Indigo Renderer Forum

Hi all.
I´ve found in the web some OpenCL benchmarks test available for Mac:
http://www.insanelymac.com/forum/index. ... pic=182874
Most popular is "Galaxy", wich I ran on my MacBook Pro (2.5 GHz dual core and NVidia 8600 GT GPU), with these results:
- Single CPU: 5 GFlops
- Dual CPU: 18 GFlops
- GPU: 39 GFlops

With a 9600 GT Mac equipped, GFlops go even further! Download and test your own!

Cheers!

OpenCL is begining to have wide support: new GeForce Nvidia Windows drivers supports it, and AMD Catalyst will bring OpenCL support very soon.
OS-X has native support since Snow Leopard, and Linux has it too with NVidia recently released drivers.
This is good news, as you can see OpenCL is platform independent.

Erm, ... David "Dade" Bucciarelli, a dev of luxrender, did a quick OpenCL version of the Mandelbrot Set
while he is porting smallpt to OpenCL too. Smallpt is a GI renderer in 99 lines of C++ by Kevin Beason.

Take a look at the result and the video and then start to pray ...




take care
psor

Cool i like that.

Very impressive stuff, Psor, many thanks.
This is the beggining, maybe from now the developing of GPUs technology will become not driven by videogame requirements.
Now we can only dream about the future developments in the field of parallel computing. And Indigo is among these .

More memory in video cards would be nice thing to see.

I guess more memory isn't needed as you wouldn't transfer the whole processing to the GPU but instead
certain parts of it as ray intersection and KD-Tree for example. Of course if you wanna let your CPU sleep
and your GPU do the whole work it's another story. ;o))



take care
psor

@psor:

38 times speedup?!?

As Lord Crc says in that thread on the lux forum, don't expect this kinda speedup on a complex renderer.
There you might get 30% or more depends on your GPU and how many you've got I'd say. Non the less,
maybe in future we're not just talking about multi-core CPU but multi-core GPU. ;o))



take care
psor

Oh yes, we will!

psor wrote:As Lord Crc says in that thread on the lux forum, don't expect this kinda speedup on a complex renderer.
There you might get 30% or more depends on your GPU and how many you've got I'd say. Non the less,
maybe in future we're not just talking about multi-core CPU but multi-core GPU. ;o))

The interesting point here is having this performance a a Extra on Top Bonus "for free" with systems that already have a nice GPU

Are we talking about 30 faster than CPU only... so its +130% with CPU + GPU??

Also it is cheaper to build a Quad SLi/Crossfire Rig than a Quad CPU system...

I think it's 230+% with single gpu + single cpu.
Because gpu shold be a bit faster than cpu because there are more transistors.
Example
HD5970: 40nm - has 4 300 000 000 transistors + insane memory speed takes 294W
Fastest cpu SPARC64 VIIIfx: 45nm - has 1 200 000 000 transistors takes 35W energy

Guys, stop speculating you might be disappointed later. As Nik said in another thread already. It's kinda black
magic as you have to take care that the cpu is not idle or vice versa. And so lets see who will implement it in a
good balance and working and not just first. Take it easy time will come. ;o))



take care
psor

Sorry to bump this thread but Dade finished his port of smallpt and it's quite sick!

video / thread



take care
psor

That is mighty sick