triton-inference-server / server

The Triton Inference Server provides an optimized cloud and edge inferencing solution.
https://docs.nvidia.com/deeplearning/triton-inference-server/user-guide/docs/index.html
BSD 3-Clause "New" or "Revised" License
7.81k stars 1.42k forks source link

Triton Server 24.05 can't initialize CUDA drivers if host system has installed Nvidia driver 555.85 #7319

Open romanvelichkin opened 1 month ago

romanvelichkin commented 1 month ago

Description I was using Triton Server nvcr.io/nvidia/tritonserver:24.04-py3 on my local machine with Windows 10 via docker container. Ie installed latest Nvidia Driver 555.85, and docker container stopped to run, returning error:ERROR: The NVIDIA Driver is present, but CUDA failed to initialize. GPU functionality will not be available. [[ Named symbol not found (error 500) ]] I downloaded container with version 24.05, but it didn't help - server doesn't start, returns same error instead.

Triton Information tritonserver:24.04-py3, tritonserver:24.05-py3,

Are you using the Triton container or did you build it yourself? Container

To Reproduce Windows 10, install Nvidia driver 555.85, run Triton Server tritonserver:24.05-py3 with model.

Describe the models (framework, inputs, outputs), ideally include the model configuration file (if using an ensemble include the model configuration file for that as well). Model: efficientnetv2 converted to onnx from tflite.

backend: "onnxruntime"
max_batch_size: 32
input [
  {
    name: "serving_default_input_1:0",
    data_type: TYPE_FP32,
    dims: [ 512, 512, 3 ]
  }
]
output [
  {
    name: "StatefulPartitionedCall:0",
    data_type: TYPE_FP32,
    dims: [ 2 ]
  }
]
dynamic_batching { }
response_cache { enable: true }

Expected behavior Triton Server has to run my model.

Tabrizian commented 1 month ago

The NGC containers are only supported on Linux. For Windows, please refer to the zip file that is delivered as a part of Github release: https://github.com/triton-inference-server/server/releases/tag/v2.44.0 (24.03 is the latest release of Windows support).

romanvelichkin commented 1 month ago

The NGC containers are only supported on Linux. For Windows, please refer to the zip file that is delivered as a part of Github release: https://github.com/triton-inference-server/server/releases/tag/v2.44.0 (24.03 is the latest release of Windows support).

Triton Server properly worked with previous version of driver. I downgraded driver already and Triton started to work again. I'm using 24.05 NGC Triton container.

Do you want to say that 24.03 will work with driver 555.85, and later Triton Server versions won't be working with latest Windows Nvidia drivers anymore?