Update README.md

README.md

@@ -25,6 +25,8 @@ tables, hyper-connections and sparse attention. It is not V4-Flash-0731.
 How to build the engine, serve it, and the gotchas, plus the current status:
 [https://github.com/smalinin/llama.cpp/tree/my_build_deepseek41](https://github.com/smalinin/llama.cpp/tree/my_build_deepseek41)
 
+New runtime binary was tested on 4xRTX4090(48Gb) + 2xRTX3090  with next config:
+
 ```
 ./llama-server \
   --model ./DeepSeek-V4.1-Flash-Q2_K-00001-of-00007.gguf \
@@ -51,6 +53,7 @@ How to build the engine, serve it, and the gotchas, plus the current status:
   --no-prefill-assistant
 
 ```
+Results:  pp = 250 t/s;  tg = 4.7t/s for prompt size=25k
 
 
 ## Status