Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s

A developer has successfully run a 125 billion instruction per second (125B) version of the Qwen 3.8 Flash Next CPU benchmark on consumer-grade hardware, specifically an NVIDIA GeForce RTX 4090 graphics card. This achievement is notable because it pushes the limits of what is thought to be the maximum performance of consumer hardware. The developer, Niko, has shared their work on GitHub, providing a detailed description of the project and the process used to achieve this feat. The benchmark, Strata, is designed to test the performance of a CPU's instruction throughput, and Niko's achievement demonstrates that consumer-grade hardware can handle extremely high-performance workloads.