From 90d4a30587f6a56d39f99fde3ac86ce88c3c47ab Mon Sep 17 00:00:00 2001 From: AlfredMoore Date: Sun, 14 Jun 2026 00:45:47 -0500 Subject: [PATCH] RTX 5090 benchmark on coco val2017; readme guidance update --- README.md | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 4f2ef02..c6b28a4 100755 --- a/README.md +++ b/README.md @@ -38,6 +38,7 @@ The numbers show end to end image processing latency per image (4K resolution) i | Jetson Thor | | | | Please contribute | | DGX Spark | | | | Please contribute | | RTX 3090 | 438 ms | 75 ms | 5.82x | | +| RTX 5090 | 120.9 ms | 24.8 ms | 4.88x | COCO val2017, TRT 10.14.1 | | A10 | 545.3 ms | 161.1 | 3.38x | GPU hits 100% utilization | | A100 | 314.1 ms | 48.8 ms | 6.43x | 40GB SXM4 variant | | H100 | 265.3 ms | 34.6 ms | 7.66x | PCIe variant | @@ -112,11 +113,11 @@ docker run -it --rm \ ```bash python python/onnxexport.py ``` -This produces `onnx_weights/sam3_static.onnx` plus external weight shards. +This produces `onnx_weights/sam3_dynamic.onnx` plus external weight shards. 5) Build a TensorRT engine ```bash -trtexec --onnx=onnx_weights/sam3_static.onnx --saveEngine=sam3_fp16.plan --fp16 --verbose +trtexec --onnx=onnx_weights/sam3_dynamic.onnx --saveEngine=sam3_fp16.plan --fp16 --verbose ``` 6) Build the C++/CUDA library and sample app