ERPC upgrades its large Solana RPC nodes in Frankfurt with EPYC 5th Gen CPUs (Turin). Advancing HTTP / WebSocket / Unlimited Endpoints to the next performance tier for workloads that compete at the millisecond level

ERPC, operated by ELSOUL LABO B.V. (Headquarters: Amsterdam, the Netherlands; CEO: Fumitake Kawasaki) and Validators DAO, has completed the deployment of EPYC 5th Gen CPUs (Turin) to its large RPC nodes in the Frankfurt (FRA) region.

With this upgrade, processing headroom under peak load and high concurrency has been expanded for RPC workloads, particularly across HTTP, WebSocket, and Unlimited Endpoints.

ERPC has continued to invest in Frankfurt because the conditions that allow results to remain stable in real-world Solana operations are concentrated in this region.

Frankfurt hosts a high density of major validators and stake, creating a structure in which operations such as Shreds reception, catch-up, voting, and state updates can proceed continuously over short network paths. This structure is not merely a matter of geographic proximity, but is based on the actual placement of execution participants and network routes.

As a result, not only is average latency kept low, but latency variance is also suppressed, allowing execution results to remain stable even when multiple external factors overlap. These characteristics form a key prerequisite for maintaining reproducible results within ERPC’s RPC execution infrastructure.

The introduction of EPYC 5th Gen CPUs (Turin) strengthens the execution base by simultaneously improving average response times, peak-time processing headroom, and stability under high concurrency.

For developers and financial use cases, including high-frequency trading, where outcomes are decided at the millisecond level, behavior during moments of concentrated load or overlapping concurrent execution directly affects results. This upgrade advances the execution base to the next performance tier so that RPC processing continues without congestion and execution can be sustained even in these scenarios.

For high-load usage, including Unlimited Endpoints, the configuration ensures stable processing headroom even as concurrent connections and request density increase.

EPYC 5th Gen (Turin) is a CPU generation designed to operate continuously at high clock levels while supporting a large number of cores and threads.

Solana workloads are characterized by rapid shifts in load state and consecutive peaks. In such environments, the ability to maintain both computational performance and operational stability under overlapping load conditions directly affects catch-up behavior and the stability of execution results.

With significant architectural improvements, the Turin generation offers a balanced combination of maximum performance and sustained performance, making it a rational choice as the core of an RPC execution platform for Solana.

In Solana RPC workloads, computational processing and memory access occur frequently and in parallel. What matters here is not only peak bandwidth, but also a configuration in which memory speed and behavior remain consistent under sustained load.

On desktop-class platforms, memory configurations with more DIMMs generally impose lower maximum memory speeds. For example, the specifications for the Ryzen 9 9950X define DDR5-5600 for two-DIMM configurations and DDR5-3600 for four-DIMM configurations.

For the large RPC nodes in Frankfurt, ERPC has adopted EPYC 5th Gen (Turin) with a server-oriented, multi-channel memory configuration. The EPYC 9005 (Turin) generation provides a 12-channel memory architecture, and supported platforms allow high-bandwidth memory conditions including DDR5-6400.

In Solana node operations, large volumes of data are continuously read from and written to the ledger and blockstore. Under these conditions, storage I/O characteristics directly influence execution outcomes.

At present, many operational setups are based on EPYC 4th Gen CPUs combined with NVMe Gen4 storage. NVMe Gen5, however, faces constraints in both availability and operational requirements, and only a limited number of configurations can be used stably in production environments.

The NVMe Gen5 storage introduced in this upgrade has been selected to maintain stable I/O characteristics even under continuous read and write workloads typical of Solana operations. In collaboration with partner data centers, multiple NVMe Gen5 options were tested, and those most suitable for Solana workloads were selected.

ERPC originated from our own project operations.

While executing processes such as NFT card minting and 100-pull gacha mechanics that generate large volumes of transactions in short periods, we encountered situations where existing RPC environments could not keep up, resulting in unstable execution outcomes. To address this, we needed to build an RPC environment capable of handling our own use cases.

Subsequently, by examining challenges faced by other projects, we identified many cases in which RPC infrastructure became the bottleneck under high-load processing. When these projects used the infrastructure we had built for their actual workloads, we received concrete evaluations regarding execution success rates and stability under load.

Based on these real-world operational experiences, ERPC has been designed with the assumption that it must continue to withstand high-load transaction processing. Improvements to the RPC execution base are ongoing, driven by actual usage patterns and feedback.

ERPC began with hands-on experience gained through operating our own projects.

To stably execute high-load transaction processing, it was necessary to design the entire execution base on the RPC side, including configuration, processing headroom, and catch-up behavior. This requirement was not unique to a single project, but common across many Solana projects.

ERPC aims to provide a reproducible execution foundation that enables high-quality Solana network operations in real-world environments. Going forward, we will continue to improve the RPC execution base based on real workloads and usage conditions.

Practical usage and feedback from users further strengthen this foundation.

As a return on this feedback, ERPC will continue to invest in improving the performance and durability of its RPC execution infrastructure.

Thank you for your continued support of ERPC.

Validators DAO Official Discord: https://discord.gg/C7ZQSrCkYR
ERPC Official Website: https://erpc.global/en