Home › Benchmarks › NVIDIA T4 (server)
Benchmark Explorer · NVIDIA T4 (server)
NVIDIA T4 (server): 17 benchmark rows, 100% measured.
Method v1.0 · dataset 2026-09-07 · verified 2026-09-07 · last updated September 2026
17 of 17 rows are measured and published (class C); the rest are EdgeAIStack-derived (class D). 5 model families. Precisions: FP16 (17). Runtimes: tensorrt (17). Tasks: detection (13), pose (2), segmentation (2). This is a reference platform in the benchmark database only; no other EdgeAIStack engine sizes it.
Family × precision: best value
The fastest matching row per family and precision (fps, or tokens/s for language models), with its evidence class. Click a cell for the full breakdown.
| Family | FP16 |
|---|---|
| RT-DETR | 114 fps C |
| YOLO11 | 666.7 fps C |
| YOLO11 | 588.2 fps C |
| YOLO11 | 555.6 fps C |
| YOLOv10 | 543.5 fps C |
Every model on NVIDIA T4 (server)
RT-DETRl
RT-DETR · rtdetr-l
RT-DETRx
RT-DETR · rtdetr-x
YOLO11l
YOLO11 · l
YOLO11m
YOLO11 · m
YOLO11n
YOLO11 · n
YOLO11s
YOLO11 · s
YOLO11x
YOLO11 · x
YOLO11m-pose
YOLO11 · m-pose
YOLO11n-pose
YOLO11 · n-pose
YOLO11m-seg
YOLO11 · m-seg
YOLO11n-seg
YOLO11 · n-seg
YOLOv10b
YOLOv10 · b
YOLOv10l
YOLOv10 · l
YOLOv10m
YOLOv10 · m
YOLOv10n
YOLOv10 · n
YOLOv10s
YOLOv10 · s
YOLOv10x
YOLOv10 · x
Links
- Live engine: https://edgeaistack.ai/engines/benchmark-explorer/?hw=nvidia_t4&model=rtdetr&variant=rtdetr-l&mv=1.0&dv=2026-09-07
- Methodology: /methodology/benchmark-explorer/
- Dataset: /datasets/benchmarks-v2.json
- All hardware: /benchmarks/
Query this hardware live.
Filter by model, precision or runtime and get the matching rows with a permanent link.