A software renderer on CELL by IBM

Discussion in 'CellPerformance@B3D' started by chris1515, Mar 8, 2007.

  1. chris1515

    Veteran Regular

    Joined:
    Jul 24, 2005
    Messages:
    3,418
    Likes Received:
    2,048
    Location:
    Barcelona Spain
    #1 chris1515, Mar 8, 2007
    Last edited by a moderator: Mar 8, 2007
  2. Cheezdoodles

    Veteran

    Joined:
    May 24, 2006
    Messages:
    3,930
    Likes Received:
    24
    xD

    impressive
     
  3. inefficient

    Veteran

    Joined:
    May 5, 2004
    Messages:
    2,121
    Likes Received:
    53
    Location:
    Tokyo
    Cool

    "The smart little system (PS3) will reach out across the network and leverage multiple IBM QS20 blades to render the complex model, in real-time, with software based ray-tracing"

    Well that's not really fair :wink: A PS3 + multiple Cell blades composed that real time image.

    Still, it is very cool. And considering that they archived that in just 3 months..
     
  4. Rolf N

    Rolf N Recurring Membmare
    Veteran

    Joined:
    Aug 18, 2003
    Messages:
    2,494
    Likes Received:
    55
    Location:
    yes
    ... would be dishonest ;)
    They had lots of prior experience with the Cell architecture before. That's somewhat inevitable if you're designing it :razz:
     
  5. chris100

    Newcomer

    Joined:
    Mar 8, 2007
    Messages:
    118
    Likes Received:
    0
    I think the "TRUE" software render in these combined system are those IBM QS20 Server,not PS3 Cell itself.Since its main Ray-tracing and modeling works are completed on IBM Blade Servers,PS3 cell only decompress/combine... what they received to do final display.

    One IBM QS20 Server has two Cell 3.2Ghz BE and 1GB Ram,which make the "REAL" High Tech Rendering could be acheiveable on server.
     
  6. Cheezdoodles

    Veteran

    Joined:
    May 24, 2006
    Messages:
    3,930
    Likes Received:
    24

    Oh, not so impressive then.
     
  7. .Melchiah.

    Regular

    Joined:
    Dec 9, 2004
    Messages:
    287
    Likes Received:
    5
    Location:
    Helsinki, Finland
    As far as I'm concerned, the most interesting part of the article isn't the software rendering itself, but how easily and quickly they created the renderer.

     
  8. Turok

    Regular

    Joined:
    Dec 24, 2006
    Messages:
    330
    Likes Received:
    0
    This is pretty misleading, at first I thought it was being done by a single Cell processor, but this isnt the case.
     
  9. Megadrive1988

    Veteran

    Joined:
    May 30, 2002
    Messages:
    4,638
    Likes Received:
    148
    not too impressive.


    edit: but then when you concider it's software rendering on CELL, and how much slower it would be on a conventional CPU, I guess it is actually quite impressive

    [/back-peddling]
     
    #9 Megadrive1988, Mar 9, 2007
    Last edited by a moderator: Mar 10, 2007
  10. Arwin

    Arwin Now Officially a Top 10 Poster
    Moderator Legend

    Joined:
    May 17, 2006
    Messages:
    17,681
    Likes Received:
    1,200
    Location:
    Maastricht, The Netherlands
    Would be interesting to see how well this scales over a number of PS3s, as they indicate you can also just use a series of PS3s for this renderer. Say that for 5000$ you can buy 10 20gb PS3s and you can hook these up for rendering, you might get pretty decent performance. Would be interesting to know.

    And I agree with .Melchiah.
     
  11. LunchBox

    Regular

    Joined:
    Mar 13, 2002
    Messages:
    901
    Likes Received:
    8
    Location:
    California
    really cool!

    I'm quite impressed!
     
  12. macabre

    Regular

    Joined:
    Jul 25, 2005
    Messages:
    707
    Likes Received:
    14
    Location:
    Porta Hercynia
    The video looks like about 15 fps maybe ? 1 cell + irt does about 2 fps. Centrino duo + default max @ 68 min/frame (check the comments in the original link).
     
  13. chris100

    Newcomer

    Joined:
    Mar 8, 2007
    Messages:
    118
    Likes Received:
    0
    (Sorry about my amateur english grammer:smile: ) )

    I think the main issue is the limited ram capcity factor.One PS3 only has 256MB SystemRam,and it cannot be expanded by user.The idea just like serveral years ago one university reseach group bought MANY ps2 to build one kind of powerful(?) ,economically relatively to the common system.

    Did their project works? I do not know.
     
  14. Carl B

    Carl B Friends call me xbd
    Moderator Legend

    Joined:
    Feb 20, 2005
    Messages:
    6,266
    Likes Received:
    63
    And there we see where the supercomputing applications lie. I'm not even going to bother calculating how many times faster that is, but it's obviously orders of magnitude. Granted, you know Barry, and he wouldn't have taken many pains to optimize for the Intel chip before running the code. But still...

    Yeah, PS2 clustering was pretty 'whatever.' But PS3 clustering is a very real option:

    http://linuxps3.net/articles/cell-broadband-engine/terra-soft-to-enable-hpc-clustering-of-ps3s.html
     
  15. macabre

    Regular

    Joined:
    Jul 25, 2005
    Messages:
    707
    Likes Received:
    14
    Location:
    Porta Hercynia
    Yeah , and it also matters if the max renderer settings were matching the irt solution or not.
    Maybe he jumps in and let us know.
     
  16. chris1515

    Veteran Regular

    Joined:
    Jul 24, 2005
    Messages:
    3,418
    Likes Received:
    2,048
    Location:
    Barcelona Spain
    The raytracer of 3Ds max is more precise. It is simply impossible to have so much difference.

    It is 8160(2*60*68) more faster than the core duo. If it was 68 second, it will be more believable and very fast. And for comparison, it is better to programming the same renderer for the two microprocessor.

    Maybe it is an error typo.
     
    #16 chris1515, Mar 9, 2007
    Last edited by a moderator: Mar 10, 2007
  17. Jesus2006

    Regular

    Joined:
    Jul 14, 2006
    Messages:
    506
    Likes Received:
    10
    Location:
    Bavaria

    No. It would be better to program the best renderer possible for each architecture (that's what they did for CELL?). But 3ds should be pretty much optimized for x86/dual core already...
     
  18. pjbliverpool

    pjbliverpool B3D Scallywag
    Legend

    Joined:
    May 8, 2005
    Messages:
    7,583
    Likes Received:
    703
    Location:
    Guess...
    If one is doing a lot more work than the other (which it clearly is) then its not even a remotely accurate test. Its akin to saying a GeForce2 is faster than a GeForce 8 because it runs 3d graphics (Quake 3) faster than the GF8 (Crysis).

    Running the same test on both and optimising it as much as possible is the only way to fairly compare, and even then it would only show performance under those circumstances, use a different test stressing different parts of the architecture and the results will be different. Thus a suite of tests is required, each one running the same test on both systems optimised to the max.

    Bottom line though is Cell is not >8000x faster than a Core Duo is anything, thats simply impossible under fair circumstances.
     
  19. Shifty Geezer

    Shifty Geezer uber-Troll!
    Moderator Legend

    Joined:
    Dec 7, 2004
    Messages:
    40,598
    Likes Received:
    11,005
    Location:
    Under my bridge
    I think you'll find that's what Jesus meant by 'best'. Best implementation, rather than best possible renderer. Which is the only sensible meaning, because there's no upper limit to a best possible renderer. You'd add more and more recursion and GI and procedural texturing, etc. It's not like Cell can power a better raytracer than any other chip. They all crunch numbers, and can all render the same visuals. The difference is in implementation and speed.
     
  20. Laa-Yosh

    Laa-Yosh I can has custom title?
    Legend Subscriber

    Joined:
    Feb 12, 2002
    Messages:
    9,568
    Likes Received:
    1,452
    Location:
    Budapest, Hungary
    A software renderer works in a completely different way and usually does a lot more which may be unneccessary for the given scene. Building acceleration structures, loading textures into RAM, etc.
    And I have to add that the built-in raytracer of 3ds max is faaaar from being the most efficient renderer for the job...
     
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...