My take on Arion

[treddie]

I don't think much of Arion's hybrid GPU+CPU approach to 3d rendering...My take on it is that it is mostly hype. GPU's are SOOOO much faster than any CPU that tagging on CPUs for extra performance is like adding a drop of water to the ocean. Sure, if all you have are CPUs, then go for it...But you won't see performance gains over any network-based solution that is not GPU-based. And by combining your CPUs with your GPUs, you will only see a slight increase in performance. And since PC's these days have to have at least one GPU, it might as well be a CUDA-capable GPU and that is not all that expensive. Especially since adding CPU's is VERY expensive compared to populating a bunch of free slots with relatively inexpensive GPUs. So if someone wants to run a non-hybrid GPU-based MLT and they don't have the CUDA-capable card, quit being cheap and just buy the damn thing!

And that logic applies to memory resources as well. If a hybrid approach is also going after any available main memory to help speed things up, that's analogous to telling your CPU to retrieve data that is 5 light years away, vs. cached memory on a GPU that is only 1 mile away! Networking only compounds the problem.

What do all of you think? I'm open to criticism.

[p3taoctane]

I have to agree. As GPU's get cheaper and have more ram they push that CPU argument further away.

Everyone has a different expectation and specific requirements for what hey need from their renderer. For me... as Octane develops their solution more and more with each release.... the speed, ease of use and quality are what I appreciate and use daily. The GPU side of the equation has it's hiccups with driver compatibility and ram etc... but the speed and instant feedback is a game changer.

[gabrielefx]

When I purchased Arion 1.6 I did some test comparing my dual Xeon 3.4Ghz exacore workstation and the first GTX 580 1,5GB.
Well, two Xeons were equivalent to one GTX580.
Today my new Xeons 2.2GHz octacores aren't equivalent to one GTX Titan.
The fact is that iray, vray-rt, live use the CPU.
Recently I did some tests with Live for Rhino and I can say that the CPUs work great with Cuda.
Because I use vray for some exterior renders I have to use the CPUs.
Also Octane uses all CPU threads during the geometry translation, I think that the Intel processors could improve the render speed.

[treddie]

Seems crazy then, that they are putting so much effort into it. That is a lot of money to get that all integrated into one seamless solution that may end up biting them in the end. The only place where I think it could be an advantage would be with already established render farms that have already invested a ton of money into a CPU solution, and are looking for a way to leverage that investment by eeking out as much performance as they can from it, until they phase completely into a GPU architecture.

[MDK]

GPU:s are getting way faster than CPU:s My workstation has 2 quadro 4000:s and 2 6core xeons. I got an alpha license for Redshift render last week. 1 quadro 4000 and redshift seems alot faster than 2 x 6core xeons and mentalray. I tried RS at home aswell with my gtx 780 and... well if you want that kind of speed with CPU:s you need a small render farm. However it might not be a fair comparison since mental ray kinda sucks. I suppose I would have to compare something like V-ray with CPU:s againt RS with a GPU.

[mbetke]

I think CPU renderer are not bad at all. Imagine how much CPU power you get for one Titan card.
Also CPU renderer are very optimized and GPU renderer are also a lot slower on complex scenes.
Sometimes I'm not that sure if one rig with two or four Titan cards is that much better/faster/cost efficient as a small cpu farm for the same price. Plus there are no that much GPU farms avaaible at the moment (subject to change with Otoy cloud some day).

If I need a quick final Maxwell render then I just submit it to Rebusfarm and get it back within minutes. So I may even only need my computer for preview renderings without need for GPU upgrades, power bills and so on.
For me I've chosen the GPU road but it may not be the best decision from an economical point of view.

[treddie]

My impression is that a complex scene should render way faster on a GPU than a CPU, if the program has been properly optimized to take full advantage of thread warps, locality and load-balancing.

[Refracty]

I like GPUs but CPUs are not that baaad.
GPUs are always good when the microprocessor units are doing a similar task all together.
If you invest 4000 € in GPUs or in CPUs (multiple computers) doesn't matters so much - You will have a fast setup in both cases.

[Dave_Yu]

I bought a license of Arion and the workflow is slower than Octane. Octane viewport is a joy to use and very fast compared to Arion. Arion have some cool features like motion blur and Displacement but the product does not seem solid.
I want to see if the upcoming version of Arion will be better.

Odd that they don't even have a user forum.

[treddie]

That's interesting about Arion.

Maxwell is the one that is puzzling to me...They are not embracing GPU rendering at all and seem to rely on render farm's for speed. If they think that will keep them competitive, I think their business model will sink them.