Quad-Processor G5 -- Quick Question
- Thread starter TimD
- Start date
Martin
0
Hi,
Cheetah3D currently only uses two processors. But that will change of course. And Randy you are right with your assumption that it shouldn't be to complicated.
By,
Martin
P.S. Randy it looks like you've upgraded to a Quad G5. So you might have to change your foot note. :wink:
Cheetah3D currently only uses two processors. But that will change of course. And Randy you are right with your assumption that it shouldn't be to complicated.
By,
Martin
P.S. Randy it looks like you've upgraded to a Quad G5. So you might have to change your foot note. :wink:
RandyWang said:
TimD
0
Yup -- It rendered the scenes in two 'tranches' not four 
Also, the new video card speed wasn't paticularly overwhelming. The machine I was playing with (at MacUser's labs) had the GeForce 6600 / 256MB. More complex models in SketchUp went to wireframe when rotated, just like they do with my ATi 9600 / 256MB. The GeForce 7800 (and the Quadro) will be another thing entirely, of course.
So, I don't feel too much envy...I thought the 16x PCI-e bus would make a big difference, but apparently the AGP 8x slot in the last PowerMacs cannot be saturated at present (by the latest games, at least), so a faster graphics card in my current Mac should keep me going for a while.
Also, the new video card speed wasn't paticularly overwhelming. The machine I was playing with (at MacUser's labs) had the GeForce 6600 / 256MB. More complex models in SketchUp went to wireframe when rotated, just like they do with my ATi 9600 / 256MB. The GeForce 7800 (and the Quadro) will be another thing entirely, of course.
So, I don't feel too much envy...I thought the 16x PCI-e bus would make a big difference, but apparently the AGP 8x slot in the last PowerMacs cannot be saturated at present (by the latest games, at least), so a faster graphics card in my current Mac should keep me going for a while.
Martin
0
Hi,
wow that was easy. My beta version of v2.6 is already using a as many threads as processors are available. :wink: So all Quad G5 users will see a serious speed up with v2.6 when all of their processors will be used.
The OpenGL drawing speed of most of the modelers is CPU limitd. So I won't expect a to high speedup from a 7800.
Games are something totally different because for games you can use some high performace OpenGL extensions which can't be used in a 3D modeler.
By,
Martin
wow that was easy. My beta version of v2.6 is already using a as many threads as processors are available. :wink: So all Quad G5 users will see a serious speed up with v2.6 when all of their processors will be used.
The OpenGL drawing speed of most of the modelers is CPU limitd. So I won't expect a to high speedup from a 7800.
Games are something totally different because for games you can use some high performace OpenGL extensions which can't be used in a 3D modeler.
By,
Martin
Martin
0
Hi,
To get the maximum performace out of OpenGL you have to define UV coords and normals per vertex and send all the data via a single array to the GPU. Most games does that.
But in a modeler UV coords and normals are normally defined per polygon corner. That save's quite some ram and is more suitable for SDS modeling etc. But you also have to send every single vertex, normal and uvcoord via a function call to the GPU which costs alot of CPU resources.
Sure you can convert the data into an array before sending to the GPU but that all costs a lot of RAM so I avoid it.
Could I have the link to that thread. Maybe I can learn something.
By,
Martin
TimD said:
To get the maximum performace out of OpenGL you have to define UV coords and normals per vertex and send all the data via a single array to the GPU. Most games does that.
But in a modeler UV coords and normals are normally defined per polygon corner. That save's quite some ram and is more suitable for SDS modeling etc. But you also have to send every single vertex, normal and uvcoord via a function call to the GPU which costs alot of CPU resources.
Sure you can convert the data into an array before sending to the GPU but that all costs a lot of RAM so I avoid it.
TimD said:
Could I have the link to that thread. Maybe I can learn something.
By,
Martin
I wish.
I found a video of it rendering here, if you're interested - but really, I just assumed that Cheetah always rendered with two threads, instead of one per core (since it uses two threads on my iBook*). I just make it look like I own a quad so I look cool.
* A 1.33Ghz, 12" machine I bought four months ago today. I love it.
