Cubix GPU-Expander for Octane Render Tweet

[ycarry]

So (just to understand) as I my pc have 1 PCIe x16 and 2 PCIe x1 slots,
I can add one Xpander (with 2 GTX480) to each PCIe x1 expansion slot
and keep my GTX260 in PCIe x16... ????
and Octane benefit of all cards?

[radiance]

So (just to understand) as I my pc have 1 PCIe x16 and 2 PCIe x1 slots,
I can add one Xpander (with 2 GTX480) to each PCIe x1 expansion slot
and keep my GTX260 in PCIe x16... ????
and Octane benefit of all cards?

yes

Radiance

[ycarry]

Wahooo
please EricNV, how much Shipment and VAT for France?
Note: here in France I can buy GTX480 for 490€ (include 19.6%VAT and free shipment)...

[ROUBAL]

Hmmm, on Cubix site support, I read :

Connect as many Xpanders as you have available PCIe x 8 slots in the host computer

PCIe x 8 slots, not PCIe x1 ! This has to be confirmed, but it is written.

For what I have been told recently, the cable link transfer speed is not important during the rendering process, but it is much important during the loading of the scene.

I am currenlty doing a test on an animation of 35 frames of a scene requiring 290 MB of VRAM, on a GTX 260 mounted in a Cubix-Xpander which is connectet to a PCIe 16x 1.0 : For a rendering time of 4 seconds (at 32 samples per pixel), the loading time of each frame is 124 seconds ! More than 2 minutes per frame .

At this speed, the whole 35 frames animation requires 72 minutes of loading for 140 seconds of rendering !

For more samples per pixels, the ratio will decrease a bit, but not so much.

If you consider that the tested scene weights only 290MB, on a PCIe x 16 (even if it is a 1.0) the loading time(so, the transfer speed of the PCIe link) seems to be a very critical factor ! I doubt that PCIe x 1 are usable.

For what I have tested so far, my fears are confirmed and the current loading time for animation is really beyond what I consider as acceptable.

[GeoPappas]

If you consider that the tested scene weights only 290MB, on a PCIe x 16 (even if it is a 1.0) the loading time(so, the transfer speed of the PCIe link) seems to be a very critical factor ! I doubt that PCIe x 1 are usable.

I'm not sure if it is the PCIe transfer speed that is the bottleneck. According to the specs, PCIe x16 1.0 has a transfer rate of 4 GB/s. That means that 290 MB should transfer in < 1 sec. PCIe 2.0 doubles the transfer rate (to 8 GB/s for x16).

There must be something else slowing it down.

[ycarry]

PCIe x 8 slots, not PCIe x1 ! This has to be confirmed, but it is written.

But I read PCIe x1 connector card here:

from Online store

[radiance]

our units here came with 8x PCI-E and they work fine.
i believe they have both 8x and 1x options.

You should be able to use octane just fine on a 1x PCI-E connection, you'll just have to wait a minute instead of 2-3 seconds for the mesh to appear when you first click on your meshnode.

Radiance

[ROUBAL]

What about animations ? Currently, frames on a slight scene are loaded in more than 2 minutes each, not only the first one. Is it the necessary time for Octane to update the geometry ? Can't it be reduced ?

[radiance]

What about animations ? Currently, frames on a slight scene are loaded in more than 2 minutes each, not only the first one. Is it the necessary time for Octane to update the geometry ? Can't it be reduced ?

do you mean the loading/voxelising or the actual upload to the GPU when you click on the meshnode the first time ?

Radiance

[ROUBAL]

I mean the loading/voxelising time.

For a still picture, it is done once, and after that if you spend time to adjust your materials parameters, you almost forget the loading/voxelising time, but when you have to render an animation, this loading/voxelizing time is repeated for each exported frame and then, it is another story : I have just finished rendering a 35 frames animation test, and while I got 2 to 4 seconds of rendering per frame at 32 samples per pixel, The loading/voxelizing time for each frame was 124 seconds... for a scene in 1024x576, with only a character in a empty screen (290MB).