Arion by Random Control

[Cesar R]

^ Makes perfect sense, This is what I am trying to find out. What is confusing is that people keep saying "use a lesser card for the UI" but in reality, that card needs to be equaly as powerful to drive the modeling application as should be on the X16.

 

Is there a way to exclude the main card from the rendering? - what happens if you don't and you have a card @ X16 and nother @X8. Does the scene load at X16 onto one card and then that card waits for the other to load it portion at X8 ?

[Fran]

^ Makes perfect sense, This is what I am trying to find out. What is confusing is that people keep saying "use a lesser card for the UI" but in reality, that card needs to be equaly as powerful to drive the modeling application as should be on the X16.

 

Is there a way to exclude the main card from the rendering? - what happens if you don't and you have a card @ X16 and nother @X8. Does the scene load at X16 onto one card and then that card waits for the other to load it portion at X8 ?

 

You can configure Arion to use only the resources you want it to use. You would exclude the card that you are using for the Windows UI, the card that you have your monitor(s) hooked into.

 

If you already have a nice quadro that you are using for max and rhino, then continue to use it for driving your UI. But you should know that if you include it in the Arion hardware configuration, available ram will be limited to the card in the pool with the least amount of memory, since all Cuda processes are done in parallel across all available resources. Just as render nodes across a network need to be able to fit a scene into memory, so does a gpu (think of each gpu in your hardware configuration as a separate render node).

 

Many motherboards are set up so that if only pci slots 1 & 2 are occupied, they will run at x16. If slot 3 or 4 (varies) is occupied, then slot 2 will only run at x8. So you could have your quadro in slot 1 running your monitors, and additional cards in the other slots.

 

I have only recently purchased any fermi (gtx 400 series) cards and haven't fully tested them, but I am finding that mixing Fermi and non-Fermi gpus in the hardware configuration can lead to diminished performance.

[Cesar R]

But you should know that if you include it in the Arion hardware configuration, available ram will be limited to the card in the pool with the least amount of memory, since all Cuda processes are done in parallel across all available resources.

 

Can you explain or rather break down the above mentioned concept about the ram being limited to the card with the least amount of memory?

What do you mean by this?

[RayOfLight]

basically when you have two videocars, say a quadro FX1800 with 768Mb and a GTX470 with 1.2Gb on it, you will have only 768Mb that you can use to render scenes. You could tell Arion not to use the FX1800, which would mean you would use only the GTX470 and thus could load/render scenes using a max of 1.2Gb.

 

(@fran, I hope I didn't offend you by answering this question)

[Fran]

You didn't offend me Stefan. I thought I'd made it clear already though.

[Cesar R]

I am slow... what can I say.

[Fran]

It isn't that Cesar I'm just not always a good explainer. Sometimes I give too much info and end up confusing people.

[Cesar R]

Fran, I think what happens is that I am not familiar or I should say very familiar with some of the concepts such as rendering in parrallell and such. I look forward to getting a licesense of the program soon. It's a shame there are no demos.

[RayOfLight]

I'd say, buy it, you won't regret it , but it's your call of course

[Cesar R]

^ I think that most subforums should be open to anyone interested in the program to allow for pre-sale research. Somehow it seem like the forums are hidding something.... random control forums

[Fran]

The support forums are for peer-based technical support for licensed users. Although most often, questions are answered by beta testers or the development team.

 

There is nothing to hide. I've tried to be helpful in every way I know how and tried to answer all of your questions, both here and via private PMs on the Randomcontrol forum.

[Fran]

 

This diagram might make my explanation a little clearer regarding multi-GPU versus LAN rendering. If you send a render over a LAN, each computer that receives the scene must have sufficient resources - mainly system RAM - to process and render the scene. If you looked at the multiple GPUs on a motherboard in the same way as multiple computers in a network, the GPUs are subject to the same constraints. Each GPU that will be rendering must have enough available memory to load and render the scene.

 

A major difference in the case of Arion GPU/CPU rendering and an engine like fryrender, is that the same scene will take far less memory in Arion than it does in fryrender. When your system has multiple GPUs of different memory capacities, you may have a scene that will fit in the higher memory cards, but not a lower one. You can exclude the lesser card(s) from the render job, just as you would exclude a computer with too little system ram from a traditional network render job.

 

For example, if GPUs 1 and 2 have 2 GB of memory and GPU 3 only has 1 GB, you could disable GPU 3 temporarily in the Arion Configuration panel if the scene you wish to render requires more than 1 GB of memory to load. I have a Tesla C1060 with 4 GB of memory and I have never seen the memory consumption go above 2 GB for my interior archviz production work. Arion has been designed to be very memory-efficient.

 

Additionally, the frame buffer is handled mainly by system RAM. Most of my machines have 8GB of RAM, but one only has 6. I was able to fit a 5100x3300 pixel render on the 6GB system.

[Cesar R]

Fran,

 

You are and have been great help in both aspects and forums ! And maybe my statement "what are they hidding" is a little crazy to say the least, but honeslty I felt like that as some point. I mean, it much more information or "nice" to be able to freely lurk and read and interact w the company and users of a product you are interested in.

 

I appreciate the graphic, I am sure many people will benefit for you effort.

I am wowed by your investment in the Tesla in a good way. That "card" must make you smile everyday. I know I would be. I would love to buy something like that someday when I grow up

 

Let me ask you this, what happens if you have (in my case - I oredered a GTX 470 (Fermi) ) with 1.2GB and you load a scene that would potentially not fit? do you get an error I presume and then you cannot render ? What from what I gather, if I had 2 GTX470 @1.2GB or ram, I have 1.2GB or ram not 3.6GB correct?

 

I presume that with the Tesla you must not need anything else. - What other things do you use that card for? or was this a purchase mainly for RTR?

 

I am starting to feel embarrased at the basic questions I am asking as it was pointed out previously... sorry =(

[Fran]

Hi Cesar,

 

If your scene doesn't fit into the GPU RAM, you can render with CPUs only. RAM is not additive across GPUs, in the same way that it is not additive across computers on a LAN.

 

There are some manufacturers that have put an additional GB in their gtx 460 cards. I am testing a machine with a pair of Palit GTX 460 GPUs with 2GB ram each. So far, they are surprisingly quiet and stay well below max temperatures at full load. I think Gainward and Zotac have one coming out too. They are priced in between the 460 with 1GB and the 470 with 1.2 GB. I paid $260 USD each for mine.

 

One thing that I have noticed is that there is no slow-down of the UI with monitors connected to one of the gtx 460s.

[Wildstyle]

hi,

sorry if it's not the good place to ask,

I'm trying to set up arion render for 3ds max, so I can apply materials in 3ds max and render it using arion render engine. I donwloaded the 3ds max plugin from the randomcontrol website. But, It seems I only have fryrender settings and I'm not abble to use arion.

 

Anyone can help?

 

Thx

[Fran]

Arion and fryrender share settings. It will be more apparent in the coming update. Meanwhile, I think this issue is mentioned in some of the "getting started" papers on the Arion wiki.