ImgTech Caustic R2500 Realtime Raytracing Accelerator

Discussion in '3D Hardware, Software & Output Devices' started by Grall, Jan 21, 2013.

  1. Grall

    Grall Invisible Member
    Legend

    Joined:
    Apr 14, 2002
    Messages:
    10,801
    Likes Received:
    2,172
    Location:
    La-la land
    It's coming to market soon now it seems. Here's an early preview (no real specifics on performance): http://arstechnica.com/gadgets/2013...t-on-imaginations-real-time-ray-tracing-card/

    It's not cheap, but the biggest model with dual ASICs and 16GB DDR2 clocks in at $1500, so it's not uber expensive either. Power draw is so low (60W max, 40W typical draw at 100% workload) it doesn't even need any auxiliary power connectors at all, it draws everything from the PCIe slot.

    There is a short youtube clip with a guy demonstrating the card real quick (mostly fumbling around with the controls, seemingly not very familiar with the software :razz:), and the tiny view of the object being rendered seems to update at about 10, 15fps maybe. Framerates are always hazardous to judge from a youtube clip anyway so this estimate could be off by quite a bit.

    Anyway, it's fast enough to qualify as realtime by most human standards. Maybe that changes a bit in a fullsize window, who can say. We'd have to wait for more infos to be presented first to know for sure, but this is just a first iteration after all. If this tech takes off we can expect some pretty intense growth in performance.
     
  2. fellix

    fellix Hey, You!
    Veteran

    Joined:
    Dec 4, 2004
    Messages:
    3,490
    Likes Received:
    400
    Location:
    Varna, Bulgaria
    Well, not real-time, but quite snappy off-line rendering. Not much different from all the other CUDA an OCL PT renderers out there. Kudos for the minuscule TDP design of the accelerator, though. And the generous amount of on-board memory is a wise choice, despite the outdated DDR2 tech -- fitting all the scene assets in one place in critical for RT performance.
     
  3. Bumpyride

    Newcomer

    Joined:
    May 19, 2005
    Messages:
    54
    Likes Received:
    0
    Location:
    MS, USA
  4. rpg.314

    Veteran

    Joined:
    Jul 21, 2008
    Messages:
    4,298
    Likes Received:
    0
    Location:
    /
    IIRC, a high end GPU does ~100M incoherent rays/s, but with a LOT more bandwidth/power and area, so the tech seems nice.

    But, BUT, you need a GPU for shading anyway, which makes this thing kinda superfluous.
     
  5. Simon F

    Simon F Tea maker
    Moderator Veteran

    Joined:
    Feb 8, 2002
    Messages:
    4,560
    Likes Received:
    157
    Location:
    In the Island of Sodor, where the steam trains lie
    Not if it means the GPU (or CPU) is running the shading code at much higher efficiency.
     
  6. rpg.314

    Veteran

    Joined:
    Jul 21, 2008
    Messages:
    4,298
    Likes Received:
    0
    Location:
    /
    Since it's made on an ancient process and uses slow RAM, I think the tech seems promising. Integrating it with a GPU would give a great chip. As a product, with all the start-up effort involved in creating the ecosystem, I dunno....
     
  7. Simon F

    Simon F Tea maker
    Moderator Veteran

    Joined:
    Feb 8, 2002
    Messages:
    4,560
    Likes Received:
    157
    Location:
    In the Island of Sodor, where the steam trains lie
    I don't understand what you are saying? What start-up effort?
     
  8. keldor314

    Newcomer

    Joined:
    Feb 23, 2010
    Messages:
    132
    Likes Received:
    13
    100 is a good realistic rate, but I see it peaking at around 250 for some scenes. Since the 80 for Caustic is almost certainly the highest number they managed to measure under ideal circumstances, comparing it to the highest GTX680 number here: http://www.tml.tkk.fi/~timo/publications/aila2012hpg_poster.pdf would be a fairer comparison.

    Since this is a discrete card, and thus can't be installed in a laptop, power consumption is hardly an issue. Maybe for render farms, but then it's competing with the K20, which is likely to have only marginally worse power efficiency. Even here, the K20 may have an advantage, since the much higher compute density means you won't have to be running as many CPUs, assuming 4 cards per CPU or something like that.

    The one big advantage the R2500 has is the amount of memory - professional rendering uses massive scenes, and doing diffuse ray tracing out of core takes a fairly large performance hit.

    The big disadvantage is that to use it, you have to rewrite your renderer to use a new API.
     
  9. mboeller

    Regular

    Joined:
    Feb 7, 2002
    Messages:
    922
    Likes Received:
    1
    Location:
    Germany
  10. Simon F

    Simon F Tea maker
    Moderator Veteran

    Joined:
    Feb 8, 2002
    Messages:
    4,560
    Likes Received:
    157
    Location:
    In the Island of Sodor, where the steam trains lie
    For incoherent rays?
    Have you looked at OpenRL? If you've done any OpenGL programming, then the shader language will be familiar. IIRC, virtually all it does is add "fire ray" to the language.
     
  11. cjo

    cjo
    Regular

    Joined:
    Mar 31, 2010
    Messages:
    350
    Likes Received:
    124
    The caustic website claims 80 million incoherent rays per second per chip. The R2500 has two of these chips, yielding 160 million incoherent rays per second.

    Pretty much. There are frame shaders (basically the camera - you emit N rays per pixel) and ray shaders (these get executed when a ray hits a surface). The information about the ray is included in some predefined variables and the code is basically OpenGL with a couple of extra functions.
     
  12. rpg.314

    Veteran

    Joined:
    Jul 21, 2008
    Messages:
    4,298
    Likes Received:
    0
    Location:
    /
    People have to use a new API supplied by a single vendor with not much money behind it...
     
  13. Pottsey

    Newcomer

    Joined:
    May 26, 2002
    Messages:
    65
    Likes Received:
    2
    Location:
    Nottingham UK
    Not sure you can really call it new as it has been around years even longer if you go by its old name. OpenRL has been around a reasonable amount of time and is used by various companies and products and even works on all major cards from ATI to Nvidia. What do you mean not much money behind it? Its seems to have all the backing it needs

    Anyway I cannot wait for this tech to get down to reasonable prices (home use). I used to love doing ray tracing on my Atari STE that tooks days if not weeks for 1 static change.
     
  14. Simon F

    Simon F Tea maker
    Moderator Veteran

    Joined:
    Feb 8, 2002
    Messages:
    4,560
    Likes Received:
    157
    Location:
    In the Island of Sodor, where the steam trains lie
    In a previous job I used have (and take home) a board with 16 T-800 transputers to do ray tracing. Compared to the 286 pc it was plugged into it was a super computer. Still, generating pictures for a film recorder (something like 4k * 2k ?) took hours.:roll:
     
  15. Lightman

    Veteran Subscriber

    Joined:
    Jun 9, 2008
    Messages:
    1,804
    Likes Received:
    475
    Location:
    Torquay, UK
    Very interested in this technology as I had my spell in 3D Ray-Tracing and animation back on Amiga 1200. Back then doing light reflections and refractions with 16 bounces would basically slow down pixel production of the fairly simple scene to 1 PIXEL every few seconds! That was on 030/50MHz with 68882 :)

    I think combination of raster GPU and RT will bring best of both worlds. Ideally they should sit next to each other on the board sharing memory or be integrated into one chip to avoid any headaches with latencies and data transfers.

    2013-2015 seems to be very interesting technology wise in PC space, contrary to 2010-2012 where there was little exciting happening!
     
  16. keldor314

    Newcomer

    Joined:
    Feb 23, 2010
    Messages:
    132
    Likes Received:
    13
    Yes, the 250 number is for diffuse rays, so as incoherent as you'll ever get in a real renderer. Honestly, once you start using various metropolis style path tracing algorithms, coherency will improve to some extent.
     
  17. MfA

    MfA
    Legend

    Joined:
    Feb 6, 2002
    Messages:
    6,810
    Likes Received:
    477
    Presumably they just used random rays ... good way to crash a GPU.
     
  18. cjo

    cjo
    Regular

    Joined:
    Mar 31, 2010
    Messages:
    350
    Likes Received:
    124
  19. Pressure

    Veteran Regular

    Joined:
    Mar 30, 2004
    Messages:
    1,340
    Likes Received:
    272
    Wait, did the number of incoherent rays just drop from "up to 80 million incoherent rays per second" down to "50 million incoherent rays per second" per chip?
     
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...