hardware-requirements

Hardware specifications and compute requirements for running professional poker solvers, from desktop applications to research-grade AI systems.

Table of Contents

• Quick Reference
• Desktop Solvers
  • PioSolver
  • GTO+
  • MonkerSolver
• Cloud vs Local
• Server Rental Options
• Research-Grade AI Systems
• Build Recommendations

Quick Reference

Key insight: Poker solvers are CPU and RAM bound. GPUs provide no benefit.

Desktop Solvers

PioSolver

PioSolver is the industry standard for heads-up No-Limit Hold’em analysis.

Minimum Requirements:

• Windows 64-bit (Mac requires Parallels/Bootcamp)
• 4-core CPU at 2GHz+
• 8GB RAM

Recommended for Postflop:

• Modern 8+ core CPU (AMD Ryzen 7/9)
• 16GB RAM
• NVMe SSD for tree storage

Recommended for Preflop:

• AMD Ryzen 9 or ThreadRipper (16+ cores)
• 64GB RAM minimum, 128GB recommended
• Fast NVMe SSD (trees can be 100GB+)

Performance Scaling:

From PioSolver Hardware FAQ:

“64GB is an absolute minimum for practical purposes” for preflop solving.

GTO+

GTO+ is a budget-friendly alternative with lower system requirements.

Minimum Requirements:

• Windows 7+ 64-bit
• 4 cores at 2GHz
• 8GB RAM
• 10GB storage

Key Advantage: GTO+ uses on-the-fly computation with compression, requiring only a few hundred KB per tree vs several hundred MB for other solvers.

Performance Formula:

Speed = CPU_cores × CPU_frequency
Example: 4 cores × 2GHz = 8GHz effective

MonkerSolver

MonkerSolver specializes in PLO and multi-way pots, requiring significantly more resources.

Minimum Requirements:

• Windows 64-bit or Mac OS X
• 8GB RAM
• Java 64-bit runtime

Recommended for Postflop (Hold’em):

• 16GB RAM
• 8+ core CPU

Recommended for Preflop/PLO:

• 512GB RAM for large PLO sims
• AMD EPYC or ThreadRipper with 64-128 cores
• Multiple NVMe SSDs (2x RAM size for storage)

From MonkerWare Guide:

“MonkerSolver will eat whatever RAM you throw at it.”

Server Configuration for Serious Use:

• 2x AMD EPYC 7282 or better
• 512GB - 2TB RAM
• 1TB+ NVMe SSD

Cloud vs Local

Cloud Solvers (No Local Hardware Needed)

Pros: Zero hardware investment, instant solving, mobile access Cons: Subscription cost, limited customization, internet dependency

Local Solvers (Hardware Required)

Pros: One-time purchase, unlimited solving, full customization Cons: High upfront cost, maintenance, obsolescence

Server Rental Options

For heavy solving without local hardware investment:

Contabo

contabo.com

OVHcloud

ovhcloud.com

• AMD EPYC dedicated servers with water cooling
• EU and US data centers
• Higher price, better support

MonkerGuide Custom Servers

monkerguide.com/site/servers

Specialized poker solver hosting with:

• Pre-configured MonkerSolver environments
• Up to 2TB RAM, 128 cores
• Quote-based pricing

Cost Comparison

Recommendation: Rent servers for occasional heavy use; buy hardware if solving daily.

Research-Grade AI Systems

Libratus (2017)

CMU’s Libratus defeated top professionals in heads-up No-Limit Hold’em.

Compute Used:

• 25 million core hours total
• Pittsburgh Supercomputing Center’s Bridges supercomputer
• 196 nodes for equilibrium finding
• 100 CPUs during live play

Breakdown:

Pluribus (2019)

Facebook/CMU’s Pluribus beat professionals in 6-player No-Limit Hold’em.

Remarkable Efficiency:

• Blueprint computed in 8 days with 12,400 core hours
• Live play: Just 2 Intel Haswell E5-2695 v3 CPUs
• Memory: Under 128GB RAM
• Used 28 cores during live play

Comparison to Other AI:

Pluribus demonstrated that efficient algorithms matter more than raw compute.

Build Recommendations

Budget Build (~$1,500)

For GTO+, basic PioSolver postflop

CPU: AMD Ryzen 7 7700X (8 cores)
RAM: 64GB DDR5-5200
SSD: 1TB NVMe
GPU: Basic (GT 1030 or integrated)
PSU: 550W

Mid-Range Build (~$3,000)

For PioSolver preflop, basic MonkerSolver

CPU: AMD Ryzen 9 7950X (16 cores)
RAM: 128GB DDR5-5600
SSD: 2TB NVMe
GPU: Basic (for display only)
PSU: 750W
Cooling: 360mm AIO

High-End Build (~$8,000)

For MonkerSolver PLO, heavy preflop solving

CPU: AMD Threadripper 7960X (24 cores)
Motherboard: TRX50 workstation
RAM: 256GB DDR5 ECC
SSD: 4TB NVMe RAID
GPU: Basic
PSU: 1000W
Cooling: Custom loop or 420mm AIO

Workstation Build (~$20,000+)

For professional/commercial use

CPU: Dual AMD EPYC 9354 (64 cores total)
Motherboard: Dual socket SP5
RAM: 512GB - 1TB DDR5 ECC
SSD: 8TB NVMe array
Enterprise cooling
Redundant PSU

Key Takeaways

1. RAM is king - More RAM = bigger trees = more accurate solutions
2. Cores matter - Speed scales linearly with core count
3. GPU doesn’t help - Save money, get a basic display card
4. AMD dominates - Better performance/dollar for multi-threaded workloads
5. Consider renting - Servers make sense for occasional heavy use
6. Efficient algorithms win - Pluribus solved 6-player poker on modest hardware

Sources

1. PioSolver Hardware FAQ
2. MonkerWare Getting Started Guide
3. GTO+ Official Site
4. Contabo Dedicated Servers
5. OVHcloud AMD Servers
6. MonkerGuide Server Rental
7. SolveOptimized Server Guides
8. Libratus - Wikipedia
9. Pluribus - Science Paper
10. PSC Pluribus Article
11. PCPartPicker Poker Solver Builds