Brief conclusions
- The 1C:Â Enterprise architecture is critically dependent on single-threaded performance; server processors must have a base clock frequency of at least 3.6 GHz (ideally 4.0 GHz in Turbo mode).
- Classic SATA SSDs are completely unsuitable for databases larger than 100 GB; the 2026 standard requires NVMe PCIe 5.0 drives with a minimum IOPS rating of 1,000,000.
- Incorrect configuration of NUMA nodes on multiprocessor servers reduces the speed of access to RAM by 30-40%, causing degradation of the entire system.
- The transition of the corporate sector to the PostgreSQL DBMS requires an increase in the server’s RAM by 20-30% compared to similar databases on MS SQL Server .
1. Specifics of the 1C:Enterprise workload
The 1C:Â Enterprise system is a classic example of a three-tier architecture (Client – Application Server – DBMS Server), where each layer requires specific hardware. The platform’s key feature is the strict sequencing (serialization) of transactions when writing registers.
Amdahl’s Law operates mercilessly in the 1C environment: the overall system speed is limited by the processing speed of a single data stream. This means that a server with 64 cores at 2.2 GHz will run 1C twice as slowly as a server with 8 cores at 4.5 GHz.
Waiting for a response from a slow disk ( I/O Wait ) immediately creates transaction lock queues. If a lock lasts longer than 15 seconds, users receive a timeout error and their sessions are terminated.
2. Processors (CPU): Debunking the Core Myth
When choosing a processor for 1C, you should focus solely on the base clock frequency and L3 cache size. For the 1C application server, the optimal choice in 2026 is the Intel Xeon Gold 65xx series (e.g., 6544Y) or AMD EPYC F-series (Frequency-Optimized).
Multi-core processors are only needed for a dedicated database server (DBMS), where each request from the 1C server can be processed in parallel. For an application server, purchasing a processor with more than 16 cores is a waste of the IT budget.
“The biggest mistake system administrators will make in 2026 is buying used servers with 2.0 GHz Xeon v4 processors for ERP databases. 1C requires Hz. If the core frequency is below 3.5 GHz, no amount of SSD arrays or terabytes of RAM will save you from block queues.”
The NUMA (Non-Uniform Memory Access) architecture plays a huge role in dual-processor servers. If the 1C server is configured incorrectly, the core of the first processor will access the memory of the second, adding up to 120 nanoseconds of hardware latency to each operation.
3. Disk subsystem (Storage): IOPS is more important than volume
The database disk subsystem is measured in IOPS (input/output operations per second) for random reads of small 4K blocks. In 2026, standard SATA SSDs , with their 90,000 IOPS limit, became the main bottleneck for enterprise databases.
The modern standard is NVMe U.3 drives with a PCIe 5.0 interface , which connect directly to the processor lanes. They are capable of delivering over 1,500,000 IOPS with latencies of less than 0.05 milliseconds. PCIe 5.0 interface performance specifications are available on the official PCI-SIG portal.
Comparison Table: Database Storage Performance
| Drive type | Bandwidth | IOPS (Random 4K) | Latency |
|---|---|---|---|
| SAS HDD 10K (Deprecated) | Up to 250 MB/s | ~150 – 200 | 5 – 10 ms |
| SATA SSD Enterprise | Up to 550 MB/s | ~90,000 | 1 – 2 ms |
| NVMe PCIe 4.0 | Up to 7,000 MB/s | ~800,000 | < 0.1 ms |
| NVMe PCIe 5.0 (Standard 2026) | Up to 14,000 MB/s | > 1,500,000 | < 0.05 ms |
The disk array architecture requires physical separation of data streams. SQL data files (mdf), transaction logs (ldf), and the temporary database (TempDB) should be located on physically separate RAID arrays (RAID 10 is recommended) to avoid contention for the controller cache.
4. Random Access Memory (RAM): Volume and Channels
The DBMS server’s RAM must accommodate the entire active database volume plus a 20% reserve for operating system needs. If your database is 300 GB, the SQL server must be provided with at least 384 GB of DDR5 RAM with error correction technology ( ECC ).
Memory bandwidth is critical for fast table retrieval. Modern processors support 8- or 12-channel memory operation. This means that to achieve maximum performance, memory modules should be installed symmetrically in the server, filling all available processor channels.
5. DBMS in 2026: PostgreSQL vs. MS SQL
The corporate database market has changed dramatically in 2026: over 70% of new 1C:ERP implementations are based on the PostgreSQL DBMS due to import substitution policies. The choice of database directly dictates hardware requirements.
PostgreSQL consumes more RAM when using aggressive caching (shared_buffers setting) compared to MS SQL . Furthermore, the dead row cleaning process (Autovacuum) in PostgreSQL creates a background load on the disks, making NVMe drives absolutely essential.
Pros & Cons (Pros and Cons of Switching to PostgreSQL for 1C)
Complete absence of license fees (saving millions of rubles), official support by the 1C platform, open source code, and independence from sanctions.
Requires highly qualified DBAs (database administrators) for fine-tuning, high sensitivity to disk subsystem speed, non-standard behavior of the query planner on complex selections.
6. Virtualization vs Bare Metal
Using virtual machines (VMware, Hyper-V, KVM) for databases over 500 GB creates parasitic overhead. The hypervisor layer inevitably consumes 10 to 15% of the processor’s clock speed for internal interrupt routing.
The main problem with database virtualization is micro-delays in the disk subsystem when a request passes through virtual switches.
For systems with over 300 active users, the DBMS server should be installed exclusively on bare metal hardware . Virtualization is only advisable for 1C web publishing servers and test environments.
7. Practical sizing: Configuration examples
Equipment sizing is calculated based on the number of concurrent users and the volume of document flow. Below are standardized configurations for the corporate sector for 2026.
Single-processor server. 8-core / 4.0 GHz processor . 128 GB DDR5 ECC RAM . Two 1.92 TB NVMe SSDs in a software mirror (RAID 1). Combined role (1C Server + DBMS).
Dual-processor server. Two 16-core processors with a base frequency of 3.6 GHz . 512 GB of DDR5 RAM . Hardware RAID controller with 4 GB of cache and an array of 4 NVMe drives in RAID 10 .
Cluster architecture . Three separate 1C application servers (high-frequency CPUs starting at 4.5 GHz). A dedicated, fault-tolerant PostgreSQL DBMS cluster with replication and all-flash storage based on the NVMe-oF (NVMe over Fabrics) interface.
Conclusion
Infrastructure for 1C: Enterprise in 2026 will not tolerate compromises like slow disks or low-frequency processors. Standard general-purpose servers with dozens of weak cores are no longer capable of delivering the required level of transaction performance. Only a transition to high-frequency architectures (starting at 4.0 GHz) and NVMe PCIe 5.0 arrays can guarantee businesses uninterrupted operation without queues, lockups, or freezes.
