[dmeyer]

numerobis, is 4 the maximum number of gpu cards possible in a pc?

I've run up to 7 with a PCI-E expansion box.

[numerobis]

numerobis, is 4 the maximum number of gpu cards possible in a pc?

The limitation of 4 cards (or two dual cards) applies only to SLI. CUDA/OpenCL is accessing every GPU separately and the number of cards is only limited by the number of PCIe slots and PCIe lanes.
The number of lanes or PCIe speed should be no big problem if only the GPU RAM is used, but i think as soon as something like unified memory will be available this bandwith limitation could be a problem.

[AndreD]

Just 2 questions:
-how much??
-how long??

[eric nixon]

thnx for info.

I dont think unified memory will happen, would need support from intel.

[choo-chee]

well it looks like something we all wait for !
but I must ask (as I'm no hardware expert...), what is needed to use it on a typical interior scene (as for exteriors FIRE now is quite good..) ?

[CodyKallas]

This is good to hear. As I was just thinking today modo 801's speed is sooo much faster that its worth the extra 5 minutes to do in photoshop. But still can not mimic the quality of maxwell. So to hear this is promising. I was just about to jump on the Octane ship. Still might, but I might hold off now until I hear a date for the maxwell GPU edition.

[eric nixon]

http://motherboard.findthebest.com/l/62 ... -Extreme11

This mobo has 7 x PCIe3.0/16x slots.

hmmm...

and this dual-cpu mobo has 7 slots too;
http://motherboard.findthebest.com/l/64 ... Z9PE-D8-WS

[dariolanza]

Hi all,

About your questions regarding the GPU development:

What is that video?

It is a very basic demo of a prototype of Maxwell running on a GPU device.

When it will be released?

It is a prototype, which means that it is neither an Alpha version yet. There are many barriers to break and features to implement before we can define a public roadmap.

Which technology does it use?

It uses CUDA 6.0. The demo is running on a nVidia Quadro K5000, but it also runs in cheaper devices like the nVidia GeForce family.

Will it run on ATI cards?

CUDA is a proprietary technology of nVidia, and can only run on nVidia devices. We are aware it means that this technology will not run on machines without nVidia hardware (i.e Macs, or windows laptops with Intel graphic chips). We will explore other alternatives that can run on ATI cards and other devices. One of these alternatives is OpenCL, although its future is unclear. The fact that almost the totality of commercial renders available use CUDA reveals that so far CUDA is so far more suitable for raytracing.

Will it replace the main CPU renderer?

So far it is a prototype useful for previewing purposes, as an alternative to the Fire Draft engine. Whether it will be useful for final shots is an open question.

Will it have any limitations in comparison with the CPU engine?

Probably it will, although it is soon to know details about that yet. In general GPUs are more limited than CPUs on a few important things, mostly in terms of memory. Therefore it is very likely that big scenes with many polygons and textures will not fit in the GPU. Multilight might also not be possible, among other features. Also there is a general perception that GPUs are faster than CPUs and that is not true. GPUs are very fast at some things, while are slower than CPUs at others. There is a lot of hype about GPUs that hides this fact, typically SIMD architectures are very good when all the cores are doing more or less the same thing. In a raytracing context it means that the performance of GPUs is good in simple scenes where rays follow similar paths but the more complex the scene is (not only in terms of geometry, a scene with a few polygons with glass, mirrors and small holes could be very complex in terms of light transport) the less the performance of GPUs is. This is a general rule with many exceptions, a full discussion on this topic is out of the scope of this FAQ.

Dario Lanza

[choo-chee]

clear answer... I guess for now we are still fighting anthrax and public enemy
("turn it up... bring the noise!")

[numerobis]

So far it is a prototype useful for previewing purposes, as an alternative to the Fire Draft engine. ...
Will it have any limitations in comparison with the CPU engine?
Probably it will ... Multilight might also not be possible, among other features.

so, FIRE only... and this sounds more like Maxwell v4 or v5... (and i thought i could quit my corona testing now... )

But it's good to see some results from the development! Thanks!

Concerning the limitations,... it seems that GPU enginges in general are adapting more and more features that where "impossible" on GPU a few years ago.
(Displacement, channels, instancing, participating media,...) Btw. Arion is supporting light layers as far as i know, at least for 10 lights, so why shouldn't it be possible to implement multilight? (if the memory problem can be solved in the future)

[choo-chee]

I wonder if it's possible to have the GPU render some kind of fast pass without too much reflections etc. so it can do some kind of merge mxi with the CPU render and get rid of the noise ...

[Mihai]

numerobis, is 4 the maximum number of gpu cards possible in a pc?

I've run up to 7 with a PCI-E expansion box.

How much power do they all draw at full blast?

[dmeyer]

numerobis, is 4 the maximum number of gpu cards possible in a pc?

I've run up to 7 with a PCI-E expansion box.

How much power do they all draw at full blast?

Too much. Typically 1400-1800w when running full tilt.

[Fernando Tella]

Too much. Typically 1400-1800w when running full tilt.

Bah! My iron eats more than that.

[photomg1]

I wonder how the Arion guy has managed it ... cpu + gpu hybrid engine that also works cpu or gpu only.

Everybody else (octane , redshift , vray etc ..) its one or the other I find this most perplexing .