Skip to content

FP16 model compilation failure with TensorRT 11.2.1.2 on 4090 RTX #4838

Description

@h6197627

Description

Some models with FP16 precision fail to compile from ONNX format with error like this:

No low-precision conv kernel available for this strongly-typed Conv/ConvTranspose. Consider keeping Conv/ConvTranspose and Q/DQ ops in FP32 (e.g. via modelopt.onnx.autocast.convert_to_f16 with op_block_list=['Conv','ConvTranspose','QuantizeLinear','DequantizeLinear']). Failing node: node_223_Conv
[08/25/2026-19:48:50] [E] Error[10]: IBuilder::buildSerializedNetworkToStream: Error Code 10: Internal Error (Could not find any implementation for node node_223_Conv. In computeCosts at /_src/optimizer/common/tactic/optimizer.cpp:4211)

TensorRT 10.16.1.11 compiles these models successfully. Also, FP32 works normally in TensorRT 11.2.1.2.

I reproduced the error with the simplest ONNX sample consisted of Input->Cast(FP16)->Conv(Grouped)->Cast(FP32)->Output (attachead
model.onnx.zip)
Probably problem related to group Conv optimizations.

Full failure log attached 11.2.1.2.log
Also 10.16.1.11 successful conversion log attached 10.16.1.11.log
Also attached engine compiled with 10.16.1.11 engine_10.16.1.11.trt.zip

Environment

TensorRT Version: 11.2.1.2

NVIDIA GPU: 4090 RTX

NVIDIA Driver Version: 610.57.04

CUDA Version: 13.3

Operating System: Ubuntu 22.04

Steps To Reproduce

Commands or scripts: trtexec --onnx=model.onnx --saveEngine=engine.trt

Have you tried the latest release?: Yes

Can this model run on other frameworks? For example run ONNX model with ONNXRuntime (polygraphy run <model.onnx> --onnxrt):Yes

Metadata

Metadata

Assignees

No one assigned

    Labels

    Module:ONNXIssues relating to ONNX usage and import

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions