Combined Quadro K4200 & Geforce Titan X experiences

[bvdsdesign]

Hi,
I've been thinking of boosting up my workstations. And I want to add a Titan X graphics card together with our standard Quadro K4200.
Does anyone here have any experience with this combination?
Does the PC recognize the 2 cards? How about drivers?
Is there a special setup for vray?

Thanks

[3LP]

Our workstations are all equipped with k4200 and we did get some Titan Black on those without any issues. Didn't recall how we managed the drivers though.

For the Titan X though, we decided to build some special nodes with 4 Titan X in so the team can easily DR through it.
More effective for the company than having each one having cards locally.

No special setup for Vray, once your OS see it, Vray will as well automatically.
If you do want to tweak though, just use the "ocldeviceselect.exe" who is the "Select OpenCL devices for V-Ray RT GPU" utility.
It's name comes from when Vray supported only opencl, but since it does cuda as well.

Hope this help
Cheers
Stan

[bvdsdesign]

Thanks,
3LP for the reply. I've got one vray machine where I would install the titan on. But I still have other workstation that i use for DR. Maybe I should get some GTX980 Ti or a Custom PCB..
Like Gigabytes GV-N98TG1 GAMING-6GD..
Whats you're idea?

[3LP]

Yes it's a great card.

It's actually the best if you have to stay with 2 pci slots.
Otherwise you have the "ZOTAC GEFORCE GTX 980 TI AMP! EXTREME EDITION 6GB" who is a bit more overclocked but uses 3 slots.
http://www.ldlc-pro.be/fiche/PB00189720.html

Keep in mind that those cards have airflow going downward and not frontwards.
meaning, if you stick multiple in a computer, they will heat one each other as they will blow each one on them.
The titan has a front face blower so the air is blown from where you plug your cables, but not downward (onto the next card). This is quite important if you want to have a multi card setup.
I have myself a 3x GTX 970 and they blow air on them-self/next-one and it's really not ideal.

Just my 2 cents on multi cards setup. So if you'd like to go down that road, maybe there are some GTX 980Ti that have air blower with front face?

Hope this helps.
Stan

[3LP]

I just had a look and I did found few GTX980Ti with the original design :
http://www.ldlc-pro.be/fiche/PB00188781.html
http://www.ldlc-pro.be/fiche/PB00188784.html
http://www.ldlc-pro.be/fiche/PB00188783.html
http://www.ldlc-pro.be/fiche/PB00188782.html
http://www.ldlc-pro.be/fiche/PB00188888.html

They are all with the base clock settings but you'll be able to OC them if you really want to squeeze everything out of the card.
I personally pushed the Titan X +25% in OC without even push the temperature when the coolers are at stock, and the Titan X has the reference design.

Stan

[bvdsdesign]

Thanks Stan for the update. Good thing you mentioned air-flow. Didn't think about that.. I think i will stay with the original design for this setup.

[3LP]

Good call

Stan

[Booska]

I just installed two EVGA 980Tis in my system in SLI config. They recommend that you keep an empty slot between them for cooling purposes, if you're going with dual SLI.

[savage309]

I just installed two EVGA 980Tis in my system in SLI config. They recommend that you keep an empty slot between them for cooling purposes, if you're going with dual SLI.

Just a note - if you are going to use them for V-Ray RT GPU, you should not use SLI.
V-Ray will use both of the GPUs even when SLI is off. If it is on it will cause troubles for RT GPU.

[Booska]

Just a note - if you are going to use them for V-Ray RT GPU, you should not use SLI.
V-Ray will use both of the GPUs even when SLI is off. If it is on it will cause troubles for RT GPU.

Wow! Good to know. I will remove that SLI bridge later today, but still keep the cards apart to keep them cool.

[savage309]

Wow! Good to know. I will remove that SLI bridge later today, but still keep the cards apart to keep them cool.

There is a tool named ocldeviceselect that comes with every V-Ray installation and lets you choose which GPUs to be used for rendering (in case you want some of them for the viewport). If during the rendering with RT GPU the UI of the OS becomes sluggish (this can happen only if the GPU for the monitor is used for rendering too), just decrease Ray Bundle Size and Rays Per Pixel until it is okay. More tips here http://docs.chaosgroup.com/display/VRAY3/GPU+Rendering.

[auke]

Our workstations are all equipped with k4200 and we did get some Titan Black on those without any issues. Didn't recall how we managed the drivers though.

For the Titan X though, we decided to build some special nodes with 4 Titan X in so the team can easily DR through it.
More effective for the company than having each one having cards locally.

No special setup for Vray, once your OS see it, Vray will as well automatically.
If you do want to tweak though, just use the "ocldeviceselect.exe" who is the "Select OpenCL devices for V-Ray RT GPU" utility.
It's name comes from when Vray supported only opencl, but since it does cuda as well.

Hope this help
Cheers
Stan

Hi, I'm currently working on setting up a very similar rendering infrastructure, but I do have some issues you might be able to help me with. Like you I'm planning to set up a render node with 4 Titan X's which will be used as a server only. Though I've learned windows will block the use of at least 1 GPU from use by another machine (distributed rendering) as it feels it needs to save it for the user in front of the machine (the server that is, which obviously has no direct user.). How did you manage to get arround this issue or are you actually only rendering on 3 of the 4 Titan X's per system?

[savage309]

Though I've learned windows will block the use of at least 1 GPU from use by another machine (distributed rendering) as it feels it needs to save it for the user in front of the machine (the server that is, which obviously has no direct user.). How did you manage to get arround this issue or are you actually only rendering on 3 of the 4 Titan X's per system?

Where did you read that ? AFAIK there should not be any problems with V-Ray RT GPU to use all of the GPUs (and it should not matter if it is being run as a render server or not).

[auke]

Where did you read that ? AFAIK there should not be any problems with V-Ray RT GPU to use all of the GPUs (and it should not matter if it is being run as a render server or not).

I have not read that on a Vray specific topic but on various other sites. (example: http://www.royalrender.de/help7/inde...rendering.html). After reading it I tested it at the office and tried to run 1 pc as a node (equipped with a K4000 as display adapter and a Titan X as compute card) and used another pc to manipulate the scene. Using distributed rendering and the Vray RT server on the node set to use both the K4000 as the Titan X. Rendering went fine but after checking the node I confirmed it was not using the K4000 (again, the display adapter).

[savage309]

Should there be a monitor attached in order that behaviour to show ? We have multiple machines in the office that render just fine with all GPUs they have.

Are you sure that in the ocldeviceselect.exe tool both of the GPUs are ticked ? Do you start V-Ray as a service and what driver you have installed ? Can you show the log of the vray.exe process on that machine ?