SoAI Bench · leaderboard

GPU leaderboard

Certified SoAIBench results published to soai.to, grouped by GPU model. Each model is ranked on the median of its published runs, not on its best single run. Open a model to read every run behind that median, with the lowest and highest score those runs produced.

Published GPU models, ranked by median score
# GPU model Vendor Runs Median overall Compute Memory Latency

SoAI Bench · scoring

What a score means

SoAIBench runs synthetic OpenCL workloads on one selected GPU and reduces them to four numbers. Only a certified run on the standard profile can be ranked, and soai.to recalculates every number from the submitted evidence before it appears here.

  • The median of five passes

    A certified run does one warmup pass and then five measured passes, and the score it publishes is the median of those five. The run only becomes rankable if the five passes agreed within 10 percent, the card stayed under its temperature limit, nothing throttled, and both the temperature and power sensors kept reporting for the whole run.

  • How the overall score is built

    Overall score is compute × 0.62 + memory × 0.26 + latency × 0.12. Compute is the mean of the ALU, matrix, transcendental, and mixed workloads in GOPS. Memory is the mean of the memory and mixed workloads in GB/s. Latency counts kernel dispatches per second divided by 20. Higher is better in all four columns.

  • What validation proves

    soai.to recomputes each pass, the medians, the variance, and the eligibility rules from the uploaded evidence, and rejects a submission whose own totals disagree with that recalculation. A validated result is therefore internally consistent with the SoAIBench V1 protocol. Nothing in it proves which binary ran on which card.

SoAI Bench · publishing

How a result gets published

SoAI cannot publish a result yet. When that ships, a finished certified run will gain a publish action that sends one request and receives a public URL in return.

  • Certified runs only

    In SoAI you open Hardware, pick one GPU, and start a certified SoAIBench run on the standard profile. Quick runs and stress runs stay in your local history and are never ranked.

  • No account

    There is nothing to sign up for. Before anything leaves the machine you see the exact payload: GPU model, vendor, driver, OpenCL runtime, the five passes, telemetry, and a display name you can set or leave as Anonymous. A public result never carries the GPU UUID, the PCI address, the hostname, the IP address, or any local device identifier.

  • Deleting a published result

    SoAI generates a secret before it publishes and keeps that secret locally. The submission carries only a one-way verifier of it. Presenting the secret later removes the result, so a published score can be withdrawn without an account.

Download SoAI Hardware & GPU docs