Page 1 of 1

All GPU Memory available

Posted: Fri Oct 14, 2016 6:59 am
by artech7
I kind of understand why Octane can't use all of the collective memory in all of the GPU's installed in a system.

So, here I'm mostly making this thread for speculation and explaining.

Do you think Octane (or any other GPU based renderer) will ever be able to use all RAM available from every GPU installed? Why? Why not?

Re: All GPU Memory available

Posted: Fri Oct 14, 2016 11:15 am
by linvanchene
Update / Edit: removed some redundant parts.

Disclaimer:
This entry focuses not on the technological limitations but more on the state of the market and the interests of the companies involved.


Speculation:


For shared GPU memory to happen first the whole GPU industry would need to start organizing themselves and actually work together to create some common standards. Then the "organized" GPU industry would need to start working together with the CPU and mainboard industry to find common solutions and industry standards.

Currently this seems far away because one major company seems to dominate the advancement of the whole field. And maybe that one company is not really that interested in providing solutions like stacking VRAM.
Why would anyone pay 5000+ $ for a GPU with 24 GB vram if you could simply buy two with 12 GB for 1'000$ each and combine their VRAM?

The same thing happend with stacking cards for increased speed. Instead of purchasing one card with 3000 cores people started buying more affordable cards with 2000 cores and combine them to have 4000 cores.

- - -

The longer version:

Adding or stacking VRAM would require for GPU, CPU and mainboard producers to work together.
Currently not even a simple cooperation between GPU producers, toolkit developers and software engineers seems ready.
Comparison:
Gaming
You can plug in any GPU into any pc and use it for gaming not matter the generation.
There is complete backwards compatibility.
You can use a generation 1 card on a generation 3 pc.
You can use a generation 3 card on a generation 1 pc.

-

GPU rendering
You can not use a generation 3 GPU to render with your software without using a generation 3 developers kit.
You can not use a generation 3 device with generation 1 software because companies expect you to buy the latest version that includes support for the generation 3 development kit.


This all seems like the GPU are intentionally limited in their usage with the help of development kits.
Backwards compatibility is not guaranteed.

What needs to happen?

Companies in the GPU rendering industry need to start organizing themselves and create industry standards.
Examples of industry standards:
USB
Users can use USB 1 devices on a USB 3 slot.
Users can use a USB 3 device on a USB 1 slot.

HDR, 4K
All companies in the media industry agreed on the same standards to make sure media is functioning during the whole production process
Capturing devices, video codecs,, monitors, editing software, broadcasting equipment, home tvs,
Requirements for change:

- If the intention is that GPU rendering one day should be as versatile as CPU rendering it needs to become standard procedure that all major companies plan technology advancement years ahead and agree on such standards.

- Company A and company B need to start producing GPU that can be used with any rendering software.

- CPU and GPU industry need to start working together and creating optimized combinations of GPU, CPU and mainboard
Solutions for the very high end market illustrate that it is possible to have faster exchange speed between GPU, CPU and mainboards. But those solutions are far away from reaching the mass market.
- - -