Working from home in the AI era has forced me to adapt and find ways to efficiently manage multiple computers as a single system. The challenge I faced was handling intensive tasks like gold trading alongside high-resource AI work. How to distribute work among three different machines with distinct but interconnected roles became an interesting project for me.
Assigning Roles to Each Machine
My Mac Mini M4 Pro serves as the primary machine for high-speed tasks like daytime work, opening MetaTrader 5 for gold trading, and daily use of the iCafeFX app. This is where I need the fastest response possible because gold trading is all about real-time decision making.
The private NAS acts as the data center, storing all data generated from the other two machines. It can be compared to an anchor that holds our computer system together as one unit. Everything is stored here - from gold trading data and price history to AI work completed.
RTX 5060 Ti Workstation for Heavy Tasks
The Windows Workstation equipped with an RTX 5060 Ti is dedicated to specialized heavy-duty tasks. I use this machine for training AI models and rendering work that requires high processing power. This is where I experiment with various models, particularly the FLUX model for image generation and the Typhoon model for complex natural language processing tasks.
This machine has enough power to run large models, especially when I need to experiment with new models that aren't ready for real-world use yet. Our team has configured this machine to work as a backend computation server, allowing the Mac Mini to operate smoothly.
System Collaboration
The most crucial aspect is the collaboration between these three machines. I use Redhat MCP AI as an intermediary for connecting data between machines, enabling data to be shared and used continuously. For example, when I'm working on the Mac Mini and need to use a model in the Windows Workstation, the data can be quickly sent to that machine.
Creating an efficient workflow takes considerable time, especially ensuring each machine understands its role and works simultaneously without confusion. Having the private NAS as the data center helps these connections occur seamlessly.
Real-World Performance
After implementing this system for a while, I measured its real-world performance, particularly in AI tasks. The initial transcription work can respond in just ~3.4 seconds, which is very fast for this type of work. The language models running on the machine can process at 25-28 tokens/second, which is sufficient for daily use.
Text similarity testing achieved results as high as 0.9956, showing that the system works with high precision. Whether it's gold trading or AI work, everything can run simultaneously without performance issues.
Lessons from a Multi-Machine Setup
Setting up a multi-computer system is not easy at all. I've learned many lessons, especially regarding network connection management and file sharing. What's important is clearly defining roles so each machine knows what it can and cannot do.
Another critical aspect is ensuring important data is backed up regularly. With the private NAS serving as the data center, I'm confident that all data is safe and accessible from any machine in the system.
Next Steps for the System
Looking ahead, I plan to further develop this system, particularly in connecting AI systems with gold trading to create models that can analyze gold trends more effectively. Additionally, experimenting with new models to improve work efficiency is another goal I'm focusing on.
For those interested in trying real gold trading, open an XM account at: open an XM account via our partner
