NVIDIA / HARDWARE FILE
L40S 48GB
data-center inference and image generation
Installed memory48 GB
Usable assumption48.0 GB
Memory typeVRAM
Software pathCUDA
Q4 MAP
What can fit
COMFORTABLE
Gemma 3 4B
small multimodal assistant
4.5 GB estimatedInspect →COMFORTABLEStable Diffusion 1.5
local image generation
4.6 GB estimatedInspect →COMFORTABLEWhisper Large v3
speech recognition
5.1 GB estimatedInspect →COMFORTABLEDeepSeek Coder 6.7B
local coding assistant
6.2 GB estimatedInspect →COMFORTABLEMusicGen Large
local music generation
6.3 GB estimatedInspect →COMFORTABLEMistral 7B v0.3
general local inference
6.6 GB estimatedInspect →COMFORTABLEQwen2.5 7B
compact multilingual chat
6.8 GB estimatedInspect →COMFORTABLEDeepSeek-R1 Distill Qwen 7B
compact distilled reasoning
6.8 GB estimatedInspect →COMFORTABLELlama 3.1 8B
chat, RAG, general assistants
7.1 GB estimatedInspect →COMFORTABLEQwen3 8B
compact reasoning and chat
7.2 GB estimatedInspect →COMFORTABLEStable Diffusion XL
high-resolution image generation
8.6 GB estimatedInspect →COMFORTABLEGemma 3 12B
mid-size multimodal assistant
9.9 GB estimatedInspect →Check power supply, cooling, physical slot size, OS, drivers, runtime support, and total system cost before buying.