RTX 3070 / TensorRT Internal Error: Assertion failed: Unsupported SM

Description

There are several issues when processing ONNX files and compiling TRT models, when launching the program on the GPU RTX 3070 with driver 460.32.03

(On other CUDA GPUs, younger RTX3070, with drivers 410/440/450 we work successfully)

Working on: model-batch4-fp16-112x112.onnx
Number of network layers: 223
Saving @ tensorrt/model-batch4-fp16-112x112.trt
STDERR: [TensorRT] WARNING: Calling isShapeTensor before the entire network is constructed may result in an inaccurate result.
[TensorRT] WARNING: Half2 support requested on hardware without native FP16 support, performance will be negatively affected.
[TensorRT] INTERNAL ERROR: Assertion failed: Unsupported SM.
../rtSafe/cuda/caskUtils.cpp:80
Aborting...
[TensorRT] ERROR: ../rtSafe/cuda/caskUtils.cpp (80) - Assertion Error in trtSmToCask: 0 (Unsupported SM.)
...
Working on: model-fp16-960x544.onnx
Number of network layers: 248
STDERR: [TensorRT] WARNING: onnx2trt_utils.cpp:198: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.
[TensorRT] WARNING: onnx2trt_utils.cpp:198: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.
[TensorRT] WARNING: onnx2trt_utils.cpp:198: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.
[TensorRT] WARNING: Half2 support requested on hardware without native FP16 support, performance will be negatively affected.
[TensorRT] INTERNAL ERROR: Assertion failed: Unsupported SM.
../rtSafe/cuda/caskUtils.cpp:80
Aborting...
[TensorRT] ERROR: ../rtSafe/cuda/caskUtils.cpp (80) - Assertion Error in trtSmToCask: 0 (Unsupported SM.)
Saving @ tensorrt/model-fp16-960x544.trt

nvidia-smi

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.32.03    Driver Version: 460.32.03    CUDA Version: 11.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  GeForce RTX 3070    Off  | 00000000:01:00.0  On |                  N/A |
|  0%   57C    P2    57W / 240W |    617MiB /  7979MiB |      1%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

Is it possible this GPU RTX3070 does not support FP16?

Environment

TensorRT Version: 7.0.0.11
GPU Type: RTX3070
Nvidia Driver Version: 460.32.03
CUDA Version: cuda10.0
CUDNN Version: 7.6.5.32
Operating System + Version: Ubuntu 18.04.5
Python Version (if applicable): 3.6.9
Numpy: 1.19.0

It seems like RTX 3070 is not compatible with TensorRT version 7.0.0.11

Hi @dmitro.biz,

You need at least TRT 7.2.1 and CUDA 11.1 to support the 3070.

Thank you.

Hello!
Thank you very much.

Status of the new tests GTX 3070 with Cuda10.2 / Cudnn8 / Trt7.2.2. 3/ Driver460.39: failure.

STDERR: [TensorRT] INTERNAL ERROR: Assertion failed: (smVersion < SM_VERSION_A100) && "SM version not supported in this NVRTC version"
./helpers.h:523
Aborting...
[TensorRT] ERROR: ./helpers.h (523) - Assertion Error in fixSMVerForNVRTC: 0 ((smVersion < SM_VERSION_A100) && "SM version not supported in this NVRTC version")

Hi @dmitro.biz,

You need CUDA 11.1 as mentioned in previous reply.

Thank you.

1 Like