RyanLee(@RyanLeeMiniMax):thanks to @xieenze_jr , now we have 3.92× on DGX Spark and 4.52× on GeForce RTX 5090

2026.08.10 23:32

thanks to @xieenze_jr , now we have 3.92× on DGX Spark and 4.52× on GeForce RTX 5090

2026.08.05 17:47

🚀 Two days later, MiniMax-H3 goes from the data center to the desktop with Sol Engine. ⚡ 3.92× faster on DGX Spark 480p · 5s · 24 FPS ⚡ 4.52× faster on RTX 5090 720p · 5s · 24 FPS Powered by full-stack kernel optimization, Sol-Attn, and cross-step caching—using the stock 33B checkpoint, with no distillation, fine-tuning, LoRA, or offline calibration. This is what agent-native optimization is built for: adapting acceleration recipes across models, hardware, and deployment settings—from GB200 to desktop AI systems. 🔗