx11 is useless now only in 2 years it wil be fesable dx10.1 is king and yes physics it`s not that much but cuda is greate for hmpc

[emax27]

But without the current gen cards, e.g Evergreen, and the time spent on actually making the games with low-poly displacement-mapped objects that are needed for tessellation in DX11 vs the old-school heavy-polygon count objects in DX10 and older, it all would have not been possible.

Right now, the few games that use DX11 do it scarcely because the PC market isn't DX11 ready. Nor is the console market.

But Evergreen has it's merits for the future!

[DarkS3phiroth17]

is just your opinion, physix is great instead, i admit there are just few games, that actually use that feature, but its awesome!

i dunno about sli drivers, but aniway i like nvidia drivers more than ati catalyst, aniway right now both company are good and bad, in the same way, people have very wide range of choice at good price from both company...about the future well we will see.

[DarkS3phiroth17]

just look at Cryostasis, sacred2, mirror edge, with and without physix, are totally differents games.

[haxxxor123]

You do understand that ATI is currently far ahead of Nvidia? ATI have DX 11 with tesselation while Nvidia have postponed their Fermi D11 cards because they are not living up to expectations, There are even reports that indicate that Nvidia is steppind down from the PC GPU war and going for mobile units instead.

Also i run Physics based on CPU better then my friends on PhysX systems because of the I7 technology, ANd ATI actually have better support on multi OSes then Nvidia (Apart from OSX)

[DarkS3phiroth17]

maybe in your dream, dx11 have nearly no use yet in the modern games, and none of these manufacters are far ahead to the other...im sure that the new card from nvidia will put ati in the shame once more...and then ati will come out with something new, like in an infinite loop...you really think that what you have now is new? the tecnology is already years ahead of what they want make you believe.

anyway nowdays you cant go wrong with any of them, both delivers great products.

[DarkS3phiroth17]

nvidia can already make use of tessellation too

"watch?v=gkI-ThRTrPY" , just wait for their new card

[emax27]

We've been waiting too long and all we are getting are clips of ghost products, IT'S SO FAST, YOU CAN'T EVEN TOUGH IT, NEVER MIND SEEING IT... yeah, just wait for them to disappoint you and I will be laughing in your face!

[LordNeb]

i7 is a great number cruncher - very true. but it will not crunch faster then a GPU - FACT. Plus, we're not to judge how far ahead ATI actually might be until the GTX 4 series are out and benched.

those reports are blatently false. no company would heavily invest in such vast different technology, only to shut down and see no return on investment. it would be daft.

Fermi is supposed to be built from the ground up. i reckon it's going to be quite special.

hell if it's not!? i'll buy a 5970.

[DarkS3phiroth17]

ATI tend to run more hot than nvidia, and have crappy drivers and support compared to them.

nvidia have better support and driver(on any OS) also better tecnology(physix,openCL,cuda), im not a fanboy i have a 4890 right now and im very happy with it, but only things where ati are better is price/performance ratio, gddr5 and dx11 @time, but i had to buy an aftermark cooling to take care of the insane temps of my gpu...(club3D 4890).

aniway how many people really need more power?

[aoakampfer]

@DarkS3phiroth17 You can't really say any of that. Drivers are hit and miss, for both companies. SLI drivers are known to be rather buggy.

Also Physx is an utter joke and not really needed. Its just a gimmick that some games use and isn't needed at all. Higher core system processors will make Physx entirely obsolete.

[masterzilf]

So mystical...

[MoldyAngle]

I want this just so I can mess with the settings XD

[Halfback1990]

1:09 like a black hole

[dban10]

The only thing i have to way is Radeon HD 5970 an unheard of

" Processing power (single precision): 4.64 TeraFLOPS

Processing power (double precision): 928 GigaFLOPS"

while nvidia Tesla C2070 which is 2 or 3 times more expensive Double Precision floating point performance (peak)520 GFlops - 630 GFlops

[Bradlz193]

wat is this

[r0galik]

openCL runs on my radeon like a charm, get the new drivers and install ati Stream SDK 2.0.0

[foxsux6000]

too bad it's only a gpu driver for opencl, it's pretty pointless if you want to use the cpu/gpu using OpenCL combined (because with nvidia u can't)

[MadsterV]

Great to read that Nvidia is also on board of the OpenCL train now.

common APIs are a good thing! now you won't have a zillion special cases to code, you won't have to pick hardware based on who develops for that platform and you won't have to cut deals with developers to "bring them" to your platform.

Everyone wins!

[SodaBurp]

Oh, and open CL too

[maypepnecro]

That would be great for making space related games. or aps.

[CypFlyer]

I dont understand why ppl are bringing the descussion ATI over Nvidia and vice versa, or talking about gaming cards.

In this situation id say Quadro, tesla give me those kinds of comparisons, you guys are comparing game cards. When it comes down to the applications of cuda its not about gaming or physics, its about taking parallel processing to a new level.

This does have its gaming applications but its scientific ones are great.

[Gallleke]

eeehm I actually think less bout that scientific stuff, I prefer the gaming side, which is way more fun

[point6000x]

@CypFlyer and openCL does that thing better than CUDA..

[WhatTehFux]

imo the only good thing about Nvidia is all the unique softwares. Like PhysX, 3D vision, etc.

But that still doesn't justify the massive price tags a lot of their cards have.

If the GT300 series are cheaper, then maybe ATI might have a problem, but right now with the 5800 series, ATI are dominating the GPU scene at the moment. 5850, DX11, *drools*

[CutOutSock]

Change unique software to gimmicks and i'd agree Oo

[Thorsupremecommander]

ATI sucks. They never managed to provide good OpenGL drivers (these drivers lack many features). They're delayed in physics and scientific platform computing. Regardless what you think, nVidia is the leader

[ZangetsuNX01]

ATI sucks huh? then how come their hardware excelleration is better then NVidia? plus their drivers are much better then that of Nvidia

[Thorsupremecommander]

I'm owner of both ATI and Nvidia cards :) and yes, it sucks! nVidia offers OpenGl 3.0 drivers, ATI doesn't . geForce 1st series offered full support for OpenGL 1.5, ATI even 9500 series (sic!) didn't have it! (only 1.4 and some features of 2.0). OpenGL on geForce 2 is faster than ATi 9500, the card more advanced and later released! Hard to believe? but it's true!

[SodaBurp]

Umm have you checked the newer cards?

ATI does offer OpenGL 3.1 support

[Thorsupremecommander]

yest, the newest card do have support GL 3.1. You can have it if you're ready to pay in comparition terms two month wages after tax, at least in Poland.

Take into account that ATI is late compared to nVidia.

[MadsterV]

actually it's the newest DRIVER.

Which means the "older" cards also do GL 3.1 now.

go FUD go!

[foxsux6000]

doesn't offer it, doesn't mean ati doesn't support newer versions.

nvidia's opengl drivers are highly unstable on windows vs ati being far more stable (i pick stable vs unstable)

[Reloaded211]

Yet on Linux, ATI drivers are crap compared to Nvidia's (games have more bugs and graphical artifacts), also glxinfo shows OpenGL version 2.1.9026 on ATI Radeon HD 4550. Way to go, ATI!

[1kinomD]

my nvidia drivers have always been stable, i havent witnessed any instabilities yet. so idk?

[marcogugo]

about time...btw i feel ashamed for intel ( most netbooks are selling with the Intel Gma950, which "partially" supports the 2003 Opengl 1.3!!)

[Thorsupremecommander]

ATI sucks huh?> you see, even their website for international users isn't working for two months now! there is only message "An unexpected error has occurred." after automatic redirection to polish version. but you can't preselect english!

[vasedeny]

what will the processors do?

[Hamorage]

ThThat makes as much sense as asking the folks standing around at a car show "what will the cars do?"

[mikeccuk2006]

openCL > CUDA

Just being universal :D

[DigitalVampire85]

Did someone just quote Wikipedia?! ROFL XD

[mnbgkjygvbu]

so this going to replace that fucking physx?

[kerogre256]

OpenCL is not designed by apple but by Kronos grup apple is just part of this grup

[cartilago1]

Not true, OpenCL was developed by Apple, and then submited to the Kronos Group

[sgtconfig]

i just went to an opencl conference last week.

apple heads the committee on opencl, but the specs are defined by kronos

[birphborph]

from wikipedia:

"OpenCL was initially developed by Apple Inc., which holds trademark rights, and refined into an initial proposal in collaboration with technical teams at AMD, Intel, and Nvidia. Apple submitted this initial proposal to the Khronos Group."

[StRoRo]

I have seen a similar thing to that demo on TV, its a simulation on how galaxys could be formed. Is that true?

[Timverbaz]

As of today I could not find the Nvidia OpenCL SDK. When will it be released ?

It's known that for now CUDA performs better but I'd prefer OpenCL as I'm aiming for compatibility.

[Foxynews]

when it's done ;)

[torentmonkey]

my dear friends ...download nvidia cuda sdk browser and select nbody simulation and it will look just like this .....and at the top of the window will show how much teraflops are processed per second bythe gpu...i had os 7 and 8500gt and it damn worked ....fake

[Traceurb]

how many different physics engines can some one have on a games engine. could i have nvidia and havok on the same games engine? and could i have more on top of that so that i can use each one for certain things like cloth or destruction or water and sand etc?

[Traceurb]

what is the hardest part of making a game and how long does that bit usually take?

[morgs2020]

GO OpenCL !

Go OpenGL

& GO Open Source

[NeoXF]

Then fuck PhysX, why are you supporting nVidia?

[Foxynews]

nvidia should make physx OpenCL compatible so everyone can enjoy it.

Nvidia is not helping themselves not releasing it on other gpu's using OpenCL, they are seriously damaging their image in the eyes of hardcore gamers!

it's so bad that nvidia will lose millions of dollars of sales, and i am not overreacting hardcore gamers are THAT important to nvidia's end-line.

[Foxynews]

real gamers don't use Nvidia anymore...

Nvidia is overpriced crud, pay for it and they'll lose yet more money.

Ati has better gpu's at this time, nvidia is just re-labeling it's old crap and giving it an entirely new name... go buy your geforce 250 which is just a re-labeled 9800GTX...

[linadragon]

You are Kidding Right? The 285/295 are heaps better than stuff ATI puts out. Admidately Nvidia has had some bad drivers off and on recently but ATI has a long history of non optimized drivers, their openGL support he heinous even on newer cards.. They dont support vertex buffer objects properly and a bunch of other openGL extensions. People still use Nvidia more heavily because it can do so much more and actually supports both api's quite well

[v4ginav0mit]

that must be why i get frequent OpenGL errors with my ati card (Soldier of Fortune 2, Knights of the Old Republic and Doom 3). OMEGA drivers work better than the ATI drivers, how sad is that. Even then i still get errors.

I'd rather go with a stable product then a buggy one, thanks.

[DjKraemer]

Thanks dude, that inspired a laugh.

Go back to playing with your Voodoo 2.

[mxviii]

well voodoo actually IS Nvidia lol. Also Nvidia bought out Agea, so as of right now Nvidia has the only cards capable of physx, wihtout having to go out and by a seperate physx card....keeping in the tradition of voodoo might I add. Voodoo being the first cards to have integrated gpus, and now nvidia being the first cards to have integrated physx....

[zompaq]

you are a dumbass

[th3flyboy]

I think they are actually doing it though, think about it. This is a physx feature being run using opencl on the gpu... Hmmm I wonder what they are doing really :P

My guess is they are working on porting PhysX over to OpenCL and working on APEX.

[BLAZE9857]

Imagine when the software is available to send threads meant for CPUs through the GPU. This is amazing stuff, but it requires recoding to run. Regardless, this is an amazing step at using the hardware's full potential by optimizing the software to run it.

[fashnek]

Then they won't be GPUs, they'll be CPUs. Amirite.

[BLAZE9857]

Its still a different architecture within the processor. Yes they would both be able to run the same tasks, and i get your point.

But a GPU would be exceedingly more efficient at running some tasks sent to CPUs regularly. Think GPU as a hummer for offroading instead of a corvette.

[vapourmile]

I don't know if I agree with this analogy. The GPU is more or less a CPU off the mainboard, yes, but it's no hummer. For floating point operations it's likely to be a lot better than the CPU. Especially desktop CPUs which have never been great at FP. Except Digital's Alpha CPU that is. But that is sadly history now.

My theory is the public are not always, or even often, intelligent enough to know what's good for them.

[BLAZE9857]

Well agreed, GPU is much more efficient at floating point calculations.

And your right, i dont see the implications hitting public use anytime soon.

I would just like to have the cabability at some point to use my GPU for non 3D based aplications, so its of constant use.

[jonharris69]

Actually since OpenCL was designed by apple. It is going to be used in Mac OS X 10.6 which comes out September of this year. so fairly soon. The tests so far have seen huge improvements when turning the GPU into an extra CPU. when using it for graphics it uses its remaining power to take load off the CPU. And adjusts accordingly in real time. So you will not see any performance lags while playing a game.

[BLAZE9857]

Sadly im stuck with vista and managing recources like Ram (1 Gig cold boot) has been a peev, so seeying what i can get my GPU to do when im not playing a game has been a little hobby.

Kinda random there, but i havnt had a good replying thread in a while. I tip my hat to you guys.

Anyone looked into the Cell Processor architecture lately ? Wait nvm, way off topic, but still....

hit 500 word limit there, had to split message

[Iainaplin]

Oh crap im done for then. there goes my career

[Iainaplin]

can anyone reply with an actual, real talk example of what this type of api will mean to the gamer. will the games be more advanced? in what way. where can a person first hand experience of what this technology is boasting

[lin545]

yeah, your avatar will explode to pieces instead of just falling down in fps.

For more advanced games, google on chess.

[Tik0t]

This will be very awesome for things like game physics engines. Because you will suddenly be able to make use of ANY modern graphics card for heavy physics calculations you will suddenly be able to have very sophisticated destroyable enviroments, realistic cloth and water etc. It will basicly mainstream what we're already seeing in nvidias CUDA physx engine, however cause it will run on any card, easily, not just nvidia graphics it will be used for more than some extra effects some users see.

[th3flyboy]

the problem becomes when applying these things on a massive scale to games. That's when the GPU acceleration really is needed....

[th3flyboy]

the reason why is the technology isn't common enough for hardware accelerated to be heavily used. If more people were able to run hardware accelerated, and it would become at least a semi-standard feature, then software makers would invest on physx heavy games. Until then, they don't want to alienate half the userbase.

[kfishy]

beugnen, OpenCL is not just about physics (in games or otherwise). it's just a standardization/replacement of existing approaches in GPGPU computing like CUDA/Brook+ (tho i suspect both will live on as hi-level interfaces to OpenCL)

one practical example of GPGPU's usefulness is Elcomsoft who use it in their password-cracking apps which supposedly accelerates up to 50x the normal speed on a scalar (even multicore) processor

[askebengtson]

???

[LCNitro]

Congratulations Nvidia. The first to demostrates OpenCL in a video.

ATi is too slow to demostrates the new API :(

[keindesign]

So I guess WWDC 08 doesn't count as public? ;-)

[nvidiacuda]

This demo of OpenCL on NVIDIA GPU was based on early non-released OpenCL API/driver interface.

[gr3go86]

Is this a real OpenCL demo?! How could it be real if there is no driver from nvidia supporting opencl?

This probably is a cuda demo...

[Tik0t]

No, this is an open CL demonstration, nvidia is showing off something they will be releasing eventually. Like a video of halo 3 before the game is launched.

[gr3go86]

That demo is available in the cuda sdk, running cuda, not opencl. That's why I was asking.....

[Tik0t]

Ok, sorry for being harsh.

[AxiomFount]

It would make sense for NVIDIA to show an OpenCL demo that mirrored one of their CUDA demos.

So nothing strange there.