Utilyze
an open source GPU monitoring tool more accurate than nvtop
Details
- External ID
- 47921626
- Source
- HN
- Company
- —
- Product
- Utilyze
- Website domain
- systalyze.com
- Launched
- April 27, 2026
- Cohort
- —
- Upvotes
- 128
- Upvotes percentile
- 0.9170951156812339
- Tags
- —
- Fetched at
- Sept. 7, 2026, 9:26 p.m.
- Updated at
- Sept. 7, 2026, 9:26 p.m.
Description
The standard GPU utilization metric reported by nvidia-smi, nvtop, Weights & Biases, Amazon CloudWatch, Google Cloud Monitoring, and Azure Monitor is highly misleading. It reports the fraction of time that any kernel is running on the GPU, which means a GPU can report 100% utilization even if only a small portion of its compute capacity is actually being used. In practice, we've seen workloads with ~1–10% real compute throughput while dashboards show 100%.This becomes a problem when teams rely on that metric for capacity planning or optimization decisions, it can make underutilized systems look saturated.We're releasing an open-source (Apache 2.0) tool, Utilyze, to measure GPU utilization differently. It samples hardware performance counters and reports compute and memory throughput relative to the hardware's theoretical limits. It also estimates an attainable utilization ceiling for a given workload.GitHub link: https://github.com/systalyze/utilyzeWe'd love to hear your thoughts!
Enrichment
- Theme
- gpu compute and acceleration tools
- Vertical
- Horizontal
- Function
- Observability & eval
- Audience
- Developer
- AI stance
- Not AI
- Project type
- Hobby / open-source project
- Normalized one-liner
- gpu monitoring tool
- Manually corrected
- False
Could you build this?
No Accurately measuring GPU utilization beyond standard kernel time requires low-level GPU profiling via NVML or CUPTI, sampling hardware performance counters (SM occupancy, memory bandwidth, tensor core usage) at the driver level.
What it would actually take: Building this requires interfacing with low-level Nvidia APIs like NVML, NVperf, or CUPTI via C/C++ or Rust to sample hardware telemetry counters (e.g., active warps per SM, DRAM throughput, pipeline stalls). This requires deep knowledge of GPU microarchitectures, driver internals, and high-frequency sampling systems that run with negligible runtime overhead on production clusters.
Discussion
20 comments analyzed.
Competitors mentioned: nvtop, nvidia-smi, vibe coding, mesh-llm
Concerns raised: Source code not open/inspectable before installation, Feasibility of memory pressure metric without instrumenting workload, Power-derived metrics overestimate utilization and have lag time, Different model architectures have different realistic utilization ceilings
Feature requests: Memory pressure view showing headroom to OOM cliff, Load balancing tool across multiple local GPUs, AMD GPU support, Scriptable GPU utilization optimization across diverse workloads
Competitors
Other products that read as similar to this one — 133 launches clear the similarity bar, closest 8 shown.
Attention rank: #17 of 134 (itself plus its competitors, highest first — normalized so YC and Product Hunt are compared fairly).
Launched 175 days after the earliest competitor.
- gputop · github · 2026-09-15 · 6 upvotes · similarity 0.59
- PantheonGPU · hn · 2026-08-18 · 13 upvotes · similarity 0.55
- Most GPU Upgrades Aren't Worth It, I Built a Calculator to Prove It · hn · 2026-03-16 · 8 upvotes · similarity 0.49
- ZeroGate · hn · 2026-06-26 · 5 upvotes · similarity 0.49
- Goodput · ph · 2026-09-25 · 2 upvotes · similarity 0.48
- Continuous Nvidia CUDA PC Sampling Profiler · hn · 2026-06-15 · 14 upvotes · similarity 0.46
- The cheapest GPU cloud · hn · 2026-09-03 · 63 upvotes · similarity 0.45
- cuTile Rust: Safe, data-race-free GPU kernels in Rust · hn · 2026-06-16 · 106 upvotes · similarity 0.45
Other launches for this product
- No other launches for this product.
Same idea, different domain
Nobody's really built a observability & eval tool for Media & entertainment yet.