LLM Build Recommendations

Three builds optimized for running LLMs locally. Each prioritizes VRAM and memory bandwidth over gaming performance. Prices are approximate mid-2026 market. Methodology.

Budget — $498

Part

Part	Model	Price
CPU	AMD Ryzen 5 5600 (6C/12T)	$95
Motherboard	B550 AM4 ATX	$90
RAM	32GB (2×16) DDR4-3200	$55
GPU	Used RTX 3060 12GB	$185
Storage	1TB NVMe SSD	$55
PSU	650W 80+ Bronze	$60
Case	Budget ATX Mid Tower	$40
Total		$498

Compatible Models (Q4 8K)

Model	Params	tok/s
MiniCPM5-1B	1B	212.1 tok/s
Qwen3.5-2B	2B	114.5 tok/s
DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx	2B	99.6 tok/s
Llama-3.2-3B-Instruct	3B	71.6 tok/s
Qwen3-4B-Instruct-2507	4B	57.3 tok/s

Why this build

Best 12GB VRAM card under $200 used. 360 GB/s bandwidth handles 8B models at Q4 with room to spare. Paired with a 6-core Ryzen and 32GB RAM for headroom. Full GPU details →

Mid-Range — $998

Part

Part	Model	Price
CPU	AMD Ryzen 5 7600 (6C/12T)	$180
Motherboard	B650 AM5 ATX	$140
RAM	32GB (2×16) DDR5-6000	$85
GPU	RTX 4060 Ti 16GB (new)	$450
Storage	1TB NVMe Gen4 SSD	$65
PSU	750W 80+ Gold	$85
Case	ATX Mid Tower w/ mesh	$70
Total		$998

Compatible Models (Q4 8K)

Model	Params	tok/s
MiniCPM5-1B	1B	169.7 tok/s
Qwen3.5-2B	2B	91.6 tok/s
DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx	2B	79.7 tok/s
Llama-3.2-3B-Instruct	3B	57.3 tok/s
gpt-oss-20b	22B	50.9 tok/s

Why this build

16GB VRAM at an affordable new price. Runs all 8B models comfortably and fits 14B models at Q4. 32GB system RAM and DDR5 give future upgrade headroom. Full GPU details →

Enthusiast — $1985

Part

Part	Model	Price
CPU	AMD Ryzen 7 7800X3D (8C/16T)	$340
Motherboard	B650 AM5 ATX	$150
RAM	64GB (2×32) DDR5-6000	$175
GPU	Used RTX 4090 24GB	$1050
Storage	2TB NVMe Gen4 SSD	$110
PSU	1000W 80+ Gold	$150
Case	ATX Mid Tower w/ mesh	$80
Total		$1985

Compatible Models (Q4 8K)

Model	Params	tok/s
MiniCPM5-1B	1B	593.9 tok/s
Qwen3.5-2B	2B	320.7 tok/s
DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx	2B	278.9 tok/s
Qwen3.6-35B-A3B	36B	213.8 tok/s
Qwen3.5-35B-A3B	35B	213.8 tok/s

Why this build

24GB of VRAM at 1008 GB/s — the ultimate consumer LLM card. Fits 32B models at Q4 and 70B models with partial offload. 64GB system RAM handles large context windows. Used 4090 saves ~$600 vs new. Full GPU details →

Prices are approximate mid-2026 estimates. Used GPU prices vary. All builds assume you have a monitor, keyboard, and mouse. No OS cost included.