RT problems GPU

[hellobard]

Hey,

I keep getting this error when I set my V-Ray to GPU (CUDA and OpenCL):

// Warning: No scene found; cannot create CUDA renderer plugin. //

Not to mention this:
// Warning: Render host nrkdt61862 (127.0.0.1:20207) is not responding

The thing is, this just worked in another scene. So it's not consistent at all...

Running the nightly V-Ray for Maya version 2.25.01, revision 22851 from Apr 5 2013, 02:24:28 for Maya 2014.

Might just try the latest nightly.

[vlado]

// Warning: No scene found; cannot create CUDA renderer plugin. //

This one is normal; you can ignore it.

// Warning: Render host nrkdt61862 (127.0.0.1:20207) is not responding

This means that the RT render server is not running. What OS are you using?

Best regards,
Vlado

[hellobard]

This means that the RT render server is not running. What OS are you using?

Hmmm... Might be a network problem then? I am running Windows 7 64 Bit. I know I usually get a windows firewall window up on the first run of V-Ray with a fresh system, maybe there's something there?

[vlado]

Well, it should be running on the same machine, so it may not matter. Do you see a command-line window with the V-Ray RT render server? You can also just manually run a DR render server on your machine and see if the RT engine picks it up.

Best regards,
Vlado

[hellobard]

I dont see a command line running no.
It's weird because it will run and function perfectly in a clean scene after a fresh boot up, so it seems the server stops working after a while. Maybe it's because I have Maya 2013 and Maya 2014 both running VraySpawners in the Services?

[hellobard]

On another note, I just installed a Tesla K20m in my PC and tried the benchmark scene with V-Ray standlalone. The cmd line showed both my Tesla and my Quadro 5000 in Maya 2014, but in the Maya 2013 command line I just got the Quadro 5000. Are there any options in terms of which cards are activated somewhere? I also got an out of memory error when running the benchmark scene which seems very odd since the K20 has 5 GB. I was also not gettign very impressive speeeds, so I am wondering if there is something wrong with the K20 card... 2600 cuda cores should be blazing fast, no?

[svetlozar.draganov]

I dont see a command line running no.
It's weird because it will run and function perfectly in a clean scene after a fresh boot up, so it seems the server stops working after a while. Maybe it's because I have Maya 2013 and Maya 2014 both running VraySpawners in the Services?

Command line doesn't appear probably because RT service is running.
Running both 2013/2014 services could cause some troubles since they both are listening on the same port - could you please stop/remove them in order to troubleshoot the issue easily.
Once the services are stopped please run two separate tests on Maya 2013 and Maya 2014. Please use a simple scene and run the tests with only one instance of Maya activated at the same time.

On another note, I just installed a Tesla K20m in my PC and tried the benchmark scene with V-Ray standlalone. The cmd line showed both my Tesla and my Quadro 5000 in Maya 2014, but in the Maya 2013 command line I just got the Quadro 5000. Are there any options in terms of which cards are activated somewhere? I also got an out of memory error when running the benchmark scene which seems very odd since the K20 has 5 GB. I was also not gettign very impressive speeeds, so I am wondering if there is something wrong with the K20 card... 2600 cuda cores should be blazing fast, no?

Could you please confirm that you are using Vray Standalone first - because Standalone is independent from Maya versions. If you have different results this is due to different version of the Vray-Standalone - please make sure that you are using the same Vray build for both Maya 2013 and Maya 2014.
By default Vray will always use all installed GPUs - we have a tool to select which device should be used and which not - it is called "Select OpenCL devices" and it is located under ChaosGroup>Vray for Maya/Tools folder.

According to the memory issue - make sure that you are using only K20 cards for the GPU rendering , if you are using multiple GPUs for rendering then the memory limit is equal to amount of memory of the weakest card (the cards with least memory). So for example if K20 has 5GB and Quadro has 2.5Gb - then the limit is 2.5GB not 5GB.

According to the speed - please try to run two separate tests with Quda engine between Quadro and Testla cards and inform us about the results.

[t.petrov]

hellobad:
Keep in mind that a windows service process cannot access the GPU, so you have to disable your vrayspawner and allow maya to autostart the render slave or start it manually.

There is one exception: If you have a Tesla card and the proper driver (TCC driver) it can work.

[hellobard]

Command line doesn't appear probably because RT service is running.
Running both 2013/2014 services could cause some troubles since they both are listening on the same port - could you please stop/remove them in order to troubleshoot the issue easily.
Once the services are stopped please run two separate tests on Maya 2013 and Maya 2014. Please use a simple scene and run the tests with only one instance of Maya activated at the same time. .

Alright, I will try that. I have turned both Spawners off now and I will install the same version of Vray on both.

Could you please confirm that you are using Vray Standalone first - because Standalone is independent from Maya versions. If you have different results this is due to different version of the Vray-Standalone - please make sure that you are using the same Vray build for both Maya 2013 and Maya 2014.
By default Vray will always use all installed GPUs - we have a tool to select which device should be used and which not - it is called "Select OpenCL devices" and it is located under ChaosGroup>Vray for Maya/Tools folder.

According to the memory issue - make sure that you are using only K20 cards for the GPU rendering , if you are using multiple GPUs for rendering then the memory limit is equal to amount of memory of the weakest card (the cards with least memory). So for example if K20 has 5GB and Quadro has 2.5Gb - then the limit is 2.5GB not 5GB.

According to the speed - please try to run two separate tests with Quda engine between Quadro and Testla cards and inform us about the results.

I was trying Vray Standalone when I wrote this, yes. But also tested it with RT inside of Maya as I mentioned.

I understand regarding the memory issue... But it's a bit strange that your test scene would run out of memory in regards to the list of cards that have successfully run the test which have less memory. Shouldn't this scene, in theory, not run out of memory then?
I guess it might just have been a general error of some sort, I will try it again with both cards and then with each individual card.

As for the Select OpenCL devices in the Chaosgroup menu, I tried both the Maya 2013 and 2014 folders and both gave me a "Problem with Shortcut" error and I can't seem to find this file in the vray/bin/ folder where it is supposed to be. Is this tool really in the nightlies?
Also, do the settings in this tool work with Vray Standalone? I was using the test-scene in Vray Standalone since it uses a .vrscene file, so that's why I am asking.

[vlado]

As for the Select OpenCL devices in the Chaosgroup menu, I tried both the Maya 2013 and 2014 folders and both gave me a "Problem with Shortcut" error and I can't seem to find this file in the vray/bin/ folder where it is supposed to be. Is this tool really in the nightlies?

I think it's missing - at least, there is an issue in our bug tracker about this, so we have to fix this.

Also, do the settings in this tool work with Vray Standalone? I was using the test-scene in Vray Standalone since it uses a .vrscene file, so that's why I am asking.

Yes, they do work in the standalone (the tool just sets the VRAY_OPENCL_PLATFORMS_x64 environment variable).

Best regards,
Vlado

[hellobard]

Alright, thanks Vlado. I will have a look for it a little later when it's in the nightlies.

I just got off the phone with Nvidia, and the reason I am not seeing any speed improvements and the memory error I got when doing the test scene is quite simple:
The K20 does not work with a Fermi Quadro 5000... It needs a Quadro K5000 to work in unison. So now we have to cough up some extra dough for that :P
I imagine I can enable only the K20 for V-Ray and turn off the Quadro 5000, but for Phoenix/Nuke/Premiere/etc. I probably can't do that... Very annoying, and not much information about this on Nvidia's site imho.

[vlado]

The K20 does not work with a Fermi Quadro 5000... It needs a Quadro K5000 to work in unison. So now we have to cough up some extra dough for that :P

Ah, yes, I found that out the hard way too. I could never get a Fermi and a Kepler card to work together.

Best regards,
Vlado

[hellobard]

Ah, yes, I found that out the hard way too. I could never get a Fermi and a Kepler card to work together.

Best regards,
Vlado

Hehe... surprising that Nvidia keeps this "secret" so hidden
We just ordered two K5000's, so I guess they just expect every commercial company to do the same when they realize the mistake.