Strata v0.1.43 has been released, offering up to 38% faster local inference on NVIDIA, AMD, and Intel GPUs while maintaining byte-identical outputs to the previous version. The most significant improvements are seen on a 16 GB Radeon GPU and Intel's Arc Pro B70, which nearly halves prompt processing times. Despite these speed enhancements, critics have raised concerns about the lack of accuracy data and inconsistencies in documentation. To run Strata effectively, users need a powerful GPU and a substantial amount of system RAM, with a minimum recommendation of 64 GB for optimal performance

Strata v0.1.43 Delivers Up to 38% Faster Local Inference

Strata v0.1.43 shipped today, adding faster decode and prompt processing on NVIDIA, AMD, and Intel GPUs without changing any output, answers stay byte-identical to 0.1.42. The biggest jump lands on a 16 GB Radeon at up to 38%, while Intel's Arc Pro B70 sees prompts nearly halve. The speed survives an independent auditor's memory-bandwidth check, but critics flag missing accuracy numbers, self-contradictory docs, and a vision regression. The engine runs a 125-billion-parameter Qwen model on consumer hardware, though that "12 GB" headline really means 64 GB of RAM.

Strata v0.1.43 Delivers Up to 38% Faster Local Inference @ Linux Compatible