| HappySqurriel said: Please, never work on a project I am working on! You seem to have very little understanding of the difference between PR bullshit and what really matters. Having built several raytracers when I was in university, I can tell you that Raytracing performance is largely dependant on the choices of intersection tests you choose and the datastructures you use to limit the number of intersection tests you have to perform. The Cell processor doesn't offer much in the way of a per-clockcycle improvement in the intersection tests performance, and offers little to improve the performance in managing the datastructures; it does offer parallel processing of the intersection tests but the performance gains will mostly be limited by the loss caused by load balancing across all the threads. In other words, the only reason they can make a claim like this is because Intel, AMD and IBM haven't made a single core processor in several years ... |
Intel and AMD still make single core CPU's
Also, go here, and read a bit.
http://www.ibm.com/developerworks/power/cell/index.html?S_TACT=105AGX16&S_CMP=LP
This is a nice paper on ray-tracing on the Cell.
A good read, but for the ppl not interested in reading, here is a part of the paper:
For benchmarking purposes the following values
were selected:
• 1280x720 (720p) output image size
• 7455x8005 Map size
• 2048 map steps to full haze
• 1.33 x (2 – 8 Dynamic) or ~2-32 samples per pixel Visualization and Computer Animation
With these settings the following rendered frame
rates were measured:
System / frames per second
2.0 GHz Apple G5 1GB memory (no image encode) 0.6
3.2 GHz UP Cell 512MB memory 30
3.2 GHz 2-way SMP Cell 1GB memory 58
