Flopper Brings GPU Specs, AI Performance, and Pricing Together

Summary

Flopper is a hardware explorer built to make GPU research for AI and other compute workloads easier. It brings together specifications from GPUs such as those from Nvidia, AMD, and Huawei, including TDP and estimated throughput expressed in FLOPS or TOPS, alongside configurations, vendors, and pricing. The project recently added price data from dozens of providers so users can track how prices change over time, although reliable quotes for systems such as the NVL72 remain difficult because they are often handled through large procurement contracts. The author says interest in consumer GPUs and local LLMs has become especially clear. A key unresolved issue is the gap between advertised specifications and real-world performance: testing so far has produced roughly 85% of advertised throughput, but the author stresses that this is only a rough observation for the workloads tested and varies with power, cooling, and workload. The measurements are partly best-effort calculations, so the author is seeking corrections and feedback. A possible next step is a CLI that would let users run benchmarks and submit results, potentially with MLPerf automation, so performance can be compared alongside workload and system configuration.