Thinking of switching renderer (& building gpu system)

[urbgeek]

I'm looking to buy a new system, undecided between the building multi xeon cpu system vs multi gpu single cpu system and switching to octane.

I'm looking for that almost instant interactive feedback to improve my workflow. My budget allows for a 3 titan single cpu build, I often render large complex architectural scenes with dense vegetation, would a 3 titan machine give me the power I need for this type of work? I have concerns about the 6gb limit on the cards, I often max out my current system with dual xeon 24gb ram on the big V renderer. I've also read that potentially the new (unreleased) graphics cards will have the ability to use system ram as well as gpu ram for exactly this problem, so is it a bad time to build a system with titans?

I have spent months debating and reading as much as I can about gpu rendering, it would be great to hear some genuine thoughts of some octane artists.

[glimpse]

Hi,

3x Titans would cost 3k$ for cards alone. 780 6Gb has come - the price is way lower (600$). Instead of 3x Titans You can get 4x 780s. For what's left You could get better cooling (full watercooling loop) - might need some extra. This would allow to run at lower temperatures..

Octane 2 is going to have what is called out-of-core rendering:

"Out-of-core rendering – Scenes that don’t fit into graphics memory will be supported through out of core rendering, storing currently unused source data in host memory while swapping it into graphics memory when required." source: http://render.otoy.com/newsblog/

As for new hardware You've mentioned (if I remember good from GTC presentations online).. this is not comming soon..I think there's not so much info 'bout that (so it's hard to say) but from what I've heared the impressions is that we'll not see this for 2-3years at least (but definatelly need to dig better to be sure).

If You pair Octane with something like PhantomScatter that could generate .CSV, or let's say ForestPro, MultiScatter (if You're workin with 3dsMax plugin) to help with Your instance placement, there's a good chance You could manage pretty huge scenes.

tom

[urbgeek]

Hi glimpse, thanks for your reply. Yes I am an avid forestpro user.
The plan was to water-cool the 3 titans. I saw the 780 6gb, seems like a good price for marginally fewer cores, is this the only difference?

could be worth considering going for the 4 780's. The aim is to achieve interactive real-time feedback on large scenes, would this kind of setup allow for this kind of workflow. Or is it best to just stick with my current render and add more cpu cores?

[glimpse]

With quad Titan/780 rig interactivity is very high, I don't think You could come even close with CPU based rendering. Even V engines sees the future in GPU & they are pushing forward that too. 780 & Titan are not the same, but for Octane apart from a bit lower speed this new 780 gives very good value. There are very good reviews on TomsHardware, Anandtech & websites liek these where You can get into detail what is the difference between those cards.

It very much depend on what You consider being large. Never the less, I bet GPUs are the future & trying to adopt this technology will bring a benefit on the long run. You'll probably run into some issues or will not be able to do just everything what You're used to using CPU engines, but..with time this will probably come very very close.

[urbgeek]

I was kinda set on the tri titan setup, do you know if octane makes use of the double precision floating point?

quad 780's (plus change) vs tri titan .. tricky

[glimpse]

no, single precision only =) that's what matters to Octane.

as for three titans vs four 780..less power usage for Titans, but I'd rather take 4 780 with waterblocks then run 3Ts on air for the same price =)
You'd probably need to spend some extra for rads & so, but.. =)

what base system You're looking to?

[rappet]

I'm looking to buy a new system, undecided between the building multi xeon cpu system vs multi gpu single cpu system and switching to octane.

I'm looking for that almost instant interactive feedback to improve my workflow. ...

I have spent months debating and reading as much as I can about gpu rendering, it would be great to hear some genuine thoughts of some octane artists.

Two years ago, as a cpu renderer, I stumbled into this forum, read and ooooh-ed a lots of time, and immediately decided to buy Octane, and Ordered a custom built with 4 GPU.
It felt a little bit as a guess at that moment, because I did not really know what to expect.
I did not regret one split of a second.
And I can not image more of instant interactive feedback in workflow.
At this moment I have two Titans in a second machine, and 4 580gtx in another machine.
I can tell you that three Titans or 4 780's will rock even more.
When I was rendering CPU it was not unusually to wait an hour or at least 15 minutes.
Right now it is more like seconds, a minute or just a few minutes. Sometimes an hour for highres clean pathtrace renders,
But that would take so much longer with CPU
In cpu I had to render over night to get a short animation, no sleep to check if everything went OK during the night,
now I did one animation in just a hour. OK, it was nothing special compare to some guys showing there stuff,
But I compare it with the cpu render I did before, and that was with a little forest of trees, not just a simple scene.
Btw, I used to cpu render in Artlantis, and that should be the fastest (CPU) render engine.
I am also in archviz, and sometimes I want heavy scenes, and the limit of textures is reached with the 580 ( 3gb) and with 64 max of certain texture images to use. The Titans have 6gb and 144 max textures.
And limitations are getting minor due to ram and vram use.. You can read somewhere else on the forum.
The GPU has doubled up in speed and vram in very short time compared to cpu, that is just leveled up a little in the past few years.
And it is so much easier to upgrade GPU then CPU.
I just added two Titans in my 4 year old machine, and maybe next year or whenever I will replace the 580's if the new GPU's are so much faster. And with future networkrendering the sky is the limit.
Or My next machine will probably be a watercooled system... not that I am a specialist and I do not now all about it..
but I get wiser and wiser by reading a lot on this forum and asking questions. And Glimpse is one of them that is very helpful and the Octane community is very nice and that helps.
My advice is.. Go for it (GPU) , immediately.
Good luck with your choice.

See the near future features: http://render.otoy.com/newsblog/?p=414

[blklab]

Currently I use a dual 6 core xeon, but the new system I'm looking to build I'm thinking a single i7 extreme 4930k, 64gb & 3 water cooled titans (although now I'm thinking 4 780's)

But I still have my concerns from reading about octane just not being able to cope with large scenes, although I believe octane has improved massively since people have posted these comments.

I'm wondering if I should not gamble so much on build a gpu rig and build another dual xeon system with fewer gpu's this way I have the fall back of using cpu and networking with my current dual 6 core xeon system..

Rappet, thanks very much for your reply. Your reply has given me a little more confidence, good to know you took the risk and your happy you did.
I guess part of the fear is leaving what I know and gambling on something I don't know. Its hard to find any videos of multi gpu setup showing the interactivity of a larger complex scene, all the videos are of like greeble boxes or one single car... so its really hard to judge. I'd love to see an arch vis scene with vegetation on a multi gpu.. The memory thing is an issue but as glimpse mentions, 2.0 will solve this, and with passes it certainly has even more going for it..

[rappet]

Hi blkblab,

Maybe you have to adjust your workflow that you are used to with CPU rendering,
but imho the pros are much more then cons.. and the cons are getting less and less with more features in Octane

I think with the future Out-of-core rendering larger scenes will be less of a problem.

Below is from the blog: http://render.otoy.com/newsblog/
"Out-of-core rendering – Scenes that don’t fit into graphics memory will be supported through out of core rendering, storing currently unused source data in host memory while swapping it into graphics memory when required."

And using instances makes a lot possible... scattering around a few tress and showing a complete forest without having to eat all the vram

This is a great archviz animation by Gabrielefx (one of the yedi masters )
he always makes super vizs http://render.otoy.com/forum/viewtopic. ... va#p101177
But when you search the Gallery forum you will find more guys that make amazing stuff, and probably with pretty large scenes as well.

This is a animation example I made.. it is archviz with a lot of trees. This was even done without scattering.. Now I could do this much more efficient and with better highpoly trees,
but I got the job done and rendered it in an hour! (with 4 gtx580, that is to compare with 2/3 Titans; 4 gtx580 are just a little faster then 2 Titans) (OK, it was one hour and a half rendering time)
Oh, and sit out at least 20 seconds, then the real animation will start
http://render.otoy.com/forum/viewtopic. ... on#p138937

Can you give any idea of how large the scenes are that you want to render... is it large because of a lot of trees, or just a lot poly in buildings modelled, or other?

greetz,

[jblessing]

So how much of a slow down do you guys see using GPUs over the network?

We are thinking of building new systems right now too, and so we are considering a main workstation and a couple GPU render nodes (all with 4 titans or 780s each). We are debating between dual xeons and i7s as well. I know there is a 12 GPU limit right now.

We are using C4D as our host application. Has anyone had any experience rendering over the network with team render?