Problem with Drivers - Petrel 2014 with NVIDIA Grid K2 - 4 GPU

Hi All,

I am deploying a VDI to be used for Petrel 2014. As long as I am using the K260q vGPU, everything seems fine. Now I want to give a full GPU per VM, i.e. Grid K2 per VM. The issue is, anytime I change the GPU the system ends up using the Standard Graphic Card and I have a yellow triangle on the GRID K2 in device manager (after installing the Drivers). All forums on the internet suggest to go to the NVIDIA control panel but since I am using Xenserver 6.2 I cannot go to the control panel as once the passthrough is enable the only way to connect to the machine is via Remote Desktop. Connecting with Citrix Receiver too is not working.

Also For those who have also deployed this infrastructure, I wish to do a pass through because Petrel is too slow in loading the files. Please what do you think I should allocate to the VM? My current specs are:

  • Virtual Machines (For Worksations)

OS: Windows 7 Pro x64 SP1
CPU: Intel® Xeon® CPU E5-2680 v2 @ 2.80GHz (4 CPUs), ~2.8GHz
RAM: 48 GB
GPU: NVIDIA GRID K260q
NVIDIA Drivers: 332.83_grid_win8_win7_64bit_english.exe

  • Citrix Xenserver Hypervisor

OS: Xenserver 6.2
CPU: 40 Cores (Intel® Xeon® CPU E5-2680 v2 @ 2.80GHz)
RAM: 393181 MB
GPU: GK104GL [GRID K2]
NVIDIA Drivers: NVIDIA-vgx-xenserver-6.2-331.59.01.i386.rpm , NVIDIA-vgx-xenserver-6.2-331.59.i386.rpm

I am opened to all suggestions.
Thanks,

Elom

Couple of suggestions.

First, use the passthrough driver, instead of the vGPU driver.

http://www.nvidia.co.uk/download/driverResults.aspx/77665/en-uk

Then, confirm that the GPU is available for the VM to connect to.

Finally, in the switch from vGPU to Passthrough, it’s possible that the connection with the VDA is broken which is why the Cirix connection fails. To resolve this you can first try running

montereyenable.exe -enable -noreset

Then reboot. You can run this via RDP.

This should reconnect the link so that the VDA “see’s” the GPU.

If it’s still problematic I’d recommend the following

Shutdown the VM
Remove the GPU profile from the VM
Boot then uninstall the vGPU driver from the VM
Reboot (you should now have a standard VM with no GPU)
Shutdown the VM
Attach the passthrough GPU
Install the passthrough driver
Run montereyenable.exe -enable -noreset
Reboot

That should get you back on track.

Hi Jason,

Thank you very much for your reply. I am currently downloading the driver from your link. I will try all that in a few hours and revert back.

Hello Jason,

Thanks again for your suggestions. I was able to get it done. Now the VMs have 2 Quadcore CPUs, 64GB of RAM and 1 GRID K2 Pass-through GPU (4GB of Video Memory) properly installed. The Citrix Receiver is able to log in as well.

Unfortunately I still record poor performance from Petrel 2014 i.e. it seems slow when working on data. I am wondering if there is any registry tweak or any command or optimization I should do (at the software level or at the hardware level).

Awaiting for your wise advice.

Thanks,
Elom

Hi Elom,

Glad to see you finally got it resolved, it can be a little frustrating when everything looks right and it still doesn’t work!

As for performance, have you got 4 or more vCPU’s allocated to the VM, and are they visible inside the VM?

Let me know how that stands and we’ll go from there.

Hi Jason,

Thanks for answering. You are right; it is reaaly frustrating, as I thought everything was solved when I finally got the pass-through working.

I used 8 cores from the server, presented as 2 quadcore processors. When I check from the task manager, I can see the 8 cores being utilized by the OS (win7). When I use the nvidia-smi command, I can as well see the utilization of the pass-through GPU, about 37%. The Ram is now 64GB and not even 25% is being used. The CPUs also look fine in terms of utilization.

Still the system is performing low.

Note the Physical server is an HP DL380z with 40 Cores (Intel® Xeon® CPU E5-2680 v2 @ 2.80GHz).

Awaiting for your advice,

Elom

Thanks Elom,

Everything looks like a good start point,

How are you measuring performance, and what are you comparing against?

What’s the storage subsystem? Sometimes IOPS becomes a bottleneck which drags everything down. I’ve seen a couple of applications that really benefit from either SSD or a technology such as Atlantis ILIO.

Hi Elom,

Can you take a look at the IOPS being consumed in the VM. It’s possible that the storage subsystem is causing performance issues with accessing data. Always worth investigating and a known bottleneck in desktop virtualisation.

Hi Guys,

I was wondering if you can also help on that one.
I am about to buy a laptop, with a quite good pro 4-core i7. The thing is this laptop has a Geforce 750M and I wonder if this graphic card would be powerfull enough to run Petrel 2012.

Am I forced to go to a laptop with a good high end NVIDIA Quadro to be able to use that software on my machine.

This may be a stale thread, but for anyone configuring the K2 for Petrel, please check your data throughput as indicated above. Petrel utilizes lots of small files generating an intensive level of disk I/O which is capable of saturating both disk I/O channels and standard enterprise networks. To effectively configure the system to support it, these factors must be considered in conjunction with available CPU, memory, and GPU.