Is RPC broken for llama-server?

[psyopsis930]

I have two machines, both Windows, both built llama.cpp from source with -DGGML_RPC=ON, both on Release b10358 and for some reason it can only successfully connect on llama-cli. I need llama-server to connect to a harness.

PS C:\Desktop\dev\llama.cpp\build\bin\Release> ./llama-server -m "C:\models\Qwen3.6-27B-UD-Q4_K_XL.gguf" -ngl 99 --rpc 10.0.15.214:50052

 E Failed to connect to 10.0.15.214:50052

The same command replaced with llama-cli works (I've already opened up the firewall for LAN access on the worker node). From digging through online discussions, it has clearly worked for people at some point but I cannot find anything recent, I am unsure what I am doing wrong.

[psyopsis930]

I've tried rebuilding llama.cpp multiple times already and even matched build tool versions. I'll leave this up for another few days before moving to VLLM for distributed inference.

[am17an]

Create an issue with a reproduction