3 ms·

Not really that impressive, esp. compared to the product offered by Caustic graphics that has been posted here before, http://www.caustic.com/ http://www.causti

by pmorici 17y ago

Not really that impressive, esp. compared to the product offered by Caustic graphics that has been posted here before, http://www.caustic.com/ http://www.caustic.com/ they do realtime ray tracing with a piece of hard ware a kin to a graphics card that fits in a desktop.

manvsmachine 17y ago
Except that the total cost of investment is a $350 piece of commodity hardware and a free/open SDK. If it's not fast enough, just throw in more / faster cards. Using Caustic's platform means putting min. $4K into a unproven proprietary platform. If performance/price didn't matter, you might as well say "Not that impressive, considering you can just build a render farm". No disrepect to Caustic, but I could easily see them going the way of Ageia if they're not careful.
- pmorici 17y ago
  I see your point but is the target audience for this technology using $350 commodity cards to begin with? High end Nvidia Quadros cost in the range of $4,000. This CUDA solution looks like it provides less impressive results and I don't think a video production company is going to sacrifice image quality to save a buck. If this were targeted toward Joe average consumer than I agree with you.
  - manvsmachine 17y ago
    I think you're somewhat misunderstanding the role that CUDA plays in all of this. You wouldn't need Quadros because the CUDA cards would only need to be responsible for performing the raytracing calculations, not actually rendering the image (unless you want it to). So they could have had 4xGTX295's, equivalent to 8 of the boards used in that presentation, just crunching numbers and then one workstation-class card to render textures, etc if they were concerned about render quality. If someone was still concerned about using GeForce cards, Tesla C1060's still cost only about $1K a board IIRC, and I think they'd be way faster at this sort of thing than the GTX285 used.

by pmorici 17y ago

manvsmachine 17y ago

Except that the total cost of investment is a $350 piece of commodity hardware and a free/open SDK. If it's not fast enough, just throw in more / faster cards. Using Caustic's platform means putting min. $4K into a unproven proprietary platform. If performance/price didn't matter, you might as well say "Not that impressive, considering you can just build a render farm". No disrepect to Caustic, but I could easily see them going the way of Ageia if they're not careful.

pmorici 17y ago
I see your point but is the target audience for this technology using $350 commodity cards to begin with? High end Nvidia Quadros cost in the range of $4,000. This CUDA solution looks like it provides less impressive results and I don't think a video production company is going to sacrifice image quality to save a buck. If this were targeted toward Joe average consumer than I agree with you.
- manvsmachine 17y ago
  I think you're somewhat misunderstanding the role that CUDA plays in all of this. You wouldn't need Quadros because the CUDA cards would only need to be responsible for performing the raytracing calculations, not actually rendering the image (unless you want it to). So they could have had 4xGTX295's, equivalent to 8 of the boards used in that presentation, just crunching numbers and then one workstation-class card to render textures, etc if they were concerned about render quality. If someone was still concerned about using GeForce cards, Tesla C1060's still cost only about $1K a board IIRC, and I think they'd be way faster at this sort of thing than the GTX285 used.