DeepSeek V4 Flash runs on RTX 5090 with 1M context via llama.cpp

A Reddit user shares benchmarks of DeepSeek V4 Flash running on a single RTX 5090 with 1 million token context via llama.cpp, using Unsloth's Q8 quantized version. The post includes configuration details and performance results.