Ray tracing in realtime with PS3's

Started by
44 comments, last by zedz 17 years ago
Quote:u miss the point of my post, A/ cell is a lot lot better at this than a top of the line pc cpu, ie the $1000+ is a lot more illsuited than cell
I doubt that the CELL is faster than a quad core in practice.
It is like a vitamin pill package: you can read on the package what is in the pill but you do not know what is really absorbed by your body :-). People buy the pill because they believe that what is described on the package would really end up in their body :-). So the description what is in the pill let's them mentally skip the part what is really absorbed by their body. Now think of the PS2 as a super-computer and the description of its raw computational power is the description on the package of the vitamin pill package. Now go slowly one step further and think about the PS3 ...
Or in other words, the environment, cache sizes, data layout and a lot of other factors have a huge influence on what can be really achieved with a hardware platform. Some people have realized that a fast processor in a PC does not make it faster as long as the bottleneck is somewhere else.

So first of all you can do this stuff easier and faster on a quad or double quad core. Second the car looks even not as good as a car in a next-gen Console racing game. This car looks pretty boring with not a lot going on on its surface. Check out Gran Turismo HD screenshots. You can see the environment reflections and the environment color influencing the car.
Advertisement
Quote: I doubt that the CELL is faster than a quad core in practice.
It is like a vitamin pill package: you can read on the package what is in the pill but you do not know what is really absorbed by your body :-). People buy the pill because they believe that what is described on the package would really end up in their body :-). So the description what is in the pill let's them mentally skip the part what is really absorbed by their body. Now think of the PS2 as a super-computer and the description of its raw computational power is the description on the package of the vitamin pill package. Now go slowly one step further and think about the PS3 ...
Or in other words, the environment, cache sizes, data layout and a lot of other factors have a huge influence on what can be really achieved with a hardware platform. Some people have realized that a fast processor in a PC does not make it faster as long as the bottleneck is somewhere else.

So first of all you can do this stuff easier and faster on a quad or double quad core. Second the car looks even not as good as a car in a next-gen Console racing game. This car looks pretty boring with not a lot going on on its surface. Check out Gran Turismo HD screenshots. You can see the environment reflections and the environment color influencing the car.

as i said above
(seeing is believing, im a great believer in facts + not theory)
i dont give a monkeys about gflops etc, all i care about is actual results, ie what it can actually do at the end of the day. im not knocking what ppl are doing WRT raytracing on the pc as some stuff ive seen is brilliant, but it does pale in comparrison to what ive seen done on the cell (albeit not in the flesh, perhaps its all bogus + done offline, until i get my hands on one i dont know, i have to assume theyre not displaying false info )

yes Gran Turismo HD is amazing (in fact the first time ive looked at a game + couldnt tell instantly if it was an actual photo or not)
hmm wasnt gthd (just a modified gt4) running on the ps2, also god of war2. the (underpowered) ps2 did end up producing some very nice things.

Quote:Original post by Johnhl
Well this demostration is just a interesting novelty now it does lend credibilty to the idea that the playstation 4, xbox720 or what ever consols may exist 7-10 years from now (hell could even be a wii2) could see the first real ray traced commerical games.


Raytracing is not going to outperform a reyes renderer, i.e. prman - pixar's renderman compliant renderer (which most people incorrectly call renderman). Prman does have some extensions to allow you to use raytracing, but it is largely a hack. The quality of the rendered images you get from prman are simply better than anything you will get from raytracing in the same amount of render time.

Entropy (from Exluna) was again another renderman compliant renderer, however it was based around ray tracing. Whilst it was always cool to play with, the render times were always significantly longer than with prman. (incidentally, Pixar killed off the Entropy renderer a few years ago with a lawsuit or two as i recall. It's interesting to note that http://www.exluna.com/ now gets re-directed to nvidia's Gelato renderer....).
Quote:Original post by zedz
as i said above
(seeing is believing, im a great believer in facts + not theory)
i dont give a monkeys about gflops etc, all i care about is actual results, ie what it can actually do at the end of the day.


Have you read the paper I have referred at least twice in this thread? http://graphics.cs.uni-sb.de/~benthin/cellrt06.pdf

In that paper they manage to optimize Cell code enough so that each SPE can provide roughly as much performance than single core K8. As Core2 has twice the SSE throughput a quadcore Core2 can provide as much real-world ray tracing performance than similarly clocked Cell with 8 SPEs. I have personally ran a few real-time ray tracers on K8 and Core2 and can confirm that the latter has double the performance in real world situations.

Of cource when you start hitting memory bandwidth wall (it is simple with lots of textures) that quadcore will start loosing speed much sooner than Cell. Also new AMD quadcore Barcelona and Intels 45nm Core2 upgrade will have SSE4 that gives quite a few nice instructions that can make things a bit more efficient.
Quote:Original post by hoho
Of cource when you start hitting memory bandwidth wall (it is simple with lots of textures) that quadcore will start loosing speed much sooner than Cell. Also new AMD quadcore Barcelona and Intels 45nm Core2 upgrade will have SSE4 that gives quite a few nice instructions that can make things a bit more efficient.


That's quite true, but then again a quadcore doesn't have exactly 4 cores available, it has a bit less than that + it's bandwidth(which doesn't even come close to cells bandwidth) doesn't scale with the number of cores.
Now all of this is purely academic since in the current way of doing raytracing you really do need a memory bandwidth close to 1Tb/s for RTRT and the processing power close to that of 10 cells, so we still got some bit to go before we are there.
But i do feel optimistic for the PS4, it could possibly prove to have enough power to transition to full RTRT.

Quote:
Have you read the paper I have referred at least twice in this thread? http://graphics.cs.uni-sb.de/~benthin/cellrt06.pdf

yes ive seen it, its a old paper (ps3 didnt exist then) they benchmark with a 2.4ghz cell + whilst the results are impressive.
im sure what with cell being in its infancy then, theyve improved a lot on their knowledge + tools that the situation has improved heaps today. ( whilst u could argue that intel core2 + 4 are also different to previous cpu's the change isnt as radical as cell's )

ill be getting a ps3 ( ive gotta lay my first $1000 on a bet though, nz to lose the rugby world cup, a bet i wouldnt mind losing )

Quote:But i do feel optimistic for the PS4, it could possibly prove to have enough power to transition to full RTRT.
the ps3's got enuf power today, true the apps will be crap, but a snooker game ( which is an ideal showcase for raytracing thinking the balls reflections of other balls + environment ) should be achievable.
btw i released this week a billards flash game for the psp http://files.filefront.com//;7204822;/
true framerate suxs, esp when the balls are at the bottom of the screen :)




This topic is closed to new replies.

Advertisement