AnalysisAI ModelsJuly 3, 2026
Deepseek V4 Flash runs on RTX 5090 MoE
User achieved TG T/S of 22.7 and PP T/S of 1105 with DeepSeek V4 Flash on RTX 5090 using MoE optimizations. The setup uses no unified KV and n-cpu-moe.
User achieved TG T/S of 22.7 and PP T/S of 1105 with DeepSeek V4 Flash on RTX 5090 using MoE optimizations. The setup uses no unified KV and n-cpu-moe.