Revision history [back]

When image delivery is slow, it can affect the inference time with GPU in the following ways:

1. The GPU may have to wait for the image data to be transferred from the CPU to the GPU memory, which can increase the overall latency of the inference process.

2. The processing of the image data on the GPU may be slower due to the limited bandwidth of the communication channel between the CPU and the GPU.

3. If the image data is streamed in real-time, such as in video analytics, slow image delivery may cause the GPU to drop or skip frames, leading to inaccurate or incomplete results.

4. Slow image delivery can also increase the contention for GPU resources, such as memory and compute units, which can further slow down the inference process.