info_LLM_GPU_GGUF_Speed

Info about what LLM in what GPU how good work with GGUF model

Test promt:"write full rich fastapi todo app"

some of P104-100 8GB

Backend	Model	SizeB	Quant	SizeGB	t/s	Rating	Reasoning	Comment	CanOutOfBoxCodeRun
llama.cpp	cerebras_GLM-4.5-Air-REAP-82B-A12B	82	Q4_K_M	53	4.3	****	Yes	Make Just Backend
llama.cpp	Qwen3-Next-80B-A3B-Thinking-UD	80	Q4_k_XL	43	17	*****	Yes	Multy fl, but wo HTML. Чутка огорчила по исправлению кода по ставнению с gpt-oss-20b
llama.cpp	IQuest-Coder-V1-40B-Instruct	40	q4_k_m	22	7	*****		Write multiple files
llama.cpp	qwen2.5-coder-32b-instruct	32	q4_0	18	10.2	****		Write multiple files
llama.cpp	Qwen3-Coder-30B-A3B-Instruct-UD	30	Q4_K_XL	17	28.5	***		Write single file
llama.cpp	gpt-oss-20b-MXFP4.gguf	20	Q4_K_XL	12	32	*****	Yes	Multy + make dokerfile Не без косячков в кодинге
llama.cpp	DeepCoder-14B-preview	14	q5_0	9	17	***	in this task -yes

RTX3060 12GB

Backend	Model	SizeB	Quant	Size	t/s	Rating	Reasoning	Comment	CanOutOfBoxCodeRun
LMStudio	DeepCoder-14B-preview	14	q5_0	9GB	29	*****	Yes	add React Frontend !)