[sldev] Re: [VWR] Real-time Raytracing

[Dzonatas]

Jason Giglio wrote:
> Dzonatas wrote:
>>> It's nice, it's the future but if we're today not even ready to start
>>> using SSE as a minimum then why dream of what you can do with a CPU
>>> that's not even on the market yet?
>>>   
>> The problem is not just SSE but cache pollution even without SSE. 
>> More cores would allow a program to assign specific tasks to a core 
>> and optimize for that cache.
>
> You are missing the point.  SL is designed for the lowest common 
> denominator.  These octo-core chips might come out next year, but SL 
> is supporting CPUs from 5 years ago.  So like, bring this up again in 
> 6 years.
>

I understood that point, but SSE and # of CPUs are unrelated. The 
attempt to argue against # of CPUs (as being the future instead of 
sooner) with an argument about the state of SSE is not fair.

That's like saying why even bother mentioning the latest GPU when you 
know it won't plug into the motherboards that are of the least common 
denominator.

The point, as with the proof-of-concept, that seemed to be missed here 
is that one can use general RT algorithms on the multicore CPUs than 
build specialized code to support different vendor's GPUs.

RTRT has been around for awhile, but to do it under general RT 
algorithms on hardware that is on the market now is the proof that no 
special hardware is needed anymore just to do it in practical sense.

That's all for this

-- 
Power to Change the Void