Choosing a model for a 384GB in-house AI server
A company is considering an in-house AI server with four s and 384GB of total memory. Two or three people would be heavy users, while peak demand could reach 10 to 20 simultaneous users.
The system would mainly handle internal policies, company data, analysis, and other tasks that require . Software development would be a minor use.
is the current p, and the machine may later expand to eight s. No model comparison, speed test, operating-cost estimate, or final decision is provided.
Key points
- The planned server has four s and 384GB of total memory.
- Two or three people would be heavy users, with up to 10 to 20 simultaneous users.
- The main uses are internal policies, company data, analysis, and tasks.
- is the current p, with a possible expansion to eight cards.
- There are no test results, cost figures, or final model choice yet.