Step 2: CPU & RAM Installation

The computing subsystem (processors and RAM) is responsible for the overall coordination of the DePIN node, data routing, and the smooth operation of virtualization containers. Errors at this stage lead to a critical drop in SLA due to delays in packet processing.

1. Processor Options: Intel Xeon Gold 6130 / 6140 (Dual CPU Configuration)

To ensure maximum fault tolerance and parallel processing, the system is equipped with two physical processors from the Intel Xeon Gold family.

Why the Xeon Gold, and not the flagship desktop processors (Core i9 / Ryzen)?

  • Multithreading and Caching: DePIN’s server tasks require a massive amount of L3 cache for instant transaction processing and GPU coordination. The Xeon features an architecture designed for continuous parallel data read/write operations. Desktop processors are optimized for short-term “peak” loads and quickly overheat (and enter throttling mode) when operating 24/7.
  • Number of PCIe Lanes: Two Xeon Gold processors provide a significantly higher number of PCI Express lanes directly from the processor. This is critical for conflict-free connection of four high-performance graphics cards via risers and high-speed U.2 NVMe drives. On a desktop platform, the graphics cards would begin to “crowd each other out” in terms of bus bandwidth.
  • Service Life: Server processor chips undergo rigorous factory screening and are designed for continuous operation at 100% load for 5–7 years without silicon degradation.
    • Engineering Life Hack:
    • Choosing a CPU for the “Quick Start” Scenario (without graphics cards)
      ​Our website’s homepage describes the “Quick Start” strategy, in which a server is launched without any graphics cards at all (relying solely on CPU and RAM power) and generates between $300 and $600 per month on general-purpose networks (for example, Akash Network).
      ​If you plan to launch using this model or want to maximize your potential for future rentals, pay attention to the following detail when purchasing processors:
      ​Base configuration: 2 processors with 16 cores each (total of 32 cores / 64 compute threads).
      ​Recommended option: 2 processors with 18 cores each (36 cores total / 72 compute threads).
      ​Why should you pay a few dozen dollars more to get 18 cores?
      ​More virtual machines (vCPUs): Decentralized networks divide your server’s resources into small packages for clients. Having 72 threads instead of 64 allows you to host more concurrent tenants on a single server, which directly increases stability and monthly revenue.
      ​Perfect compatibility with RAM: The memory controller architecture in these processors is identical. Our base configuration of 8 DDR4 ECC memory modules (4 modules per CPU) will operate with the same stability in both the 16-core and 18-core versions, without any risk of “crashes” or a drop in the node rating.
      ​A Foundation for the Future: When you eventually move up to the top tier of AI computing (after purchasing additional graphics cards), more powerful processors will ensure better overall system performance and eliminate any bottlenecks in data transfer from storage devices to the GPU.
      Summary: Don’t be lazy—find and install 18-core processors. The price difference between them on the secondary market is negligible, but as a standalone unit operating without a graphics card, this setup will prove to be significantly more effective.

2. An Important Detail: CPU Cooling and the LGA 6130/6140 Socket Pitfall

When choosing coolers for our Supermicro X11-DAI-N motherboard, there’s a major pitfall that many PC builders run into:

  • Mounting Form Factor (Square vs. Narrow): The LGA 6130/6140 socket has two different mounting frame variants— Square and Narrow (rectangular). Our board uses a specific revision. You need coolers that strictly match the mounting type on the board. Rectangular coolers with elongated metal “ears” will either not physically align with the mounting holes or will not provide sufficient pressure against the processor’s heat spreader.
  • Active cooling: Use only active coolers (heatsinks with their own high-performance fans). In standard data centers, servers are cooled by airflow passing through a powerful rack (“wall” of fans”), while the processors themselves are equipped with passive aluminum heatsinks. In a custom build inside a 4U chassis, where four hot graphics cards are installed side by side, passive cooling will cause the processors to overheat instantly.

3. RAM: 256 GB (8 x 32 GB) DDR4 ECC

Server memory is a buffer through which terabytes of data are processed for AI training or rendering.

ECC (Error Correction Code) Technology — A Prerequisite

In ordinary home computers, heat, electromagnetic interference, or cosmic radiation can periodically cause a “bit flip”—when a 0 turns into a 1. For a home PC, this means, at worst, a game crash. For a DePIN node, a single bit flip in operational data means the failure of the entire computing task, a loss of network trust, and a penalty (slashing) resulting in a drop in SLA. Hardware-level ECC memory detects and corrects single-bit memory errors on the fly, guaranteeing absolute data accuracy.

Module Configuration (8 modules, 32 GB each)

  • Multi-channel mode: Installing 8 identical memory modules allows all memory channels of both Xeon processors to be fully utilized. This doubles the memory subsystem’s bandwidth compared to the standard dual-channel mode used in desktops.
  • Symmetry: Memory must be distributed strictly equally between the first and second processor slots (4 modules per processor), as shown in the diagram in the Supermicro manual; otherwise, the system will not balance the load on the Ultra Path Interconnect (UPI) bus.
    • Engineering Life Hack:
    • RAM Optimization and Enabling 6-Channel Mode
      ​RAM in DePIN servers isn’t just “capacity for the sake of it.” Its throughput (the speed of data exchange with the processor) directly determines how efficiently the system will process tasks in “Quick Start” (in networks such as Akash) and how quickly it will transfer large data sets to graphics cards when working with artificial intelligence.
      ​We offer two DDR4 ECC RAM configuration options for you to choose from.
      ​Option 1: Basic Standard (Recommended for cost-effectiveness)
      ​Configuration: 8 modules of 32 GB each (256 GB total). Four modules are installed in the slots of each of the two processors.
      ​Pros: Minimal initial investment, excellent capacity—more than enough to handle 90% of standard DePIN network operations. An excellent solution for a quick return on investment.
      Option 2: Top Optimization (True 6-Channel Mode)
      If you want to squeeze the absolute maximum performance out of the Intel Xeon Gold processor architecture and eliminate any micro-latencies, we recommend this configuration:
      ​Configuration: 12 32 GB memory modules (384 GB total). Six modules are installed on each processor.
      ​Process Physics: Why Are 6 Transistors Per Processor the Technical Ideal?
      ​Unlocking Hidden Bandwidth: Every processor in the Intel Xeon Gold family (architectures compatible with our Supermicro X11 motherboard) features a built-in memory controller that is physically designed for 6 independent channels.
      The “bottleneck” effect: When we install 4 memory modules in the CPU, we’re only using 4 of the 6 channels. The system runs stably, but the CPU isn’t utilizing its memory bus to its full potential. Installing six memory modules in parallel activates all six channels. The data transfer rate between RAM and the CPU cores instantly increases by 20–30%.
      ​Priority in “Quick Start”: Cloud networks instantly detect not only the capacity but also the response speed of the memory subsystem. A server with true 6-channel mode and 384 GB of RAM receives the highest technical score and secures the most expensive and resource-intensive contracts from clients.
      ​Full utilization of resources: With this configuration, there are no unused “locked” reserves left in your system. The processors (especially if you’ve chosen the recommended 18-core models) start running at full capacity, utilizing every thread 100%.
      ⚠️ Engineering Tip: The beauty of our server platform is that you can perform this upgrade in stages. If your budget is limited at the start, begin with the basic 8 memory modules (256 GB). In the future, once the server starts generating a net profit, you can purchase 4 more identical 32 GB memory modules at any time, install them in the empty slots, and enable the top-tier 6-channel mode.
      Summary: To save money and get started quickly, go with 8 memory modules. For uncompromising engineering precision, maximum speed, and future-proofing for heavy AI computations—don’t hesitate to upgrade your system to 12 memory modules. The choice is always yours.