OTOY Forums

Hello Guys,

I’m planning to build up a new render machine.
I will use it for architectural visualisation, indoor & outdoor.
No Animations, but pics will have high resolution (e.g. 7000x3000 pixel) for printing.
The software i use is Rhino, Sketchup and Adobe Photoshop, Illustrator and InDesign.
And of course i would buy a octane license, when i'm done.

My set will be something like that:

Win7
i7 2600/2700
Gigabyte Mainboard (not chosen yet)
about 16 gb Ram
and a CUDA based GPU

Of course the gpu is my problem. I hope you can help me to chose the right one with your experience considering my needs.

I thought about these cards.

2x GeForce GTX 560 2GB GDDR5 CUDA Cores: 336 or
2x GeForce GTX 560 Ti 2GB GDDR5 CUDA Cores: 384 or
1x GeForce GTX 580 3GB GDDR5 CUDA Cores: 512


I’m hopefully awaiting your suggestions.

Dude, there is no doubt that the best choice is the GTX 580 with 3GB of VRAM!
In addition to 3GB, it has 512 CUDA cores!
Let me remind you that Octane uses only the memory of one graphics card. The second graphics card is only to help in the acceleration render.
Two graphics cards 2gb each, does not mean you have 4GB for your projects. Hence the best option is for the 580 with 3GB.
I have a GTX 470 with 1.28 GB, but I intend in the near future to acquire a 580 with 3GB.
I hope it helped!

THX for your answer, marcio.

I've read some more threads about vram and the 580 and with your answer I'm sure now.
Let it be a 3GB Phantom !

Thx a lot !

I will go with 2x GTX560 Ti 2GB. You will have 768 CUDA cores and 2GB memory per card.
Or even better - 2x GTX570 with 2.5GB per card.
Just my 2 cents

"Let me remind you that Octane uses only the memory of one graphics card. The second graphics card is only to help in the acceleration render."
so, if I can find one day a cheap tesla when kepler will be there, I could mix a large memory card with a brand new kepler with low memory but beast in calculation ?

HHbomb wrote:"Let me remind you that Octane uses only the memory of one graphics card. The second graphics card is only to help in the acceleration render."
so, if I can find one day a cheap tesla when kepler will be there, I could mix a large memory card with a brand new kepler with low memory but beast in calculation ?

no
octane needs to upload the scene to each (active) graphics card vram individually!

by the way, don't expect miracles from kepler nvidia made significant changes to the architecture by lowering the shader clock to be equal to the core clock (up to fermi the shader clock was twice the core clock); therefore kepler needs twice the shaders to push the same performance than fermi. also the first kepler cards to appear on the market are mainstream models which will hardly beat the top-range fermi cards available today. the higher-end kepler chips are awaited to arrive q2/q3 this year, and will have about 30-40% better performance - in 3d-gaming. how cuda apps will benefit from kepler is yet to be discovered...

thanks for answer... even if it not the answer I would like...

HHbomb wrote:thanks for answer... even if it not the answer I would like...

what's your budget?

in fact have a i7 2600K with a gtx 580 (but 1,5gb..) and I wanted to see what are my options for GPU rendering (actually I use Vray)

HHbomb wrote:in fact have a i7 2600K with a gtx 580 (but 1,5gb..) and I wanted to see what are my options for GPU rendering (actually I use Vray)

the first gpu where you connect your display takes more memory using Max, about 300Mb. Then you have to consider this amount of vram for the Max viewport.
With a single 3GB gtx you'll have a slow system no matter how many GBs you have in your gpu, a Quadro 6000 doesn't help.
You can use the second gpu for the previews and when the scene is ready you will turn on the first one.