Ray tracing is the current holy grail of gaming graphics, the rendering technique that might finally make the licensed game based on Pixar's latest look as good as the film itself. But, the typically random nature of rays has made rendering them on traditional hardware inefficient, a problem Caustic Graphics
claimed to have solved, and is now backing that up by giving
PC Perspective some further details and demos. The company's tech will rely on a new graphics co-processor called the Ray Tracing Processing Unit (RTPU), working in concert with existing 3-D accelerators to deliver rays at frame rates high enough for interactive applications. How high? Early hardware dubbed CausticOne (that giant slab of silicon above) manages 3 - 5 frames-per-second in the demonstration video after the break. That's not nearly enough for twitchy first-person shooters, but second-gen hardware due next year is looking to deliver 14 times that -- plenty to get your high-reflectivity frag on.
posted April 20th 2009 9:21AM
Elmer you completely miss the point.
This wasn't using the GPU at all. This ISNT a hardware renderer. This lets you use the full power of your raytracing shading pipeline. But accelerates it. It's a hardware co-processor for a SOFTWARE renderer.
A GPU is a hardware renderer. It's extremely limiting and extremely frustrating to work with. The Caustic card ISN'T A RENDERER. It's a render accellerator. It would be like calling PhysX a game engine. It's not a game engine, it's a co-processor.
Now if your game wants to use raytracing it can call data like it would the CPU. This is exactly the same as using a CPU only faster. This is also very clearly still 'developer art'. The simplicity of the shaders is a result of the demonstration not the capability of the card. Anything that uses Raytracing can use raytracing EXACTLY the same as it does currently just faster.
if this thing ever makes it way to mainstream then they could have a winner on their hands..btw before anyone asks...yes is can play crysis!
I think its quite the news that these guys are developing this on a mac. Which stands to have a very interesting possibilities, since mac gaming hasn't been the best place to find gaming hardware or software.
Yah because this won't ever be on the PC. Nope. Not at all because the core components of the Mac are so different from the PC that it will never happen. Sorry, but Mac gaming has been, and always will be a joke.
*shrugs* There is nothing wrong with co processors. They generally are the stop gap before having it integrated into the system board. Take a look throughout computer history and this has almost always been the case. Everything from audio, to video, to NIC, to modem, to memory (Look at some of the smaller laptops out there memory is integrated onto the board now.), to the HD controller. It all eventually gets rolled into one set of chips as will this with time.
bad part is once they're integrated progress slows down.
Physx is integrated into most of the newer video cards as well.
Co-processors don't work for the reason PhysX failed: chicken and egg problem for consumers over whether to buy the hardware when the software doesn't exist, and for developers on whether to spend resources to develop for the hardware when nobody owns it.
However, it's possible for this to work for the reason why PhysX didn't fail completely: if this card can do raster based graphics as well at a good speed, then people would buy it in good conscience while they wait for developers to get on board. Because as far as I can remember, the whole problem about RT is that it requires a massive, massive amount of parallel processing, but that the architectures wouldn't actually be that different, just that today's GPUs are optimized with things like shader units in mind that RT wouldn't need.
It actually needs a discrete video card if I read it correctly...
Same reason Physx failed - additional card that doesn't do anything and costs twice as much as a normal video card. Either Nvidia/ATI/someboy buys them or they die from hunger.
Yeah, going to be a tough sell for the mainstream crowd unless they can get it cheap and fast enough. If it really adds a new level of realism to games they might get my Benjamins, but from the looks of the demos they're simply giving me what I already have (or less) simply without visual "tricks" and I simply don't effin' care.
So, it's a Poser-accelerator? THAT'S just what the world needs.
More desktops of highly reflective balls in space.
Why not just toss in another GPU and run raytracing code on it?
___Read the article.___
It explains exactly your question as much as the company was willing to release. Essentially, GPUs do not have the per stream memory to be able to modify the models in real time and other things.
Ray tracing doesn't look as good as scanline rendering anyway. All the techniques we've developed for things like anti-aliasing don't work with ray tracing. Plus, for any non-trivial scene, ray tracing takes exponentially more work.
I don't get why people are interested in real-time raytracing anymore. I thought the idea of it died in about 1998 when scanline rendering started to look so much better than raytracing.
Actually, if you read the article, you can see AA being implemented.
I thought we were moving away from add-on boards and graphics cards;
Graphic acceleration cards, Riser boards, co-processors, were all developed back in the day when processors were single core running at 500Mhz and not able to processes information fast enough.
Developing more add-on boards and such seems like a step backwards for me. Why not work on better core APIs that allow multicore CPUs to shift a core into GPU mode another into a Physics mode, and then us the remaining 2 cores for maths (they were talking about this when multicore became mainstream).
no matter how good your add-on is, in the end, you still get chocked by the Mobo's bus, the more the processor core(s) do on their own the better.
They because even modern multicore cpus still suck at math crunching, just look at the i7-965@3.2 GHz 51.20 gflops source
http://www.intel.com/support/processors/sb/cs-023143.htm
This company is gonna come out with it's first gen cards and the Nvidia is license the tech and through it in their drivers.
I'd rather Nvidia just buy them out and do like they did with Physx. That way I wont have to buy a separate card and risk it becoming redundant with my next videya card purchase.
The accelleration happens thanks to hardware not software.
PhysX was a problem which could easily be solved with GPGPUs. Raytracing isn't. In short it requires access to memory in a different way than a GPU operates. So Raytrace accelleration isn't terribly fast. As the movie describes it isn't just a matter of throwing more CPU clocks at the problem, there are other bottlenecks that need to be addressed. The caustic card addresses these hardware bottlenecks.
Now NVidia could license the hardware specs and add it onto their GPUs. But they can't emulate it and get any kind of performance improvement.
All this talk about Pixar and raytracing is bollocks anyway. The vast majority of Pixar's work has been done with REYES, which is a micropolygon rendering system, much more similar to scanline than raytracing.
OK thats nice you can have a low spec System with this Card in there and still get that same performance as a highend system
+1
im still trying to figure out what is supposed to be impressive in this video.
LOL. "raytracing is the future"
I've been hearing that for twenty years. The nice thing about the "the future" is that it never fucking gets here you so can keep on saying that and it'll still be true!
Raytracing is a correct model of our world, no cheats. If it could work in high fps it would be extremely easy to create nice/correct looking games. This is true simulation. Nice work.
@Cornelius
Ray tracing may optimize for O(ln(n)) for singular searches, but that's not the bottleneck. At least in the present incarnations of technology, we're talking about "how much geometry is represented in the scene". Adding more geometry means adding more memory. The bottleneck is the datapath to main memory. Ray tracers actually do not reduce the total geometry needing to be touched to render a scene by anything significant. Current 3d rendering methods can employ ray-tracer esk geometric culling to get that number more in line with a ray tracer, with significantly less silicon. That is the present bottleneck for rendering geometry.
It boils down to cost, the theoretical "advantage" of ray tracers in being able to view a large amount of geometry is just that, theoretical, it breaks down in application and will never cost anywhere near current hardware for the same level of performance.
Cost will drive adoption. Sum, I never expect mainstream adoption of ray tracing 3d hardware.
Ray Tracing is undoubtedly the future, but it still has a long way to go. When we can sustain a solid 60 frames per second, then I'll be interested.
F#!$% im going to need more PCIE slots for one of these....
I don't understand too well the implications of this product but everyone here is mostly way off on what is being implied.
Caustic One is an enterprise product not a consumer product, hence why it has a $4000 dollar price tag and is being advertised to developers on their website.
What Caustic One is supposed to do is reduce the time it takes companies to reduce their time creating ray traced scenes by a factor of 20. Reducing every dollar and minute they spend to 5 cents and 3 seconds would be a ridiculous amount of savings if it works.
As it stands Caustic One is useless for gaming. Ray tracing takes a lot of time to render and even the reductions this version would create would be insufficient to work for gaming in real time.
Caustic Two is the next iteration that they hope to have available by next year. That's the one that will make ray tracing for gaming feasible and this will only be considered if developers that currently ray trace now adopt Caustic One as the new standard in ray tracing development.