OTOY Forums

glimpse wrote:Single CPU, 7 watercooled 980Tis ~> 1000 in terms of OctaneBench.
Total budget under 10k$ (the best value You can get for Your money).

Any specific parts list you'd want to share. I've seen a wide gamut of motherboards and configurations most limited to 4 double cards...but have seen up to 8 on a 4u rack system.

3dreamstudios wrote:

glimpse wrote:Single CPU, 7 watercooled 980Tis ~> 1000 in terms of OctaneBench.
Total budget under 10k$ (the best value You can get for Your money).

Any specific parts list you'd want to share. I've seen a wide gamut of motherboards and configurations most limited to 4 double cards...but have seen up to 8 on a 4u rack system.

Use single slot moded GPUs or Kingpin cards that comes out of the shelf with single slot bracket included if You want to keep warranty =) x99ews works fine.. but if You wish more than 7 GPU..supermicro boards will come handy =) also ethere are backplates with up to 18 slots..but You'll hit other issues..

If you put Xeons in the same box, loading time and transfer will slow down.
I personally would not recommend that for Octane. A fast overclocked i7 will be much faster (Single threaded) then a Xeon.
Keep us updated with your rig.

Refracty wrote:If you put Xeons in the same box, loading time and transfer will slow down.
I personally would not recommend that for Octane. A fast overclocked i7 will be much faster (Single threaded) then a Xeon.
Keep us updated with your rig.

I have not heard this. What would cause a slower load with Xeons? My current machine is dual xeons. I know that a single i7 that has less cores but individually are faster than Xeon's would be better for modeler and the slice tool...lol.

If you could elaborate or point me to some information I could research on that would be great. Even if Xeon's share pcie lanes if they are not being used at the same time I can't imagine this being the case....

Thanks....

speaking simple, 200$ i5 could do work faster than pair of Xeons costing 2000$ pluss - that's the way OctaneRender works now.. so for Octane build.. You do not need to go expensive route.. it's better to stick with CPUs that offer You best single core performance & then good enough performance on four cores.. There rest of cores are simply sitting waiting doing nothing.. so unless You need them for other tasks..do not buy them for Octane (right now, as things might change i nthe future =)

glimpse wrote:speaking simple, 200$ i5 could do work faster than pair of Xeons costing 2000$ pluss - that's the way OctaneRender works now.. so for Octane build.. You do not need to go expensive route.. it's better to stick with CPUs that offer You best single core performance & then good enough performance on four cores.. There rest of cores are simply sitting waiting doing nothing.. so unless You need them for other tasks..do not buy them for Octane (right now, as things might change i nthe future =)

yes I understand this. But my machine will be doing Native LW rendering (CPU) and other CPU intensive tasks on other occasions.

Some scene compilation time where Octane prepares models for rendering are single threaded or poorly multi threaded, meaning Octane will sometimes use only 1 core to compute it and this would be marginally faster on the best non Xeon CPU since it has a higher clock and boost clock.

In my case of dual Xeon E5-2687W scene preparation CPU usage is as follows:

LOADING..... 3% CPU usage ( I guess this is just pure object loading into memory, no special computations here, I guess SSD and memory speeds are crucial here )
UPDATING SCENE INTO GPU..... 14%-25% variable ( we see CPU multi threading here, it uses max 4 out of 16 physical cores )

Devs have noted that they haven't touched this piece of code in years and that there was a room for improvement. So if you go with Xeons things can only improve in the future and you would benefit from the improved compilation code.

Another way to make use of more Xeon CPU currently is to render using more instances of Octane. I use this when scene compilation for each frame is substantial. ( when rendering animated fur )