---
date: 2026-06-12
generated_at: 2026-06-12 10:23
artifact_type: 5090-report
status: benchmark-failed
gpu_name: "NVIDIA GeForce RTX 5090"
driver_version: "610.47"
memory_total_mib: "32607"
power_limit_w: "575.00"
benchmark_rows: 0
benchmark_failures: 1
source: nvidia-smi + Reports/5090/benchmarks
canonical: https://bmdpat.com/5090-reports
raw_markdown: https://bmdpat.com/5090-reports/latest.md
failure_artifact: https://bmdpat.com/5090-reports/failures/2026-06-12-gemma4-26b.md
---

# The 5090 Reports - 2026-06-12

Production AI agents on hardware I own. The 5090 is the hook; the artifact is the lab notebook.

## Verdict

Hardware capture is live. The first bounded Ollama benchmark failed; see the failure artifact before publishing tokens/sec claims.

## Hardware Snapshot

| GPU | Driver | Memory used / total MiB | Power draw / limit W | Temp C |
|---|---:|---:|---:|---:|
| NVIDIA GeForce RTX 5090 | 610.47 | 3918 / 32607 | 68.73 / 575.00 | 39 |

## Benchmark Inputs

- No benchmark CSV found yet.
- Add CSVs to Reports/5090/benchmarks/ with columns: model, quant, engine, workload, input_tokens, output_tokens, tokens_per_second, watts_avg, notes.

## Benchmark Failures

- /5090-reports/failures/2026-06-12-gemma4-26b.md

## Benchmark Table

| Model | Quant | Engine | Workload | Input | Output | Tok/sec | Watts avg | Notes |
|---|---|---|---|---:|---:|---:|---:|---|
| pending | pending | pending | agent workload | pending | pending | pending | pending | no tokens/sec claim yet |

## Distribution Contract

- Public posts must include one artifact: this report, a benchmark table, a diagram, or a repo note.
- CTA is the email list on bmdpat.com. No calls, no DMs, no cold outreach.
- One new experiment per week. It must feed production AI agents on owned hardware.

## Next Measurement

Run one real agent workload through the local model stack, capture tokens/sec, watts, latency, and API-equivalent cost.