A few days ago, I upgraded my system with a 3090 that I purchased.
I was thinking about trying dual 3090s, but after searching, it seems the effect isn't that great and there aren't many uses for it, so I decided to put it on hold for now.
It also seems like running dual 3090s would be difficult with a 1000w power supply, so I only installed one for now.
My previous CPU and RAM were a 5700x and 2666 96G, but the system that came with this one has a 5900x and 3600 64G.
After pre-installing the SSD and using the hermes agent to measure,
the TPS which used to be around 45 increased to about 52.
It's said to be about 10% faster, but personally, I feel that the response start is a bit faster.
My local llm is running on llama.cpp-server and the model is Qwen3.6-27b-MTP.
The CPU and memory have seen a significant performance improvement, it seems that llm almost entirely utilizes the GPU.
