is 8 Gtx980ti will faster than network rendering ?

[hyubrad]

HI all

I'm is considering to buy 8 gtx 980ti for working but have a question .

Is 8 GTX 980TI in one desktop fater than 2 Quad GTX 980 ti of 2 desktops (by network rendering method) ?

[rappet]

As far as I know it it the renderspeed itself is the same, only loading the scene has some delay over network.
Fast network helps minimize this delay ofcourse.
I'll get back to you later with numbers... I got same card on master and slave and will test them seperately being the second card to render a scene.

PS
Did some test for you:
rendered the OctaneBenchmark with a local TitanZ,
and then added the two TitanZ's as network slaves.
If lineair the results should be triple renderspeed and it is.
The rendertime to load this scene over network was about 3 seconds.
(after 3 seconds the renderspeed jumped from 72 to 216 MS/s).

PS2
When I did some animation rendering using the network slaves and having just a short time rendering for ech frame (10seconds)... then the 3 seconds delay was a bottleneck.
So for short time rendering 8gpu local is preferable the 4 local and 4 over network, but for longer rendertimes (per frame) it does not make a difference losing a few seconds.
Costs for building a 8gpu PC is very high compared to splitting up?
And you need an extra licence for each slave fusing in networkrendering.

[hyubrad]

Rappet

we decided to buy 2 quad GTX 980ti desktops ,and an extra license after read your test . you did a good test for me

Thank you very much

[garytyler]

I've done testing with a similar setup as well. It was 4x GTX 980 Ti in one machine and 3x GTX Titans in another which also had a slower processor. With this setup while running the render from the 980 machine, the machine with the Titans and slower processor only contributed a negligible amount. I think the earlier generation GPUs load slower so with the slower processor and network communication, it got bogged down. In your case, and if you have similar processors, it should work out nice.

[hyubrad]

I've done testing with a similar setup as well. It was 4x GTX 980 Ti in one machine and 3x GTX Titans in another which also had a slower processor. With this setup while running the render from the 980 machine, the machine with the Titans and slower processor only contributed a negligible amount. I think the earlier generation GPUs load slower so with the slower processor and network communication, it got bogged down. In your case, and if you have similar processors, it should work out nice.

thank you for the tested

[gabrielefx]

the answer is YES

If you have a gigabit network and all workstations are plugged to the same switch you will have few seconds of delay before the Octane frame buffer is updated.
You can't compare the hardware bus with a wired connection.
I own 4 quad gpus workstation very silent. It's a risk having 8 gpus in one case, If your system will fault you loose an entire renderfarm. An 8 gpus barebone sounds very loud.

[glimpse]

one case with 7-8 GPUs water-cooled shouldn't be too noisy =) plus in case of 7 water-cooled GM200 based GPUs (aka 980 to TitanX depending on Your needs) should give You 1000+ in terms of OctaneBench. Check this for inspiration: viewtopic.php?f=40&t=43597&hilit=best+p ... 60#p249916

Lan rigs is an option & not a bad one, as Gabriel mentioned offering a bit of redundancy, however You can beat GPUs sitting inside single box..unless You have very expensive expander boxes.. =)

[hyubrad]

thank you all for answered

Finally ,my boss decided to buy 5 GTX980 ti not 8 GTX980ti , it's over budget
here's my rig

I7-5820K

ASUS X99-E

DDR4 - 32GB

480GB SSD *2 and Sata HDD * 2

GTX 980 Ti * 5

1200W power * 2