gpu-benchmarking

CUDA matrix multiplication benchmarking on Jetson Orin Nano. Four implementations, three power modes, five matrix sizes. 99.5% mathematical validation. C++/CUDA and Python.

LLM benchmarking, GPU workload orchestration backend server | Deepseek-R1, Qwen2.5, Llama3.1 | 4xRTX-5090 inside PRU2500, 2xH100 inside PRU2500, 8xMI210 in SuperMicro

Analytical performance modeling, VRAM capacity planning, speculative decoding sizing, and hardware roofline prediction for LLM serving.

Standalone LLM inference benchmarking pipelines on AMD GPUs using ROCm, vLLM, MAD, and data visualization scripts.

LLM inference: compare activation and KV-cache sparsity through memory-traffic models and GPU decoding benchmarks.

Reproducible research for measuring agentic AI inference outcomes on NVIDIA Blackwell across cloud environments.

Standardized bare-metal hardware benchmark suite measuring LLM prefill TFLOPS, decode memory bandwidth saturation, and attention kernel efficiency

benchHUB is a Python-based project to parse, aggregate, and visualize system and performance benchmarks. It includes a Streamlit dashboard to display and compare results.

Public, one-way mirror of jgdynamite10/blackwell-agentic-inference-lab. Development, issues, pull requests, releases, and security reporting live in the canonical repository.

Reproducible GPT-2 distributed-training benchmarks on 1-8 V100 GPUs using Slurm, PyTorch, DeepSpeed, NCCL, NVTX, and Nsight Systems.

Measured CUDA graphs trade-off in vLLM 0.23 on H100: per-token decode always faster, but throughput inverts at 32B under saturation (-5.15%) alongside -1.8% tok/J; cold-start cost decomposed into capture and torch.compile.

Artifact-backed LLM serving performance lab for vLLM baselines, official metrics, GuideLLM checks, and SGLang/PD scaffolding

RTX 4090 measurement harness - paired A/B levers, drift and control-spread gates, retraction chain

Run a 2-min local benchmark → predict how long your AI job will take on cloud GPU (T4/V100/A100). No guessing, no wasted money.

Single-GPU LLM inference optimization on NVIDIA DGX Spark and AMD MI300X.

Ramanujan's mathematics meets the NVIDIA stack: CUDA-Q/cuQuantum quantum simulation + NIM/Nemotron analysis, consumer RTX to cloud H100

Reproducible benchmarks for state-space, attention, and hybrid sequence models on NVIDIA GPUs. Measures latency, memory bandwidth, compute utilization, and energy per token to quantify the GPU-SSM mismatch.

Evidence-backed, workload-conditioned autotuning for local LLM inference. Local-first control plane, G0–G8 validation, PyTorch · llama.cpp · vLLM.