How can I run an OptiX 6 application in AWS Windows instance?

I’m trying to run my Optix app in AWS Windows.

The connection is provided via RDP and so I followed the tutorial in this link: https://www.cloudar.be/how-to-use-aws-ec2-gpu-instances/ to try and get access to the GPU, which doesn’t come in default mode in Windows Server.

The G2 instance allows an NVIDIA GRID connection, but it doesn’t contain the optix libraries in the driver, and that’s the only instance type that I managed to connect to the GPU with.

Does anyone have any experience with AWS? Please help…

Thank you

Hi @roman.vlk18,

Which version of OptiX are you trying to run?

On the OptiX team, we don’t have any experience with AWS, it would be better to check with the AWS forums for help on this issue, or request support from Amazon directly.

A few tips I can offer that may or may not help:

It appears AWS G2 instances run on the Kepler architecture, which is not supported on OptiX 6. It looks like G3 and P1-P3 instances might work, as they are Maxwell and later.

For OptiX 5.1 and earlier, you would need to install the OptiX SDK on your instance in addition to a driver that meets the minimum driver requirement for your OptiX version.

There’s nothing we can do if your instance type doesn’t have the driver version you need. You might investigate whether you can manually install your own driver, it looks like AWS has some help pages devoted to that topic.


David.

I was able to run the Optix 5.1.1 Precompiled samples on an AWS EC2 G3 (Tesla M60) instance using the AWS provided driver “412.16_grid_win10_server2016_64bit_international”

NVIDIA GRID Drivers for (G3):

Samples: C:\ProgramData\NVIDIA Corporation\OptiX SDK 5.1.1\SDK-precompiled-samples

I could not get the Optix 6.0.0 samples to run as they require driver 418.81 or later.

Same problem. I could not get the Optix 6.0.0 samples to run on AWS p3.x2large instance (Tesla V100).

±----------------------------------------------------------------------------+
| NVIDIA-SMI 419.69 Driver Version: 419.69 CUDA Version: 10.1 |
|-------------------------------±---------------------±---------------------+
| GPU Name TCC/WDDM | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Tesla V100-SXM2… TCC | 00000000:00:1E.0 Off | 0 |
| N/A 36C P0 24W / 300W | 0MiB / 16258MiB | 0% Default |
±------------------------------±---------------------±---------------------+

The GPU (V100) in the AWS P3 Instances runs in ‘TCC’ or Tesla Computer Cluster mode vs ‘WDDM’ mode, so it does not provide visualization support. I tried adding an AWS Elastic GPU which provides OpenGL support to a P3, but could not get the Optix 5.x or 6.x samples to run.

-->  | GPU Name TCC/WDDM | 
-->  | 0 Tesla V100-SXM2... TCC...

The GPU (M60) in the AWS G3 instance runs in WDDM mode and runs Optix 5.1.1 with the currently available NVIDIA driver (412) and should run Optix 6.0.0 once AWS releases a newer driver.

The Tesla M60 Nvidia driver for AWS EC2 G3 Instances has been updated to version 425.31_grid. Instructions for downloading and installing can be found here:

NVIDIA GRID Drivers for (G3):

I tested running the Optix 6.0.0 sample files via a RDP connection.

Samples: C:\ProgramData\NVIDIA Corporation\OptiX SDK 6.0.0\SDK-precompiled-samples