Refractive Software® is discontinuing Cubix Xpander support

[ROUBAL]

If I remember well, most quoted remarks here above were from me.

But at this time I used some useless parameters for exporting, and now my exporting time is divided by two.

I want to say that in my quoted posts I made a confusion between the exporting time and the loading time. As It was my first attempt to render animations with Octane, I measured the time from exporting start to rendering start, and didn't really notice that the trouble was with export.

In fact, the critical factor was the exporting time per frame, due to slow execution of Blender export Python script (slow interpreted language).

IT IS NOT the loading time on the GPUs ! When exporting, it is only CPU work, and has nothing to do with the speed of the PCIE ports and Cubix !

About the Magma box, its motherboard has 7 slots, but can support only 4 double width cards, like GTX 480 or equivalent. Added to that its Power supply is a 400 W ! So, you can connect the cards, but you can't even feed two of them ! This makes this product totally unusable for GPU rendering with more than one card... and it costs twice the price of the recent Cubix GPU-Xpander Desktop 16x version, supporting two big GPUs.

Added to that, even if it was possible top replace the PSU by a 1500 W power supply, as the only fans would be the fan of the PSU and individual fans of the GPUs the box would probably not be cooled enough for 4 cards, and would require modifications and additional extracting air fans...

[Carl S.]

Carl, it is physically impossible that PCI 1.0 is as fast as PCI 2.0 even onto medium scenes. Roubal said on another post:

I am not talking about the time it takes to export the scene each frame, but the time it takes to load the frame onto the card.
When you add a mesh to the scene via the nodes the scene does not instantly start rendering. When you click the node, the time it takes from the moment you click to the moment it starts to render is the scene being loaded onto the cards. Have a great day I can't be bothered about this anymore.

[ROUBAL]

Maybe I also forgot to say that the PCIE port on my Asus P5K on which my Cubix is connected is physically a 16x connector, but is electronically configurated in 4x.

This loading time has also been slightly improved in recent releases of Octane.

Obviously a true 16x PCIE port and a Cubix with 16x connection would certainly give better loading speed.

[bob30]

Perhaps, but I doubt that the Tianhe-IA and its 7,168 Nvidia Tesla M2050 are using Cubix Xpander boxes for running its 2.507 petaFLOPS.

[mcgarianala]

Hi!

I was looking at these systems as alternatives:

http://www.supermicro.com/products/syst ... GT-TRF.cfm

http://www.tyan.com/product_SKU_spec.as ... =600000150

http://www.cyclone.com/products/expansi ... 0-2707.php

Has anybody tried either of these products.

Best regards,

Gari

[ROUBAL]

I thought that the goal was finding a solution cheaper than the Cubix. Most of these seem to be more expensive and the PSU is also too weak in some cases for the number of usable PCIE slots.

[Inciner]

We're currently working with another company

Which one's GPU Expander is preferable for work with Octane?

[bob30]

I think we all agree that we want the best and cheaper GPUs utilities of the market

[adrencg]

I think the whole idea of the expander is to keep the heat generated from multiple cards away from the case that houses the cpu and other components.

I think the new gtx580 runs cooler than the 480, so putting 2 in the main case might be a viable option.

[Burkhard]

Feel free to comment on this information, and provide what you'd like to see added to our new solution instead. We will do what's possible and have a lot of flexibility when it comes to 'compactness'

Back to the roots:
I would like to see an option which allows me to switch between my different machines. Connecting to my MacBook Pro, my Office Desktop and my Windows PC
gives me more flexibility while working and with my trainings on Notebook.
That means also we need some equipments.