OTOY Forums

Dear all,

I've worked in the field of archiviz and CGI for digital media and animation for the past 8 years, in that time I have probably spent several thousand hours (at least) setting up unbiased renders for production, I'm now going through the process of migrating to the unbiased GPU rendering pipeline; having done as much research as I can I've decided to go with Octane, I have been blown away by the richness and quality of work produced by you guys and also the active community that exists here, a huge selling point in my book.

There are a few questions I have pertaining to this system:

• I need to get a new GPU set up to get started - so I have a few options. I either get 2 x GTX 580's each with 3GB ram or I get potentially 2 x GTX 590's, I believe that in the latter case, although the number of CUDA cores and the processing speed is higher, I would be limited to 1500MB of onboard memory, which may be a problem? Which of these purchases would be a better bet, I'm guessing 2 590's would be faster owing to total core count but might make it impossible to render certain scenes or at a high enough resolution for production?

• A lot of the work we do has to be for print, that means a render size of around 5000 x 3500 is unavoidable in certain situations, is this currently possible with Octane with the cards listed above (assuming that you have to downsample from higher resolution to remove some of the noise? I think this is probably the major issue for CGViz use.

• I might have made a total fail here but is there anywhere here where benchmarks for a representative scene are posted with different GPU setups? If someone could direct me there or to something similar it'd be much appreciated!

IF anyone could answer any or all of the above I'd be very grateful indeed, thanks in advance! I really am looking forward to getting started, as GPU speeds will only ever increase this is clearly a good time to get involved.

: )

P3D

p3d wrote: • I need to get a new GPU set up to get started - so I have a few options. I either get 2 x GTX 580's each with 3GB ram or I get potentially 2 x GTX 590's, I believe that in the latter case, although the number of CUDA cores and the processing speed is higher, I would be limited to 1500MB of onboard memory, which may be a problem? Which of these purchases would be a better bet, I'm guessing 2 590's would be faster owing to total core count but might make it impossible to render certain scenes or at a high enough resolution for production?

Personally, I would get the (2) 580s. As you stated, the 590s will probably only show 1.5 GB to Octane, but that hasn't been confirmed yet. You might want to purchase one 580 at first and see if it will fit your needs. You could then add the second one later.

FYI: If you purchase a motherboard with more slots, you could get more video cards installed (which will increase speed).

p3d wrote: • A lot of the work we do has to be for print, that means a render size of around 5000 x 3500 is unavoidable in certain situations, is this currently possible with Octane with the cards listed above (assuming that you have to downsample from higher resolution to remove some of the noise? I think this is probably the major issue for CGViz use.

It really depends on the complexity of your geometry and textures. I regularly render at 4000x3000 and only have a 1 GB card, but I don't do archviz. My scenes are much more simple in nature. But I plan on purchasing a 2 GB card soon anyway.

I would go for the 580s. Scene size matters. 3GB are enough for most scenarios. Keep in mind that the texture size can blow up your scenes. So 1.5 G are sometimes easy to break. Check out the space that the 580s require on your Motherboard.

The cuda cores give a good indication for speed.

Fyer flies can be an issue. If you are willing to photoshop a bit you might not have to downsample a lot.

Thank you guys for the responses, very much appreciated!

It's really good to know that resolution won't be too much of a problem, even if I could reliably render out at 4000 x 3000 I can use resampling software to increase that about 1.5x with no visible problems.. Same goes for fireflies, I don't think I've ever had an image output from a renderer that hasn't been extensively retouched anyway, most especially with biased render output I always have to manually retouch to create the richness and depth that seems so natural in unbiased renders.

In terms of cards I have also looked at twin 480's but then we're still back to square one in terms of the ~1500 mb RAM limit - the textures I use are usually quite large, but to be quite honest, needs must - I'm sure a lot of the sharpness could be restored in post if necessary.

I'll look out for the size limitations, I'm using a DX5SO which has a 4x PCI inbetween the two X16 slots, hopefully there'll be enough room for two.. but I'll definitely try one first and see how it goes.

Once again, thanks.

P3D