65fbdb836b
- Based on working minting_plan repository - Configured for threefold.info/economics deployment - Added ops documentation for server deployment - Updated baseUrl and URL configuration
7.4 KiB
sidebar_position
| sidebar_position |
|---|
| 1 |
Understanding Slices
Slices are the fundamental unit of capacity allocation in TFGrid v4. Think of them as standardized "packages" of compute resources that make it easy to price, allocate, and utilize node capacity efficiently.
What is a Slice?
A slice is a defined portion of a node's total resources, including:
- CPU Cores: Processing power allocated to the slice
- Performance (Passmark): Benchmark score indicating CPU performance
- Memory (RAM): Gigabytes of system memory
- Storage (SSD): Gigabytes of fast storage
- TPS (Tokens Per Second): For AI workloads, processing capability
Rather than selling "whole nodes" or arbitrary resource amounts, slices provide:
- Standardization: Predictable resource bundles
- Flexibility: Users can consume 1 slice or 100 slices
- Efficiency: Better matching of supply to demand
- Clarity: Transparent pricing per slice
Why Slices?
The Old Problem
Without slices:
- Users need to understand complex hardware specifications
- Pricing is inconsistent across different node types
- Hard to compare value between providers
- Resources often underutilized or over-provisioned
The Slice Solution
With slices:
- Simple, standardized packages (like buying "small", "medium", "large")
- Consistent pricing model across the network
- Easy comparison: "This slice costs X, provides Y resources"
- Better utilization: Divide nodes into right-sized portions
How Nodes Divide Into Slices
Each physical node can support a maximum number of slices based on its specifications.
Example: 3Node Home
Total Resources:
- 16 CPU cores
- 22,000 Passmark score
- 32 GB RAM
- 1,000 GB SSD
Maximum Slices: 25
Per Slice (resources divided):
- 2.3 cores (16 ÷ 7 slices actually used)
- 3,143 Passmark performance
- 4.0 GB RAM
- 143 GB SSD
:::note Resource Allocation Not all nodes use their maximum slice count. The actual slices used depends on the node's resource constraints. For example, a 3Node Home is limited by RAM, so it provides 7 slices despite having capacity for 25. :::
Slice Types
Different workloads need different resource balances. ThreeFold offers several slice configurations:
Standard Compute Slice
Best for: General applications, web services, APIs
Resources:
- ~2,000 Passmark performance
- 4 GB RAM
- Standard storage
Price: ~1.2 CC/month minimum
Memory-Heavy Slice
Best for: Databases, caching, memory-intensive applications
Resources:
- Standard performance
- 8 GB RAM (double standard)
- Standard storage
Price: ~2.4 CC/month minimum
AI Slice (Standard)
Best for: AI inference, small models, development
Resources:
- Moderate performance
- 124 GB RAM
- Shared GPU access
Price: ~50 CC/month minimum
AI Slice (Dedicated GPU)
Best for: AI training, large models, high-performance inference
Resources:
- High performance
- 124-250 GB RAM
- Dedicated GPU (96-192 GB GPU memory)
Price: 500-1,000 CC/month minimum
See TF Bid Packages for complete specifications.
Slices and Farmer Income
Slices directly determine farmer earnings:
Income Calculation
Monthly Income = (Number of Active Slices) × (Price Per Slice) × (Utilization Rate)
Example:
- Node: Large TFGrid Node with 20 slices
- Price per slice: 7.6 CC (minimum pricing)
- Utilization: 80% (16 slices rented)
- Monthly Income: 16 × 7.6 = 121.6 CC/month (~$15.56 at 1 CC = $0.128)
Income Ranges
Each node type has:
- Minimum Income: Conservative pricing, partial utilization
- Maximum Income: Premium pricing, full utilization
See Pricing Model for detailed projections and ROI analysis.
Slices and the 80% Revenue Share
The slice pricing includes the revenue share model:
- User pays for slice: Example: 10 CC/month
- Revenue distribution:
- 80% → Farmer (8 CC)
- 10% → TFT Burn (1 CC worth of TFT burned)
- 10% → ThreeFold Operations (1 CC)
:::tip Farmer-First Economics With 80% going directly to farmers, the slice model ensures capacity providers receive the majority of value generated. :::
Oversubscription
One important note from the current calculation: CPU oversubscription is not currently factored in.
What is Oversubscription?
In practice, most workloads don't use 100% of allocated CPU 100% of the time. Cloud providers often oversubscribe CPU (sell more CPU cores than physically exist) because:
- Most workloads are bursty (high usage occasionally, low usage normally)
- Statistical multiplexing: not all users peak at the same time
- Better hardware utilization and economics
Future Implementation
TFGrid may implement CPU oversubscription carefully:
- Benefit: Higher effective income per node
- Risk: Must ensure quality of service isn't degraded
- Balance: Conservative ratios (e.g., 1.5x or 2x) rather than aggressive oversubscription
:::caution Current Status The income projections in this documentation assume no CPU oversubscription. Actual income could be higher if careful oversubscription is implemented in the future. :::
Slices in Practice
For Users
Simple shopping experience:
- Browse available slice types
- See clear pricing and resources
- Deploy workloads to slices
- Scale up or down as needed
No complex calculations about "how many cores do I need vs. RAM vs. storage?"
For Farmers
Predictable capacity planning:
- Know your node specifications
- Calculate maximum slices
- Set competitive pricing
- Track utilization and income
Clear comparison between different hardware investments.
Example Scenarios
Scenario 1: Web Hosting Company
Needs: Run 50 WordPress sites
Solution:
- Rent 10 Standard Compute Slices
- Each slice hosts 5 sites
- Cost: 10 × 1.2 CC = 12 CC/month (~$1.54)
- Clear, predictable pricing
Scenario 2: AI Startup
Needs: Train a medium language model
Solution:
- Rent 2 AI Slice Big (Nvidia 2x)
- Dedicated 192 GB GPU memory each
- Cost: 2 × 1,000 CC = 2,000 CC/month (~$256)
- Direct comparison to AWS/GCP alternatives
Scenario 3: Farmer Planning
Hardware: Investing in Mini 3 node
Capacity: 25 slices maximum
Income Target:
- Conservative: 48 CC/month minimum (partial utilization)
- Optimistic: 558 CC/month maximum (full utilization, premium pricing)
- Realistic: ~200-300 CC/month with good network adoption
See Node Specifications for all node types.
Key Takeaways
✅ Slices standardize capacity into easy-to-understand packages
✅ Better resource utilization than whole-node or arbitrary allocations
✅ Clear pricing model for both users and farmers
✅ Flexible scaling - consume exactly what you need
✅ Direct income correlation - more slices rented = more income
✅ Farmer-first revenue share - 80% goes to capacity providers
Next Steps
- Node Specifications - See detailed specs for all node types
- Pricing Model - Understand income projections and ROI
- TF Bid Packages - Explore standard ThreeFold offerings
:::tip Related Concept Slices work seamlessly with the Cloud Credits (CC) stable pricing system, ensuring predictable economics for everyone. :::