IBM announces QS21 server & new SDK; releases Interactive Ray Tracer binary

Discussion in 'CellPerformance@B3D' started by Titanio, Aug 30, 2007.

  1. Titanio

    Legend

    Joined:
    Dec 1, 2004
    Messages:
    5,670
    Likes Received:
    51
    IBM had a small flurry of Cell related announcements yesterday.

    The Interactive Ray Tracer we have seen demoed before is now available for download for PS3 and QS20 servers:

    http://www.alphaworks.ibm.com/tech/irt

    It is the binary it seems, not the libraries or anything for use in your own projects. Might be inspirational though..

    They also announced a new QS21 server and SDK for Multicore Acceleration v3.0:

    http://money.cnn.com/news/newsfeeds/articles/marketwire/0296425.htm

    More detail on the new SDK (available Oct): http://www-03.ibm.com/technology/cell/software.html
     
  2. one

    one Unruly Member
    Veteran

    Joined:
    Jul 26, 2004
    Messages:
    4,823
    Likes Received:
    153
    Location:
    Minato-ku, Tokyo
    Maybe it's a 65nm version of QS20? Smaller PSU etc.
     
  3. Shifty Geezer

    Shifty Geezer uber-Troll!
    Moderator Legend

    Joined:
    Dec 7, 2004
    Messages:
    40,509
    Likes Received:
    10,882
    Location:
    Under my bridge
    How do you use the raytracer? Is it just the fixed car demo? (where's the fun in that?!) Or can you load other scene formats or perhaps move the camera around?

    Still, very limited without the source code. Might be good for a performance comparison if the data can be imported/exported somehow.
     
  4. Titanio

    Legend

    Joined:
    Dec 1, 2004
    Messages:
    5,670
    Likes Received:
    51
    I doubt you can change the scene, but you can probably move the camera around, switch between shaders etc.
     
  5. chris1515

    Veteran Regular

    Joined:
    Jul 24, 2005
    Messages:
    3,378
    Likes Received:
    2,002
    Location:
    Barcelona Spain
  6. patsu

    Legend

    Joined:
    Jun 25, 2005
    Messages:
    27,614
    Likes Received:
    60
    Here's another attempt:
    http://proliferationofniches.blogspot.com/

    "Real-Time Ray Tracing on the Playstation 3 Cell Processor"
    http://eric_rollins.home.mindspring.com/ray/ray.html
    [Updated in June/July to include PPE pthread, Cell SDK 2.1 and XLC performances]

    "Real-Time Ray Tracing with NVIDIA CUDA GPGPU and Intel Quad-Core"
    http://eric_rollins.home.mindspring.com/ray/ray.html
    [Updated in July to include Intel compiler performances]

    He has a different conclusion. :) (Source code provided)
    What could have caused the differences ?
     
    #6 patsu, Sep 9, 2007
    Last edited by a moderator: Sep 9, 2007
  7. Shifty Geezer

    Shifty Geezer uber-Troll!
    Moderator Legend

    Joined:
    Dec 7, 2004
    Messages:
    40,509
    Likes Received:
    10,882
    Location:
    Under my bridge
    Your GPU link is the same as your PS3 one. I presume you meant to link here
    http://eric_rollins.home.mindspring.com/ray/cuda.html

    As for what's different, I think his implementation is just poor. He's rendering a 1024 by 564 image with 10 spheres and one light source in the scene and it only runs at 13 frames per second on the PS3 using all 6 SPEs, without any secondary rays. That is one seriously unoptimized raytracer! For just spheres, you'd ideally use CSGs and massively speed up ray tests. As the IBM case shows, they can render a larger image with far more detail at a far better framerate.

    Given the very low performance obtained from the Cell, Id say the comparison is very far from comparing best performance. What you're seeing is his results with his one implementation, and they're not too useful except to compare something of a brute-force raytracing performance between chips that you wouldn't ever employ outside of test cases.
     
  8. Arwin

    Arwin Now Officially a Top 10 Poster
    Moderator Legend

    Joined:
    May 17, 2006
    Messages:
    17,678
    Likes Received:
    1,196
    Location:
    Maastricht, The Netherlands
    So, basically, this gives an indication of how far away from creating fully ray-traced games were are currently? ;)

    40fps for 1024x1024 image, and no idea about the complexity of the LOD ... but a ray-traced pong in 3d with lots of tiles and some effects could be on the cards, then. ;)
     
  9. Shifty Geezer

    Shifty Geezer uber-Troll!
    Moderator Legend

    Joined:
    Dec 7, 2004
    Messages:
    40,509
    Likes Received:
    10,882
    Location:
    Under my bridge
    Not only that, but 40fps on a single recursion. The moment you add secondary rays, which is the whole point to RT, framerates drop more.

    However, the Stanford Bunny is a polygonal mesh. A CSG game would be great buckets faster. A game built entirely out of primitives with no textures would have minimal memory accessing and be almost entirely processing bound. I imagine Snooker could be handled in realtime with reflections, or something with the graphical style of Katamari.
     
  10. Arwin

    Arwin Now Officially a Top 10 Poster
    Moderator Legend

    Joined:
    May 17, 2006
    Messages:
    17,678
    Likes Received:
    1,196
    Location:
    Maastricht, The Netherlands
    Well yes, ok, but the framerate drops to about 25fps for secondary rays, and that's still close enough to 30fps for me to result in something playable. I hope someone does it. Could be good at least for the Linux PS3 crowd, but I'd even like it as a PSN game/demo. I don't think I'll get around to installing Linux until I can afford that 2nd PS3 for in my workroom ... ;)
     
  11. patsu

    Legend

    Joined:
    Jun 25, 2005
    Messages:
    27,614
    Likes Received:
    60
    Thanks Shifty for the correction :)

    I will email IBM's result to the site owner to see if he can improve on his work. :)
     
  12. Shifty Geezer

    Shifty Geezer uber-Troll!
    Moderator Legend

    Joined:
    Dec 7, 2004
    Messages:
    40,509
    Likes Received:
    10,882
    Location:
    Under my bridge
    I suggest you ask him to employ CSGs if he's just experimenting. A pure sphere demo, space partitioned and with sphere<>sphere collisions, ought to produce something akin to the old 16 bit demo-scene output - 65535 blit-balls and all that malarkey! I dare say an occlusion factor could be calculated effectively too, as the occluding term will be a factor of normal, distance traveled by secondary rays, and diameter of the occluding sphere. The end result could look quite convincing.
     
  13. 3dcgi

    Veteran Subscriber

    Joined:
    Feb 7, 2002
    Messages:
    2,432
    Likes Received:
    261
    Have any technical details been revealed about how the IBM ray tracer is implemented? I didn't see anything of substance on the site, but maybe I just missed something. I'm specifically interested in how data is managed to make the most efficient use of the local store memory.
     
  14. Vitaly Vidmirov

    Newcomer

    Joined:
    Jul 9, 2007
    Messages:
    108
    Likes Received:
    10
    Location:
    Russia
    such a game is already exist ;)
    http://www.virtualray.ru/eng/download.html

    some screenshots:
    http://www.ixbt.com/video/theor/virtualrayengine/planets.jpg
    http://www.ixbt.com/editorial/images/whoneedrt/bioform-sm.jpg

    CPU Rightmark also uses similar engine (source is available)
    http://cpu.rightmark.org/download.shtml
     
    #14 Vitaly Vidmirov, Sep 10, 2007
    Last edited by a moderator: Sep 10, 2007
  15. Vitaly Vidmirov

    Newcomer

    Joined:
    Jul 9, 2007
    Messages:
    108
    Likes Received:
    10
    Location:
    Russia
  16. 3dcgi

    Veteran Subscriber

    Joined:
    Feb 7, 2002
    Messages:
    2,432
    Likes Received:
    261
    Thanks for the links.
     
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...