Real time ray tracing has finally arrived in an affordable package

[L'Ancien Regime]

http://arstechnica.c...y-tracing-card/

The faster of the two cards, the $1500 Caustic R2500, is physically large, but the actual hardware driving it is much less powerful (and power-hungry) than what's found in a high-end workstation graphics card. Two of Caustic's ray tracing units, or RTUs, are located under the card's small cooling fans. Each of them has 8GB of memory dedicated to it for a total of 16GB on the entire card.

This sky-high amount of memory, which far exceeds the amount available on most graphics cards (and even many computers), is the highest-end spec the card has. The heavy lifting is done by silicon that seems ancient by today's standards: the RTUs are manufactured on a 90nm process and uses DDR2 memory, both of which were cutting-edge circa 2005 or so. Despite this, the card still requires a relatively small amount of power. While its peak power consumption is rated at 60 watts we were told that realistically it maxes out at about 40 watts—much less than a modern high-end (or even a low-end) graphics card, and well under the amount that would require the card to have a separate power plug.

For those with less cash, the $800 Caustic R2100 offers one RTU paired with 4GB of RAM and consumes about half the power of the R2500, with a peak power consumption of between 30 and 40 watts—because it has half the RTUs and a quarter of the RAM, its rendering speed should be less than half that of the R2500, though we weren't able to see the low-end card in action to be sure.

The cutting edge secret sauce of the R2500 isn't in its raw power, then, but the patented algorithms that Imagination and Caustic are using to solve the real-time ray tracing problem.

"The way the algorithm works, it turns ray tracing from a high-performance compute, memory-intensive problem to one that's more like a database problem," Imagination Technologies Director of Product Management Michael Kaplan told Ars. "It's highly optimized. We can store about 120 million triangles on that card."

[michalis]

Have you tried new blender and cycles? On a nice ~600$ GPU, or on a hi-end dualXeon CPU?

It isn't that fast exactly but it still is considered as real time rendering. Of greater quality than this on the demo. (excellent GI)

The interactivity is great, especially when you setup a scene.

IMO, the problem is not the hardware exactly. It's the software itself. Lot of great improvements on this are.

My wish: To be able one day to sculpt under GI illumination. Even now, in blender/cycles, I can have very fast previews as I sculpt. SImilarly as in 3dcoat but using a real raytracer/GI of excellent quality.

GI eats carving as hell. It's very important to know how deep you have to carve sometimes. Capturing the light.

Thanks for posting. I love this technology.

[L'Ancien Regime]

Have you tried new blender and cycles? On a nice ~600$ GPU, or on a hi-end dualXeon CPU?

It isn't that fast exactly but it still is considered as real time rendering. Of greater quality than this on the demo. (excellent GI)

The interactivity is great, especially when you setup a scene.

IMO, the problem is not the hardware exactly. It's the software itself. Lot of great improvements on this are.

My wish: To be able one day to sculpt under GI illumination. Even now, in blender/cycles, I can have very fast previews as I sculpt. SImilarly as in 3dcoat but using a real raytracer/GI of excellent quality.

GI eats carving as hell. It's very important to know how deep you have to carve sometimes. Capturing the light.

Thanks for posting. I love this technology.

http://www.tomshardware.com/news/Titan-Nvidia-GK110-gpu-tesla,20614.html

http://news.softpedia.com/news/NVIDIA-GeForce-Titan-Means-No-GTX-780-Has-85-Performance-of-Dual-GPU-GTX-690-325712.shtml

The NVidia Titan is due out by the end of this month ( February 2013)

The article is saying three Titans (each with 6 GB of DDR5 RAM) and with 2500 CUDA cores each ($890.00 ea) will give you RTR with Ray Tracing and god knows what else...

[L'Ancien Regime]

here it is! The GeForce Titan's first public appearance;

http://www.geeks3d.c...+Geeks+Of+3D%29

The first pictures of the GeForce GTX Titan are available, few hours before the launch (if still planned on Feb. 18 2k13)… The GTX Titan is powered by the GK110 GPU with 2688 CUDA cores (837MHz base clock, 878MHz boost clock), 6GB of GDDR5 memory (6008MHz effective speed), 224 texture units, 48 ROP units and a TDP of 250W.

[L'Ancien Regime]

http://www.geeks3d.com/20130219/nvidia-geforce-gtx-titan-arrives/?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+TheGeeksOf3d+%28The+Geeks+Of+3D%29

It's here. It'll be a three or four month wait but I'll be getting two or three of them for sure.

[Carlosan]

around 1000 bucks is a good price... and 2 can be used in SLI...

render engines that only use CPU must move fast to GPU or hybrid fast, lol

Now the move is app related... to use for tasks cpu+gpu processors at the same time.

I dream to have a supercomputer at home, lol... (who does not want that )

//edit

@carlosa

A 32 bit exr displacement map at 16k or 8k or even 4k, lot of such textures are in use, in film industry. (see the making of prometheus)

Do you have any idea how much ram you need for such tasks?

erhhhhh... no

[michalis]

@carlosa

A 32 bit exr displacement map at 16k or 8k or even 4k, lot of such textures are in use, in film industry. (see the making of prometheus)

Do you have any idea how much ram you need for such tasks?

Indeed a hybrid system may be a solution.

Or, well, maybe in the future we won't even have a GPU (or a CPU as we know them today)

In blender community, we all are waiting for these new Nvidias, we'll have the first tests soon. (Cycles)

BTW good news for OSX users.

OSX 10.7 or better 10.8 do now support any Nvidia (drivers) natively.

[AbnRanger]

I too think developers of 3D Render engines need to get on board the GPU revolution now. Why? Cause the RAM on the cards are now becoming less and less of an issue. But most importantly, CPU development has slowed to a crawl, now that AMD has publicly stated they are no longer competing with Intel in the CPU arms race. They are now focused on Value and the lower end market. 6-8 cores is enough computing power for 80% of computer users these days. The enthusiast market for CPU's is only a small percentage of that group. Intel would normally have gone to 8-12 cores by now, if AMD were still putting up a fight.

This is also why I have been trying to get Andrew to look into GPU acceleration throughout the app. It gave Mudbox the performance crown overnight, when they did (MB 2009). The arms race for GPU's is alive and well, so this is where technology is headed.

[Tony Nemo]

I will switch renders from C4D to Octane when their integrated plugin is available. The last 2 minute animation I did took 70+ hours using all 8 cores of two CPUs.

[michalis]

@Tony Nemo

Octane has still some limitations. Whatever GPU you have. On the other hand, all these octane renders are so lovely! I like this engine.

@Abn

True! I have a 3.5 years macpro (dualXeon 16 threads). Still after all these years is much faster than a modern (logically prized) i7 engine. This doesn't feel natural, regarding CPU development.

[L'Ancien Regime]

I'm wondering whether it's worth going with the 22nm i7s or just keeping my 2xquad core harpertowns and just adding more RAM and a few Titans..

[AbnRanger]

I'm wondering whether it's worth going with the 22nm i7s or just keeping my 2xquad core harpertowns and just adding more RAM and a few Titans..

Depends on what you use most. If you regularly use GPU accelerated render engines, then the latter makes the most sense.

[popwfx]

http://arstechnica.c...y-tracing-card/

....

The faster of the two cards, the $1500 Caustic R2500, ..............

......

The cutting edge secret sauce of the R2500 isn't in its raw power, then, but the patented algorithms that Imagination and Caustic are using to solve the real-time ray tracing problem.

"The way the algorithm works, it turns ray tracing from a high-performance compute, memory-intensive problem to one that's more like a database problem," Imagination Technologies Director of Product Management Michael Kaplan told Ars. "It's highly optimized. We can store about 120 million triangles on that card."

This is interesting, but no matter how awesome or cost-effective the hardware is, it's nothing without rock solid driver software. Any idea on their ability to deliver in that arena? Drivers (and CUDA) are the main reason I steer clear of ATI - although I suspect that it used to be much more of an issue and I gather ATI drivers are much better these days. I just personally don't have the problems whenever I go NVidia.

One other question while we're on the topic, do any of you know about the Quadro K5000 (for Mac)? Is it shipping yet? vaporware? I really want one card that does awesome 3D and 4K video streams for editing and that seems to fit the bill, though overpriced (as macs don't have many options). I did however recently pick up a modded Nvidia 570 for an old Mac Pro in my studio modded by a guy on ebay to have EFI and bootscreens and it works awesome - though certainly not a card for the future.

[popwfx]

@Tony Nemo

Octane has still some limitations. Whatever GPU you have. On the other hand, all these octane renders are so lovely! I like this engine.

@Abn

True! I have a 3.5 years macpro (dualXeon 16 threads). Still after all these years is much faster than a modern (logically prized) i7 engine. This doesn't feel natural, regarding CPU development.

The Mac Pro I mentioned is also getting long in the tooth, but (other than the video card - which has just been rectified by this new modded card) it is still very very fast - especially with SSDs and bootcamp is awesome with Win 7x64. Certainly this hardware (despite being overpriced for "equivalent" roll your own PC hardware) has been fantastically solid, and never had any downtime (also the cleanest insides of any PC I've ever owned), So i too am also surprised at how useful such an old computer is in the pipeline -- though without significant updates from apple, I can't imagine getting another one for our next machine. The GPU / EFI problem/lack of updates/support for new card models is just too much of a limitation despite the many Xeon cores...

[L'Ancien Regime]

This is interesting, but no matter how awesome or cost-effective the hardware is, it's nothing without rock solid driver software. Any idea on their ability to deliver in that arena? Drivers (and CUDA) are the main reason I steer clear of ATI - although I suspect that it used to be much more of an issue and I gather ATI drivers are much better these days. I just personally don't have the problems whenever I go NVidia.

One other question while we're on the topic, do any of you know about the Quadro K5000 (for Mac)? Is it shipping yet? vaporware? I really want one card that does awesome 3D and 4K video streams for editing and that seems to fit the bill, though overpriced (as macs don't have many options). I did however recently pick up a modded Nvidia 570 for an old Mac Pro in my studio modded by a guy on ebay to have EFI and bootscreens and it works awesome - though certainly not a card for the future.

That card has nothing to do with ATI. That is a Caustics card. It talks about it in the article. It's running OpenRL...it's real time ray traced rendering of millions of polygons..

[popwfx]

yup I realize it is not ATI, I just mentioned that because I was saying how driver stability is important. Even though ATI had good hardware I found (in the past) their drivers were not super stable. So I guess I was asking how you thought this Caustics driver software was or if you had heard anything about it....

[L'Ancien Regime]

yup I realize it is not ATI, I just mentioned that because I was saying how driver stability is important. Even though ATI had good hardware I found (in the past) their drivers were not super stable. So I guess I was asking how you thought this Caustics driver software was or if you had heard anything about it....

Good question; obviously it does what it does really well which is real time ray tracing but as for running the eon vue interface or maya interface I don't know.

[michalis]

Meanwhile... from a macuser perspective.

http://blenderartists.org/forum/showthread.php?281817-Guide-CUDA-GPU-Rendering-With-Apple-Mac-Pro&p=2307456&viewfull=1#post2307456