Whats your opinion of real time ray tracing.

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Nemesis 1

Lifer
Dec 30, 2006
11,366
2
0
Ray tracing has long been considered too expensive for mainstream rendering purposes. Movie production studios have only recently begun the transition to using it; however, the true cost of ray tracing has been very poorly understood until recently. It is now poised to replace raster graphics for mainstream rendering purposes. Its behavior is very well suited to CPU processors, and scales well with hyper threading and multi-processor configurations. The traditional cache hierarchy associated CPUs is very effective at managing the external memory bandwidth requirements. For ISVs, a transition to ray tracing is a huge step forward freeing them from all the limitations imposed on them by today?s raster-based approaches. Ray tracing is one tool that can enable ISVs to aspire to achieving high fidelity photo (or cartoon) realistic imagery.

Now lets look at a dual socket Nehalem setup. 16 cores large L2 cache = 32 threads with H/T . Add in 1 larrabe card 16 cores 4 threads per core=64 threads large shared cache Vertex unit. Total of single 8 core cpu with larrabe= 80 threads . Add one more cpu and an additional larrabee card = 160 threads.

Looks to me like intel can pull this off . Will know more after spring IDF.
 

Nemesis 1

Lifer
Dec 30, 2006
11,366
2
0
Not promoting anything here . I do not need to promot for intel their products speak for themselves.

This thread is about RtRT and I got some really good replies . I really liked the radiosity
reply it was very interesting and good links.

Bigest problem is the nay sayers and how intel is simply promoting something for some hidden agenda.

Intel seems to be introducing products in the 09 time frame .That On a 8 cores =16 threads on one die. Larrabee with 16 cores 4 threads per core=64 threads for a total of 80 threads .

The good replies all doubted intel could have enough processing power. 80 threads is a lot. With intels present arch they showed RtRT with only 8 threads. 8 cores .

I may be over estamating Intels abilities here . But I can count 80 threads by 09 . So it looks good to me . Its not me saying 09 its Intel .

Now if it was AMD saying this I agree not possiable.
 

Nemesis 1

Lifer
Dec 30, 2006
11,366
2
0
Fot those of you who believe that. RtRT will be better on a GPU . I will give you support on your debate theory with this link. But before you draw any sort of conclusions from this. A lot more research is required. Intel's method pawns GPU method.I will give just 1 of many factors which is shared cache that Intel will offer. There is much more and if your truely interested you will do your own research . Without me listing a bunch of links few will read.

http://rapidmind.net/pdfs/RapidMindRaytracer.pdf

 

SickBeast

Lifer
Jul 21, 2000
14,377
19
81
I could see this being a huge boon for architects/engineers if it can actually fully render a ray-traced scene of decent complexity in real time.

It used to take me around an hour to generate a single high-quality rendering on my old Athlon XP-M at 2500mhz. :Q

Once a scene in 3D Max reaches a certain threshold, it used to cause my 9700Pro to chug along. Working with a wireframe or the primitive half-rendered 3D Max view was difficult and made it hard to envision the end product. I'd say that this type of advancement will cut most rendering job times in half at the least.

IMO it sounds too good to be true. I'll believe it when I see it.

I didn't like radiosity for architectural stuff. I can see its use for games/movies.