← Toutes les actualitésLire l'article ↗
The RAM, processing power, and trade-offs you should know about
Synthèse
The obvious question: what kind of computer do you need? Running large language models locally requires substantially more computational resources than typical software. The key constraint is memory — specifically, RAM on most systems, or VRAM if using a dedicated graphics card for acceleration. Block's documentation suggests that 32 gigabytes of RAM provides "a solid baseline for larger models and outputs." For Mac users, this means the computer's unified memory is the primary bottleneck. For W
The obvious question: what kind of computer do you need? Running large language models locally requires substantially more computational resources than typical software. The key constraint is memory — specifically, RAM on most systems, or VRAM if using a dedicated graphics card for acceleration. Block's documentation suggests that 32 gigabytes of RAM provides "a solid baseline for larger models and outputs. " For Mac users, this means the computer's unified memory is the primary bottleneck.
For Windows and Linux users with discrete NVIDIA graphics cards, GPU memory (VRAM) matters more for acceleration. But you don't necessarily need expensive hardware to get started. Smaller models with fewer parameters run on much more modest systems. Qwen 2. 5, for instance, comes in multiple sizes, and the smaller variants can operate effectively on machines with 16 gigabytes of RAM.
"You don't need to run the largest models to get excellent results," Sareen emphasized. The practical recommendation: start with a smaller model to test your workflow, then scale up as needed. For context, Apple's entry-level MacBook Air with 8 gigabytes of RAM would struggle with most capable coding models. But a MacBook Pro with 32 gigabytes — increasingly common among professional developers — handles them comfortably.
Source originale
block.github.io