My rap system is currently running Gold 6146 with DDR4 2400t in a 6-channel full configuration.
The memory bandwidth is estimated to be around 90~100GB.
I'm using two 5060ti 16GB in dual mode here.
TP 10,000 is supported. <- Although there are two 5060tis installed, only one can be used.
When I ran DeepSeek-V4-Flash-0731 with a single GPU, it was about 13~14 t/s
It does run... but it doesn't seem to be practical.
I don't know how the internal structure is...
It seems that only the models listed as supported are running.
I tried downloading Qwen3-Coder-Next-FP8 just in case, but it doesn't work. ㅠ.ㅠ