External Graphics Cards PC

[Notiusweb]

Itou31, what exactly happens once you bake a texture. Is it exported and then it acts as an importable texture map for the given mesh that it was rendered on? If it is an exported texture, is it a giant 'full body texture', or a collection of texture materials, like how Daz meshes typically have.

Also, do you forsee getting new cards to take over for the 3GB ones you have, It might help with sone if the rendering quags you have at times. Hopefully V3 build progress resolves the GPU timing interrupt thing ("the freeze") we have on the baking camera. I saw big performance gains going from 3-6GB when I took my gtx 660 Ti out of the equation. Although, in my case, that card was a very handy testing card when I was experimenting with getting 10 GPU to 11 because it was lower demand. So I could run tests in principal and then try with the bigger cards after.

[itou31]

Hi,

I don't know how to use the bake camera feature. It was just a test to see what happen with my system. I though that it could be a "daz texture atlas" with a pre-rendering (not sure how to use it after). Need more information.
I don't know about getting 6GB cards. Just wait for the progress on the V3 as I see great performance over the V2, but some fears of the freeze...

[Notiusweb]

Hi, okay that's where I am too. Im on the same page as you with Daz, I'll report back with anything I can learn too.

[Gray_Parrot]

Hi everyone,
I've been following this thread and trying build up as much information as I can to help me add more GPUs to my current setup. Notiusweb and itou31, I would be very keen to have your advice about the Amfeltec GPU Cluster in particular.

Currently I have 2 Titan Z cards cooled by air on a Sabertooth X79 motherboard with one slot left open. I have been thinking that buying three of the 980ti Hybrid cards would give me good performance in the case, where there would also be mounts for the radiators. Then I was planning to add the Amfeltec and run the Titan Zs externally. This way I would double my render performance and still have room for 2 additional cards in the future.

Does this sound like a good idea? Can you think of any significant limitations to this hardware plan?

I greatly appreciate any feedback in advance.

GP

[glimpse]

..about the Amfeltec GPU Cluster in particular.

Currently I have 2 Titan Z cards cooled by air on a Sabertooth X79 motherboard with one slot left open. I have been thinking that buying three of the 980ti Hybrid cards would give me good performance in the case, where there would also be mounts for the radiators. Then I was planning to add the Amfeltec and run the Titan Zs externally.

Does this sound like a good idea? Can you think of any significant limitations to this hardware plan?

one of the limitations is upload speed, 1x to be precise. Going this route seems to be logical solution, but maybe adding one dual slot & one single slot GPU (maybe watercooling both Zs) in the case, seated directly in the motherboard & using 4x splitter (let's say from the same Amfeltec) would be better idea?

GPU cluster is nice & fairlly organised - interface is prety reliable, but.. that upload speed might cause some issues (especially with new alpha builds).

Spliter's cables are 12 inches (if I remember good) thus for placing GPUs You do not get too much of flexibility. Other thign is that splitter seems not to keen in "cooperating" with riser cables (read: no possibility to extend).

it depends how clean of a result You would like to get..

[Gray_Parrot]

Hey thanks for the information Glimpse!

I have a long term plan of water cooling the Titan Zs, but I will have to wait until the summer when I have some downtime for that project. Water blocks would also help me fit all the cards into the chassis.

Just a quick follow up question about 1X speed: would this have a significant impact on animation load times? If I could "load all frames in Octane" would that reduce the gap in speed?

Finally, would it be a better idea for an animator to buy a second license to octane and just run two systems concurrently?

I appreciate the feedback,

--GP

[tipt0p]

Hi guys,

Sorry if this is off topic but i just wanted to find out if anyone is running 16x GPUs on a 8x lanes server board such as SuperMicro? I'm planning on installing 8 to 10 GTX 780/980 cards on the board but these server boards are 8x slots which obviously aren't going to fit, what's the best work-around solution for this? beside 16x-to-8x adapter cable?? And have you noticed any loss in performance if you're using risers / splitters?

Thanks

[Gray_Parrot]

I'm curious to know the answer to this too.

[Notiusweb]

Hello, I typically think the more the merrier, and 1x is pretty much without any meaningful drawback. However, as it stands right now in the development phase of OR3, there are issues with setups involving PCI 1X speed. This may be rectified in some time (hopefully sooner, rather than later), but it remains an unknown. What I am finding is that any high res render is unfortunately subject to crashing. 2 other users have reported the same thing, so it is not unique to my setup. Thus, if this issue is never 'rectified', it could be that you are better off having a series of cards with a 16x connection than a 1x connection as this will kill a high res render. PC hangs, sometimes immediately, and goes into a BSOD or straight reboot. And, here is the key thing, this all occurs while the V2 builds do not crash at all.

I would say if you have no plans to use OR3 anytime soon and are completely happy with current V2 build, then yes, more GPU at 1x is great, and the Amfeltec is great. I have 1 Titan X (16x) and 6 Z (1x) and sometimes a non-intensive render at 1080p takes 5 seconds (2,000 - 3000 samples). I notice with the X can with its 16x speed make the to be rendered images appear in the viewport faster than with the Zs connected at 1x. Sometimes the render finishes right as the image appears in 5 seconds or so with my Zs at 1x

As to 8x vs 16x, I never really saw anyone have that setup where they ran a 16x to 8x connection. I never read complaints or saw negative comparisons other than it meant very little to gamers to have 16x vs 8x. In a nutshell, the 8x would probably be just fine. Have no idea if the same Octane Render 3 issue that currently occurs with a 1x speed would also occur with 8x or 4x speed. Those who have brought to development's attention, including myself, only have 1x and 16x I think.

My thoughts here are really subject to change. If, for example, the developers found a solution, a lot of what I said about the unknown factor gets dismissed.

[Seekerfinder]

Hi guys,

Sorry if this is off topic but i just wanted to find out if anyone is running 16x GPUs on a 8x lanes server board such as SuperMicro? I'm planning on installing 8 to 10 GTX 780/980 cards on the board but these server boards are 8x slots which obviously aren't going to fit, what's the best work-around solution for this? beside 16x-to-8x adapter cable?? And have you noticed any loss in performance if you're using risers / splitters?

Thanks

I would say the best workaround for you depends on the specific Supermicro board you're using. If it has double spacing between the pcie slots, I would just cut the back of the slots to allow the 16x cards to physically fit. They should work fine and operate at 8x speed that way.

If the slots are single spaced, use 8x to 8x risers (coax type are the best but they're expensive). You may again have to cut the back of the riser's slot for your cards to fit. There may be 4x to 16x or 8x to 16x risers available but shielding is important and some of them are not good quality.

Best,
Seeker