CUDA cores / RenderTime relationship?

[rappet]

using the same gpu at the same clock rates and having enough memory bandwidth it is just like that. but usually gpus with different cores have also different specs in terms of vram speed and bandwidth, so the picture is a bit more complex; if you look at the benchmark chart, you can get an impression: http://www.refractivesoftware.com/forum ... f=9&t=7383

This relates to the question I'm posting in the forum. How much improvement in terms of time/% do I get going from 3x580 GTX 3Gb to 4x580 GTX 3gb?

Who are allowed to see these numbers? Perhaps I just have to joint another forum, or is it for already established customers?

Thanks!

Hi,
 This was my result a while ago:
16.72 Ms/sec (4x580, OR 2.57, driver 285.62)
4.18 Ms/sec with just one GPU.
As you can see 4xgpu equals 4x 1 gpu
To my surprise with SLI enabled it's just a little difference: 16.53 Ms/sec
You can see more tests on this post:
http://render.otoy.com/forum/viewtopic. ... k&start=60
greetz,

[Gadgety]

Thank you! Thank you for posting your numbers! This is great to know.

(I cannot view the link, though "You are not authorised to read this forum.")

[karanis]

using the same gpu at the same clock rates and having enough memory bandwidth it is just like that. but usually gpus with different cores have also different specs in terms of vram speed and bandwidth, so the picture is a bit more complex; if you look at the benchmark chart, you can get an impression: http://www.refractivesoftware.com/forum ... f=9&t=7383

This relates to the question I'm posting in the forum. How much improvement in terms of time/% do I get going from 3x580 GTX 3Gb to 4x580 GTX 3gb?

Who are allowed to see these numbers? Perhaps I just have to joint another forum, or is it for already established customers?

Thanks!

Hi ,

Octane uses multiple GPUs almost linearly. I mean, if a scene is rendered in 10 mins with a single GPU, 2 GPU makes it 5 mins, 4 GPU makes it about 2.5 mins... 3 GPU makes it like 3.75 mins (assuming identical GPUs). I`ve read it in forums and experienced it by myself. Hope this helps...

[buzby]

[/quote]

Hi ,

Octane uses multiple GPUs almost linearly. I mean, if a scene is rendered in 10 mins with a single GPU, 2 GPU makes it 5 mins, 4 GPU makes it about 2.5 mins... 3 GPU makes it like 3.75 mins (assuming identical GPUs). I`ve read it in forums and experienced it by myself. Hope this helps...[/quote]

Thanks, but...does it make a difference in the rendering speed if those cards are all 1.5GB as opposed to all 3GB? What is the determining factor? Is it brute strength in the sheer number of cards, or is it further discriminated by factoring in the size of vram in any one card or all cards? i.e. could/should all the cards be 3gb or could/should 1 be 3gb and the rest be 1.5? Does it matter? The reason I ask is I am about to sell off my 3 1.5gb gtx580 to purchase 3 3gb 580 gtx cards.

[buzby]

So, you mean that it doesn't matter to have the same power in 1 card or divided in 2, right?
In that case would be better to have all the power condensed in 1 card to have more free space.

exactly.
just in case you are looking for a gtx 590, be aware that this one doesn't have 3gb vram at once - as most retailers suggest - but only 1.5gb per gpu, because it's a dual-gpu card...

So a single gpu gtx 580 with 3gb of Ram is more "powerful" than a dual gpu 590 with 3gb of Ram, (or for that matter a 690 with 4gb of Ram) divided x2 in regards to the way octane renders?

[buzby]

bump- anyone? need to know before purchasing cards

[glimpse]

dual cards like 590 are clocked downd a bit compared to 580. Why? probably because of heat or maybe other things..- noOne knows for sure, just nVidia.

690 scales more nicelly then 590to580 & runing nearly as fast as two 680 =) vRam will not influence speed at all..Though You need more time to transfer the information (what is logical)if You want to top out the vRAM.