To SLI or not to SLI - *that* is the question.

[Streetwise]

I just acquired two GTX 780 6GB cards for pretty cheap. My current workstation setup has a Quadro 5000 and a GTX480 (both very similar cards from what I can find).

At this point, all I know is that I'm about to really speed up any RT work using Active Shade or RT GPU. My work doesn't consist of any animation but rather 3K x 3K still images. So what I'm primarily interested in is the ability to visualize lighting in real time, plus quick render tests to send off to the client for camera placement and lighting. As I understand it, for final production, CPU is still the way to go. So I'd like to know how best to use these new little cards... While I'm planning on replacing the Quadro and GTX480 with them, do I keep them separate? SLI them together? Maybe use them in SLI, but keep the Quadro for running the viewport as a third card? What would you do? What are some possible configurations and/or uses?

Thanks.

[benb32]

No need to use SLI (nVidia) or xFire (AMD) with any graphics cards - in fact such a setup might even cause problems with Max and Vray RT. All the graphics cores are addressed independently, memory is viewed on a per card basis and is not joined: i.e. two cards one 4GB and one 6GB, both added to the RT list, the largest scene that can be rendered is 4GB. So you could render a larger scene by removing the 4GB card from the active GPS list. This might change for the new release of Pascal GPUs, where normal RAM and GPU might be unified and different GPU memory combined - but I wouldn't hold your breath for it in the near future, plus is only likely to work well on motherboards with nvlink connectors, only available on systems which are astonishingly expensive!

Anyway, your best bet is to use the quadro for your Max view port, it is likely to be quicker (but do try the new cards as well to see how different it is). To stop lag in the max view port, you can reduce the ray bundle size in the RT render config, or remove the card from doing RT all together. Then have all your other cards active and doing the RT render. The more cards you can have the quicker it will be - motherboard space and power supply output not withstanding

Try not to have too large a resolution with your RT window, and you'll always get good feedback. Also, look out for installing the PhysX part of the nvidia drivers, some people here have reported issues where their cards are not seen properly and don't work with vray RT in process.

All the best Ben

[Streetwise]

Thanks for that. Exactly what I was looking for. I have an HP Z820 for my workstation. Pretty sure it will support 3 cards. I'll look into that. I'm reasonably sure though, that a GTX 780 6GB would be superior to a Quadro 5000. But like you said I'll have to test that out on viewport performance.

[Syclone1]

Check your power supply in that Z820, The Z820 I have, had to have the power supply upgraded for the 2 780's that I have in it. Also there is a cooling shroud that comes from the factory that you will have to remove as it wont work with one card, let alone 2.

-dave

[Streetwise]

Pretty sure it came with the 1125PSU, but I will double-check it. I've got two cards in it now and had to modify the shroud to make it work. Sounds like its not a big deal to remove it though?

[Streetwise]

So another question: Does ActiveShade use the GUI card, or the card assigned for RT?

[Sbrusse]

You have to choose the cards you assign in RT (both in active shade as in production render)
By default it uses all the cards available but to keep a fluid user experience it's better to uncheck the cards you use for your monitors