Prev: FAKE CONFERENCE 2nd call - INFORMATICS 2010: submissions until 31 May 2010
Next: Micro Men on BBC4 today 22:30
From: Skybuck Flying on 26 May 2010 18:47

"Nicolas Bonneel" <nbonneel(a)cs.ubc.ca> wrote in message
news:htk745$3nm$1(a)swain.cs.ubc.ca...
> Skybuck Flying wrote:
>> "Nicolas Bonneel" <nbonneel(a)cs.ubc.ca> wrote in message
>>> I don't say "everything is solved". I just say there *are* efficient
>>> compression schemes - as in the links I sent, with octree based
>>> compression. So maybe stills things have to be done to reach Shannon
>>> limit, but if people *are* able to render billionS of voxels, it *means*
>>> it is sufficiently compressed for interesting use with current hardware.
>>> If it was *not* compressed, a 8192^3 voxelization would take 2TeraBytes
>>> (with just 1 float opacity per voxel, without any color) which obviously
>>> does not fit in the gpu memory.
>>
>> Perhaps, but perhaps the renderer is also "cheating" by using the CPU to
>> do the decompression.
>
> Instead of saying "perhaps", read it!!!!!!

No time for it, besides it contains way to many details.

> If the data was decompressed on the CPU, there would be no way to send
> that amount of data on the fly to the GPU.

It could do it partially/little bits at a time but that would not be very
usefull.

Actually I think the documents mentioned "blocks" being decompressed on the
cpu and being fully sent to the gpu.

> They achieve at worse 20fps on this old hardware.

20 fps is already very low, and won't be enjoyable for shooters.

> If you meant "compressed", then it is completely fine to compress data on
> the CPU.

I remain unconvinced... a demo would answer these questions what cpu
utilization is actually done.

There is hope though for these kinds of renderer's at least for the cpu...
because of multi-core...

However for now systems only have 1 memory and therefore the memory will be
the bottleneck even on multi-core systems.

The proof is in the pudding ! ;) :)

Bye,
Skybuck :)


From: Skybuck Flying on 26 May 2010 18:49
Oh I forgot to mention another important problem with these shaders and
custom renderers:

"Re-usability of code".

It will probably be very hard/difficult to re-use this code.

An API would be much easier to re-use.

So in another words: Shaders do not seem to be the most re-usable codes...
especially CG seems to be lacking the "units/objects" concepts.

Could be a big obstacle.

Bye,
Skybuck.


First  |  Prev  | 
Pages: 1 2 3 4 5 6
Prev: FAKE CONFERENCE 2nd call - INFORMATICS 2010: submissions until 31 May 2010
Next: Micro Men on BBC4 today 22:30