The GT300/Fermi Thread - Part 2!

[Behemot]

Maybe some manufacturer's insignia and the MB maker don't want to be blamed for leakage...?

[Teemax]

Just to clear things up about CUDA + Just Cause 2.

OpenCL and DX Compure Shaders are accelerated THROUGH the CUDA architecture standards. As such, the effects in Just Cause 2 will likely be accelerated on AMD hardware through their Stream architecture as well if they use OpenCL or Compute Shaders.

Everyone has to remember that CUDA is an all-encompassing term for the architecture that accelerates all types of APIs on NVIDIA GPUs (OpenCL, PhysX, folding, Compute Shaders, ray tracing, Adobe Photoshop GPU acceleration, Flash acceleration, etc., etc.). It is WAS proprietary but is now acting as a type of umbrella term for everything GPU accelerated. Much like Stream on AMD products.

So, which GPGPU framework will Just Cause 2 use? DirectComputer or OpenCL?

[SimBy]

whats with the pictuires with white outs and black outs that makes no sense whatsoever. remeber leaked HD5 series pics? were any of them whited out?? nope why the secrecy? maybe because this isnt what it will actually be? i just dont get it

Dude making the photo prolly tried to hide some identification markers? Just guessing.

[H2O]

whats with the pictuires with white outs and black outs that makes no sense whatsoever. remeber leaked HD5 series pics? were any of them whited out?? nope why the secrecy? maybe because this isnt what it will actually be? i just dont get it

Because there are several codes on there that can be traced back to the AIB. And no AIB wants to expose themselves to the wrath of Nvidia.

[prava]

Jeeez resize your pictures lads, its pissing me off to be scrolling right, left up and down all the time in order to see anything. Why there are no rules about image size?

[-Sweeper_]

thats GTX470 or GTX480 cooler?

[mindfury]

thats GTX470 or GTX480 cooler?

470.

http://www.computerbase.de/bildstrecke/28704/4/

[Sam_oslo]

naked 470 and 480
..]

Cool pictures The cooler looks totally different.

[mapel110]

So, which GPGPU framework will Just Cause 2 use? DirectComputer or OpenCL?

These effects will only be available on nvidia-hardware. So they don't use DC or OpenCL.

[Serra]

You really don't have to go too far in this thread before you can find some posts to delete. It's a bit of a challenge really, given how quickly this thread grows.

I understand that people are frustrated with Fermi's late launch and some of nVidia's other practices. But let's check our frustrations at the door when we come to this thread because it is about Fermi news, info and updates... that's it.

[SKYMTL]

These effects will only be available on nvidia-hardware. So they don't use DC or OpenCL.

I don't think that's possible. Unless the developer is using PhysX (which needs to be used in conjunction with another API to accurately model hair, water, etc.), there isn't anything out there other than DC and OpenCL right now that can use the GPU for this type of acceleration. If it is a proprietary engine that can do this we're talking some major $$$$$ invested on the part of the developer.

In order to accurately add water to a scene you need tessellation and selective geometry shading on the rendering side, DirectCompute for animation and physics if the water is interacting with anything. Unless the developer found away using DirectCompute for the animations, I can't see how this would be "exclusive to NVIDIA". Basically, they may use this to show off the GF100's power in DC or something along those lines.

[trinibwoy]

What do you mean? Of course you can use CUDA directly to implement these effects. There are hooks in CUDA for interoperation with OpenGL and DX buffers. Remember, while CUDA is the wider compute architecture it also refers to the "C for CUDA" language on which DC and OpenCL are very much based/similar to.

[Farinorco]

I don't think that's possible. Unless the developer is using PhysX (which needs to be used in conjunction with another API to accurately model hair, water, etc.), there isn't anything out there other than DC and OpenCL right now that can use the GPU for this type of acceleration. If it is a proprietary engine that can do this we're talking some major $$$$$ invested on the part of the developer.

In order to accurately add water to a scene you need tessellation and selective geometry shading on the rendering side, DirectCompute for animation and physics if the water is interacting with anything. Unless the developer found away using DirectCompute for the animations, I can't see how this would be "exclusive to NVIDIA". Basically, they may use this to show off the GF100's power in DC or something along those lines.

I don't think that, for example C/C++ for CUDA are any lower level than OpenCL so... why not?

Indeed, what I have read about it is that programming something with OpenCL might be less straight forward than with C for CUDA because of a lower level (and harder) setting up process.

But don't take my word on any of this, I have no clue about GPGPU programming.

[Chickenfeed]

Everyone has to remember that CUDA is an all-encompassing term for the architecture that accelerates all types of APIs on NVIDIA GPUs (OpenCL, PhysX, folding, Compute Shaders, ray tracing, Adobe Photoshop GPU acceleration, Flash acceleration, etc., etc.). It is WAS proprietary but is now acting as a type of umbrella term for everything GPU accelerated. Much like Stream on AMD products.

Thank you sir People get caught up in marketing terms like always...


At this point it looks like I may just have to go crossfire if I want a more cost effective performance boost (as much as multi gpu setpus are facepalm inducing affairs). I can see the 480 being nice at high res / IQ due to its vram and bandwidth advantage over Cypress (and unless you use 8x + AA at 1920x1200 and more often 2560x1600, 1GB of vram should remain adequete at 1920x1200 4x well into the future eg HD6000 ), but beyond that all these rumors so far are merely (WTB facepalm smilely face) If the 480 truely creeps up on 300 watts, that is quite strange. Would another shader cluster and some extra ram and similar ( perhaps lower? ) clocks really result in 80 more watts power usage (max) over the 470? For comparisons sake what was the max board power of the original GTX 260 and 280?

[ajaidev]

I wonder why the GTX 480 has a two pin fan connector and GTX 470 has a 4 pin one?

[trinibwoy]

I don't think that, for example C/C++ for CUDA are any lower level than OpenCL so... why not?

Indeed, what I have read about it is that programming something with OpenCL might be less straight forward than with C for CUDA because of a lower level (and harder) setting up process.

But don't take my word on any of this, I have no clue about GPGPU programming.

No, you're right. The runtime C for CUDA interface is higher level than the OpenCL API which is more similiar to CUDA's driver interface. Nvidia has absolutely zero motivation to use OpenCL in any situation where CUDA would suffice because that would effectively give AMD a free invite to the party. Although I don't think OpenCL is yet in a stable enough state to ship with a commercial game so it's a moot point anyway.

[zalbard]

I wonder why the GTX 480 has a two pin fan connector and GTX 470 has a 4 pin one?

Maybe an earlier sample, just a guess.

[SKYMTL]

What do you mean? Of course you can use CUDA directly to implement these effects. There are hooks in CUDA for interoperation with OpenGL and DX buffers. Remember, while CUDA is the wider compute architecture it also refers to the "C for CUDA" language on which DC and OpenCL are very much based/similar to.

Didn't I just say that in plain English?

CUDA allows for interoperability with DC, etc. but it goes to reason that same interoperability can be created (though not through CUDA) for ATI's Stream. Maybe I just didn't explain what I was saying well.

Allow me to post a slide directly from NVIDIA:

[trinibwoy]

Didn't I just say that in plain English?

Nope you claimed that you have to use either CS or OpenCL to access the CUDA infrastructure. That is incorrect as CUDA is also its own language.

"there isn't anything out there other than DC and OpenCL right now that can use the GPU for this type of acceleration."

[SKYMTL]

Nope you claimed that you have to use either CS or OpenCL to access the CUDA infrastructure. That is incorrect as CUDA is also its own language.

"there isn't anything out there other than DC and OpenCL right now that can use the GPU for this type of acceleration."

See my edit above. I should have used more words instead of less it seems. My mistake.

[kull]

I wonder why the GTX 480 has a two pin fan connector and GTX 470 has a 4 pin one?

GTX 480 also has a 4pin fan connector just look at the bottom of the card, it's black not white.
White 2-pin connector in the picture powers an illuminated GeForce logo on the top of the card.

[trinibwoy]

See my edit above. I should have used more words instead of less it seems. My mistake.

Gotcha. Don't know about Stream though, it was never very mature/stable and now seems all but abandoned.

[Sam_oslo]

I've been programming all my life, and have been looking into these stuff lately.
The CUDA architecture enables developers to leverage the parallel processing power of NVIDIA GPUs. CUDA enables this via standard APIs such OpenCL and DirectCompute, and high level programming languages such as C/C++, Fortran, Java, Python, and the Microsoft .NET Framework.

[SKYMTL]

Gotcha. Don't know about Stream though, it was never very mature/stable and now seems all but abandoned.

ATI needs SOMETHING to help developers use their GPUs for DC and OpenCL. Then again, they may just wait in the background and let others blaze the trail.

[Dimitriman]

Here's my general feeling about PhysiX

In about 1 year time, many people will be wondering this:

"Humm, with 4 cores a bit busy now with this game, I wonder how it would be like if I could use my other 4 Bulldozer cores to do something useful like run physics, too bad I had to waste 150$ on this graphics card just for that"