Systalyze just open-sourced Utilyze, a GPU monitoring tool built for a simple reason: your GPU dashboard is probably lying to you. The standard 'GPU utilization' metric reported by nvidia-smi, nvtop, and cloud monitoring services from AWS, Google, and Azure doesn't measure how hard your GPU is actually working. It just checks whether something, anything, is running. A single kernel executing on one of an H100's 17,424 cores reads as 100% utilization. Tools like zml-smi want to replace nvidia-smi for everything, offering real-time performance metrics for GPUs, TPUs, and NPUs. The real numbers are brutal. Systalyze found production deployments where GPUs ran at as little as 1% of actual capacity while standard tools showed full saturation. Teams buy hardware they don't need because they can't see what they already have. GPU costs keep climbing. NVIDIA H100 one-year rental contracts rose nearly 40% between October 2025 and March 2026. Lead times stretch months. Every percentage point of throughput recovered from existing GPUs is hardware you didn't acquire and power you didn't burn. Utilyze uses hardware performance counters to measure actual compute efficiency in real time with negligible overhead. Manya Ghobadi, MIT Professor and CEO of Systalyze, frames it plainly: 'The gap isn't awareness. Engineers who write CUDA kernels know what accurate utilization looks like. The gap is tooling. There has never been a way to see true GPU efficiency continuously, in production, without slowing down the workload.' Some Hacker News commenters pointed out that power draw can indicate actual GPU load, and NVIDIA's nsight systems exists for profiling. But Utilyze targets continuous production monitoring, not one-off debug sessions.