Mounting Hardware (multi GPU & CUDA)

A public forum for discussing and asking questions about the demo version of Octane Render.
Forum rules
For new users: this forum is moderated. Your first post will appear only after it has been reviewed by a moderator, so it will not show up immediately.
This is necessary to avoid this forum being flooded by spam.
Post Reply
User avatar (Default)
bartosh44
Licensed Customer
Posts: 26
Joined: Mon Oct 31, 2011 8:06 am

Hi, i have few questions about buying new hardware. I know there are few posts like mine on this forum but every case is different.

I have got sli ready motherboard and GTX295 (480 CUDA cores) and i don`t know what to buy. I may buy one GTX 590 3GB & 1000 CUDA cores (it would give me 1480 in summary) or 2 x 580 1.5GB and this second option also will give me 1480 CUDA cores but it is faster about 5% than GTX 590. In GTX 590 i read here that Octane see only 1.5GB from one card so i thought to do this more scalable for future (near GTX 600) maybe i should buy only one 3GB GTX 580 and i should wait for Kepler but Kepler may be in march next year and lowest cards at that time and no one knows when would be release time of GTX 690 for example it might be in 2013 :)

So i`m confused a bit and i don`t know for example what is difference for Octane: 2 x same clas GTX 580 1.5GB and 2 x GTX 580 (one 1.5GB and second 3GB).

And i don`t know also what render time would be when i will have 1000 cuda or 1500 cuda. I don`t know also is that 3GB is needed for my scene textures or not.

In summary i don`t know what is hardware efficiency of Octane Render and it makes me confused and i don`t know what to buy to improve Octane (i have got about 700 euro to spend on it and i know its not much ;) so please do not advice me 3 x gtx 590 and one gtx 580 3GB becouse i`m to weak for that ;) ).
Top
User avatar
gabrielefx
Licensed Customer
Posts: 1701
Joined: Wed Sep 28, 2011 2:00 pm

go for 2 gtx 580 3gb...
it's better to have more memory than more cores...this is valid for cpus too.
quad Titan Kepler 6GB + quad Titan X Pascal 12GB + quad GTX1080 8GB + dual GTX1080Ti 11GB
Top
User avatar (Default)
bartosh44
Licensed Customer
Posts: 26
Joined: Mon Oct 31, 2011 8:06 am

Thx for answer.

Tell me please what is the difference betwen 2 x 3GB GTX280 and 1 x 1.5GB GTX280 + 1 x 3GB GTX280.

Why both should have 3GB? What if first has got 3GB and the second has got 1.5 GB?

I thought that Octane is using only from one card? If you are using GTX295 3GB that Octane sees only 1.5GB from first card? Am i thinking right?
Top
User avatar
kubo
Posts: 1377
Joined: Wed Apr 21, 2010 4:11 am
Location: Madrizzzz

it's not so much that it only sees an amount as that since the whole scene has to fit in both cards it can only use as much memory as the GPU that has the less vram, son in your second example you could only use 1,5 Gb. About how much memory you need, I only have 1,2 since the OS uses about 200Mb by default (if I turn off aero I can use a little more) then I have roughly about 1Gb if I want to use both cards (which is not all the time, just for finals) and I can fit a decent scene there, mostly is not about polygons but about texture size, load plenty of 4kx4k textures and you'll top your scene fairly quick, but if your render is going to be 2k there is no need for such huge textures anyway. Beside soon we'll have instances, so for most cases that'll mean that we'll be able to load tons of polygons with no problem (well besides render times, hehehehe).
windows 7 x64 | 2xGTX570 (warming up the planet 1ºC at a time) | i7 920 | 12GB
Top
User avatar
gabrielefx
Licensed Customer
Posts: 1701
Joined: Wed Sep 28, 2011 2:00 pm

bartosh44 wrote:Thx for answer.

Tell me please what is the difference betwen 2 x 3GB GTX280 and 1 x 1.5GB GTX280 + 1 x 3GB GTX280.

Why both should have 3GB? What if first has got 3GB and the second has got 1.5 GB?

I thought that Octane is using only from one card? If you are using GTX295 3GB that Octane sees only 1.5GB from first card? Am i thinking right?
Example:
GTX280 1.5GB + GTX580 3GB + TESLA 295 6GB = total 1.5GB

GTX580 3GB + TESLA 295 6GB = total 3GB

TESLA 295 6GB = total 6GB

...:)

Nvidia CUDA developers are working hard to implement the use of cpu memory too.

In the meanwhile Intel developed the 'Knights Ferry' Intel Many Integrated Core (MIC) board with 50 co-processors, sdk tools, etc....hard times for gpus lovers...
quad Titan Kepler 6GB + quad Titan X Pascal 12GB + quad GTX1080 8GB + dual GTX1080Ti 11GB
Top
User avatar (Default)
bartosh44
Licensed Customer
Posts: 26
Joined: Mon Oct 31, 2011 8:06 am

So...

@kubo
I would like to get animations and 2k is enough for me. In that case do i need 2 x gtx580 3gb? If i don`t need 3gb so maybe i should buy 2x gtx580 1.5gb or 1x gtx590? What is better for Octane in that case?

@gabrielefx
its wired :) but now i understood it :) I am curious what will be price of knight ferry and market will decide what is best.

I am Modo user and i like render of this software. Only thing i can`t stand is time of rendering. I bought Octane to give me possibility to faster this process (adding new hardware). I am glad that will be more options on market becouse that sick that in nowadays you are must wait for rendering most of your precious life (or pay fortune for render parks and other stuff). I think it should be realtime whithout waiting...
Top
User avatar (Default)
bartosh44
Licensed Customer
Posts: 26
Joined: Mon Oct 31, 2011 8:06 am

Please tell me also what is more important for Octane: to have 3GB VRAM or 1480CUDA (becouse i`ve already have one gtx295), becouse i can`t have both for this price...
Top
User avatar
kubo
Posts: 1377
Joined: Wed Apr 21, 2010 4:11 am
Location: Madrizzzz

well it is really really hard to answer that, it's more about your personal-profesional needs and what you can afford now. I usually go for speed, mainly cause most of my profesional renders come with crazy time schedules, but that means that I have to plan ahead very carefully to fit all in the amount of ram I've got. Some times I've got to break the scene in several parts and comp them out later. I know having the extra vram would be great but I couldn't afford it. On the other hand with enough vram if you don't mind waiting longer then you can add extra detail and don't worry so much about optimization. Like I said I don't think there is a right universal answer.
windows 7 x64 | 2xGTX570 (warming up the planet 1ºC at a time) | i7 920 | 12GB
Top
User avatar (Default)
bartosh44
Licensed Customer
Posts: 26
Joined: Mon Oct 31, 2011 8:06 am

Thx for helpfull advice. Now i know that i need more CUDA then more vram :)
Top
User avatar (Default)
bartosh44
Licensed Customer
Posts: 26
Joined: Mon Oct 31, 2011 8:06 am

What about loading scene? It depends on the amount of ram or vram memory?
Top
Post Reply

Return to “Demo Version Questions & Discussion”