Support Knowledge Base

Why Do the Inference Results Differ When Compiled With the GPU Plugin?

Content Type: Troubleshooting | Article ID: 000100594 | Last Reviewed: 05/23/2025

Description Resolution

Description

Converted clip-vit-large-patch14 model into Intermediate Representation (IR).
Ran inference using the CPU plugin and obtained the expected results. However, when running inference with the GPU plugin, observed different output results.

Use INFERENCE_PRECISION_HINT: "f32" to force the FP32 inference on the GPU plugin.

This article applies to 3 products.

Intel® Xeon Phi™ Processor Software OpenVINO™ toolkit Performance Libraries

Contact support