HARDWARE-AWARE PERFORMANCE OPTIMIZATION OF YOLOV8 FOR REAL-TIME APPLICATIONS: GPU ACCELERATION WITH TENSORRT AND AN ANALYTICAL FPGA COMPARISON

Tursunaliev, Ulugbek, Tulkinov, Bakhromjon

Innovation science and technologiy · 2026-yil

Annotatsiya

Real-time computer-vision systems deployed on edge platforms are constrained as much by the surrounding softwarepipeline and the host–accelerator interface as by raw compute. This paper presents a reproducible, hardware-aware optimization studyof the lightweight YOLOv8n object detector on an NVIDIA Tesla T4 GPU and contrasts the measured GPU behavior against ananalytical projection of a Xilinx Zynq UltraScale+ (ZU9EG) FPGA accelerator. Applying TensorRT graph optimization—verticalConv–BatchNorm–activation layer fusion together with FP16 precision—raised single-stream inference throughput on the T4 from aPyTorch FP32 baseline of 62.92 FPS to 72.72 FPS, a 1.16× (15.6%) speed-up, while mean average precision on COCO val2017 fellonly marginally (mAP@0.5 from 52.8% to 52.6%). A controlled batch-size sweep then exposes a host-side preprocessing bottleneck:as the batch grows to 16, host CPU utilization saturates to 98%, while GPU utilization decreases to 8%, and end-to-end throughputdegrades to 2.05 effective FPS. We show analytically and empirically that this is a data-ingest (decode/resize) limitation of the hostpipeline rather than a compute limit of the accelerator. Finally, using device datasheets and published FPGA-accelerator figures, weproject that a spatial INT8 dataflow accelerator would deliver lower peak throughput (≈45 FPS) but markedly better energy efficiencyand deterministic per-frame latency, characteristics desirable for latency-critical edge deployment. All numerical claims are clearlyseparated into measured (GPU) and projected (FPGA) categories.

Maqola ma’lumotlari
MualliflarTursunaliev, Ulugbek, Tulkinov, Bakhromjon
JurnalInnovation science and technologiy
Nashr sanasi2026-06-01
Jild2
Son6
TilIngliz
DOI10.5281/zenodo.20781147

Kalit so‘zlar

object detection; YOLOv8; TensorRT; layer fusion; FP16/INT8 quantization; GPU; FPGA; edge inference; energy efficiency; deterministic latency.

Ilmiy soha

Innovation science and technologiy jurnalidan boshqa maqolalar

Innovation science and technologiy — barcha maqolalar