Merwansky·Nov 4TensorRT or OpenVINO? One Vision Model, NVIDIA and Intel Hardware, MeasuredComputer Vision & AI
Israr Ahmed·Aug 17I Made YOLO-World 2.9x Faster and My FPS Counter Didn’t MoveYour benchmark is measuring your webcam
Sridhar💲·Jul 14Becoming an AI Infrastructure Engineer, Part 3: The kitchens that actually serve your modelThere is a specific kind of confusion I used to feel early on, and I think a lot of people working with LLMs go through it too. You take…
Michael Hannecke·Jul 13The Race Car in the Parking Garage: Which Inference Engine Wins on the DGX SparkWhy the vLLM-vs-TensorRT-LLM fight is decided before you install anything — and what the second Spark is really for. Everything here…A response icon2A response icon2
InTowards AIbykyon·Jun 28Finding One Pose in a Crowd — Detection and Matching, Split, at 0.16 ms/personOut of a crowd of people moving around, light up only the ones holding a specific pose (say, a tai chi stance), in green, in real time. Not…A response icon2A response icon2
Mantha Sai Gopal·Jun 24Model Quantization and Why We Need It?#1 : Taking your model from the notebook to the production
Mohit Mathur·Jun 23The TensorRT Rabbit Hole That Led Us to torch.autocastHow chasing a complex infrastructure migration taught us the real lever behind a 45% inference speedup — and why we didn’t need TensorRT at…
Sharat Nellltla·May 13The GPU Inference Stack: TensorRT, vLLM, Triton, and ONNX Runtime ComparedBy Sharat Nellutla — Principal Engineering Manager, Microsoft. All content is original and copyright-protected.A response icon2A response icon2
InBumble TechbyAndrei Potapkin·May 1Why 100% GPU Util Wasn’t EnoughA practical loop for optimizing GPU inference without guessing
Maher Hanafi·Apr 5Deep Dive Analysis: Architecting High-Throughput LLM InferenceA Technical Comparison of vLLM, TensorRT-LLM, and Triton Inference Server for Machine Learning Engineers, MLOps Architects and…
