Chapter 2 -- The 50-Day Plan For Building A Personal Assistant Agentic System (PAAS)

Daily Resources Augment The Program Of Study With Serindiptious Learning

Outsource Your Big Compute needs

Regardless of whether it is for your work [unless you work as a hdw admin in IT services and would benefit from a home lab], your ventures or side-hustles and any startup that you are contemplating. There are numerous reasons:

  • Outsourcing compute needs instead of purchasing and managing hardware WILL save time, energy, and money

  • This approach teaches extremely valuable and timely lessons about how economic ecosystems have evolve for today's needs.

  • Helps you learn the principles. especially for computing needs. Default to service-based consumption until you can demonstrate with financial precision why ownership creates superior economic value. Only transition to ownership when you can articulate and show specific, quantifiable advantages that overcome the flexibility and scalability benefits of renting. The most successful organizations operate with this discipline rigorously -- the winners defer ownership until comprehensive understanding justifies the commitment; suckers and fools buy cheap, obsolete crap for more than it's worth to save money.

Investigate what is going with alternatives such as ThunderCompute, ie don't just understand their value proposition for the customers vs their competitors, but also understand something about their business model and how they can deliver that value proposition.

  • GPU virtualization achieving up to 80% cost savings ($0.92/hour for A100 GPUs vs $3.21/hour on AWS)
  • Increases GPU utilization from 15-20% to over 90%, ensuring efficient resource allocation
  • Seamless setup process - run existing code on cloud GPUs with a single command
  • Generous free tier with $20/month credit
  • Optimized specifically for AI/ML development, prototyping, and inference
  • Instances behave like Linux machines with physically attached GPUs
  • U.S. Central servers ensuring low latency for US customers
  • Integration with VPCs or data centers for enterprise users
  • Backed by Y Combinator, adding credibility
  • Ideal for startups and small teams with budget constraints

Be sure to routinely update your research on ThunderCompute and other top competitors in cloud GPU computing for startups; for example, VAST.ai has compelling pricing has very interesting auction spot pricing business model which makes it a viable competitor to Thundercompute.

  • Hypercompetitive dynamic auction marketplace with spot pricing starting at $0.30/hour for RTX 3080
  • Real-time benchmarking and ARM64 support
  • Competitive spot market pricing possibly undercuts ThunderCompute
  • Supports graphics and data-intensive workloads
  • Offers wider variety of GPU types
  • Known for flexibility
  • Provides 24/7 support
  • Large user base
  • Hourly billing like ThunderCompute
  • Less focused exclusively on AI/ML than ThunderCompute

Runpod is another with compelling pricing also has very interesting vetted supply chain model that makes it a viable competitor to either VAST.ai or Thundercompute.

  • Active GitHub community developing amazing projects and resources
  • Offers two services: Secure Cloud and Community Cloud
  • More competitive prices than AWS or GCP, though comparable to ThunderCompute
  • Serverless GPUs starting at $0.22/hour
  • Pay-by-the-minute billing
  • Intuitive UI and easier setup
  • Scalable for both short and extended workloads
  • Over 50 pre-configured templates
  • Known for ease of use and community support
  • 24/7 support with community-driven approach (less comprehensive than ThunderCompute)