Skip to content

Dear customers! Please send orders and inquiries to [email protected] or to Telegram @servix_uz instead of calling by phone: due to the high workload, the operator cannot answer all incoming calls.

Servix

Virtualization server in Tashkent: sizing by number of machines, licenses and ready-made configurations

Updated

In brief. A virtualization server is sized by the number of virtual machines and their total vCPU and memory, not “by power”. For 3–⁠4 machines at the level of a domain controller and file server, one four-core Xeon E and 32 GB are enough — Dell R350 for 20,886,000 UZS. For 8–⁠12 machines with 1C and a DBMS you need two processors, 128 GB and SSDs in RAID 10 — HPE DL380 Gen10 for 45,761,000 UZS. From 20 machines, two nodes with shared storage are used: a pair of Dell R650xs with 2×Gold 5320 and PowerVault ME5012 iSCSI 25GbE. A ready-made kit for 3–⁠4 machines is 46.5 million UZS, for 10–⁠12 — 104.0 million UZS, a cluster for 20–⁠30 machines — price on request. For VMware a processor should have at least sixteen cores: the license is counted by cores with a minimum of 16 per socket.

Request a virtualization server calculation → — send us a list of machines with vCPU and memory and the hypervisor (VMware, Proxmox, Hyper-V), and an engineer will select nodes and drives and calculate licenses for three years.

Quick links: four figures for the calculation · matrix: how many machines on which server · processor and licenses · memory · drives: local or storage system · network · a two- or three-node cluster · ready-made kits

Four figures the calculation starts with

A server for a hypervisor is selected by the sum of the virtual machines' needs. Write down each machine with four values: vCPU, memory, disk size and the nature of the load. Then it is arithmetic.

Typical machine

vCPU

RAM

Drive

Disk load

Domain controller, DNS, DHCP

2

4–⁠8 GB

80 GB

low

File server

2

8 GB

by data volume

sequential

1C application server

4–⁠8

16–⁠32 GB

120 GB

medium, sensitive to clock speed

DBMS (MS SQL, PostgreSQL)

4–⁠8

32–⁠64 GB

by database + logs

high, random writes

Terminal server for 20 users

8

32–⁠48 GB

200 GB

medium

Web, CRM, mail, Zabbix — each

2–⁠4

8 GB

100 GB

low

Backup server

4

16 GB

separate volume

sequential, at night

  • Cores. vCPUs are overcommitted onto physical cores: for office services 3–⁠4 vCPU per core, for 1C and DBMS 1–⁠2 per core, because what matters to them is not the total core count but clock speed and the absence of a queue. Twenty-four vCPUs of office machines fit on 8 cores, eight vCPUs of DBMS and 1C on 6–⁠8 cores.
  • Memory. Not overcommitted. Add up the RAM of all machines and add 10–⁠15% for the hypervisor itself. Eight machines with 8–⁠16 GB each is 96 GB, which means a 128 GB server.
  • Drives. Sized in operations per second, not terabytes. Ten machines with a DBMS and 1C generate thousands of random write operations — that means SSDs in RAID 10; hard drives remain for archives, files and backups.
  • Headroom. Add 30% to everything: in a year there will be more machines. And check that the core count is consistent with how the hypervisor is licensed — more on this below.

Matrix: how many machines on which server

Guidelines from the catalog as of November 2026. The number of machines is for a typical office mix with one or two heavy ones (1C, DBMS). Prices are per server in the specified configuration, without drives unless stated otherwise.

Machines

Total vCPU / RAM

Cores in the server

Platform from the catalog

Price

3–⁠4

8–⁠10 / 24–⁠32 GB

4

Dell R350 8SFF, Xeon E-2314, 32 GB

20,886,000 UZS

5–⁠7

16 / 48–⁠64 GB

8

HPE DL360 Gen10, Xeon Silver 4208, 32 GB or the tower HPE ML350 Gen10

25,043,000 UZS / 34,912,000 UZS

8–⁠12

32–⁠40 / 96–⁠128 GB

16–⁠20

HPE DL380 Gen10, 2×Silver 4210, 64 GB, DL360 Gen10, 2×Silver 4208, 64 GB, DL380 Gen10, 2×Silver 4215R, 128 GB

price on request / price on request / 45,761,000 UZS

12–⁠18, a lot of memory

40–⁠50 / 192–⁠256 GB

14–⁠24

HPE DL380 Gen10, Gold 5120, 256 GB, Dell R750xs 12LFF, 2×Silver 4310, 64 GB

54,457,000 UZS / 55,284,000 UZS

20–⁠30, cluster node

80–⁠100 / 128–⁠256 GB

40–⁠52

Dell R650xs, 2×Gold 5320 with 26 cores each, 128 GB, Dell R650, 2×Silver 4316 with 20 cores each, 128 GB

71,163,000 UZS / 86,673,000 UZS

New generation, DDR5

made to order

from 24

Dell R760 8SFF, 2×Silver 4410Y, Dell R660 10SFF, HPE DL380 Gen11 8SFF, Lenovo SR650 V3

price on request / price on request / price on request / 60,000,000 UZS

Gen11 platforms and 16th-generation PowerEdge are listed in the catalog in a base configuration with 8–⁠12-core processors and 32 GB: this is a starting point, and processors and memory are changed to suit the task when ordering. Full lines with prices are in the selections HPE ProLiant Gen11 and Dell PowerEdge.

Processor: cores and licenses are calculated together

In 2026, choosing a processor for virtualization is first a question of licensing and only then of performance.

  • VMware is sold as a per-core subscription, and each processor is charged a minimum of 16 cores. A server with two eight-core Xeon Silver 4215R is licensed as 32 cores with 16 physical ones. The guideline for vSphere Standard 8 is 50–⁠55 USD per core per year, for vSphere Foundation 135–⁠155. The conclusion: for VMware, processors with 16 or more cores are used, and one dense processor is better than two small ones.
  • Proxmox VE is licensed per socket and only for support: Community 120 EUR, Basic 370 EUR per socket per year; the features are the same. The core count does not affect the price, so two inexpensive Xeon Silver processors give a cluster with HA without a penalty.
  • Hyper-V comes as part of Windows Server: Datacenter is licensed for all cores in packs of 16, 6,771 USD per pack, perpetual, and gives an unlimited number of Windows Server guests on the host. If the guests are Windows (1C, SQL Server, domain controller), Datacenter is bought anyway, and the hypervisor comes for free.

Configuration

Physical cores

Cores as counted by VMware

vSphere Standard, USD per year

Proxmox Basic, EUR per year

Windows Server Datacenter, USD one-time

One node 2×Silver 4215R (kit 2)

16

32

1 600–⁠1 760

740

6,771

Two nodes 2×Gold 5320 (kit 3)

104

104

5 200–⁠5 720

1,480

54,168

Software prices are list-price guidelines without discounts and CALs; a full breakdown with VVF, VCF and the vSphere 8 support term is in the article Proxmox or VMware in 2026. Clock speed matters more than core count for 1C and DBMS: a Silver 4215R with a 3.2 GHz base clock outperforms a Silver 4210 at 2.2 GHz for 1C at the same server price. A comparison of generations and lines is in the selection Xeon and EPYC server processors and in the article Xeon 6 or EPYC 9005 for virtualization.

Memory: no overcommitment and across all channels

Memory is the most common host limit: cores sit idle, but no more machines can be started. Three rules.

  • The sum of RAM for all machines plus 10–⁠15% for the hypervisor and cache. Memory consolidation technologies (ballooning, compression) are reserved for emergencies, not for planned savings.
  • Modules are installed across all processor channels: second-generation Xeon Scalable has six channels per socket, fourth-generation has eight. Two processors and two 32 GB modules give 64 GB on one channel out of six — bandwidth several times lower than with 6×16 GB. Catalog configurations such as “2×32 GB” are better rebuilt for the channel count when ordering.
  • A single machine should not be larger than one processor's memory: NUMA crossings slow a DBMS down by tens of percent. For a 128 GB DBMS, take a node with 2×128 GB or one processor with 256 GB, such as the DL380 Gen10 with Gold 5120 and 256 GB.

Module types, ranks and what is compatible with what are covered in the guide RAM for servers.

Drives: local SSDs or shared storage

For one host, local drives are enough. For a cluster with VM migration between nodes, you need shared storage — either a storage system or distributed storage on the nodes' drives.

  • One host. Enterprise SSDs in RAID 10 on a controller with cache: four Intel D3-S4520 7.68 TB at 10,782,352 UZS each give about 15 TB usable with a load of tens of thousands of operations. The system goes on a boot module — Dell BOSS for 7,907,058 UZS or HPE NS204i-p for 10,063,529 UZS, so as not to waste drive cage bays. Controllers with cache and battery: HPE P408i-a — 7,188,235 UZS, Dell PERC H755 — 8,625,882 UZS.
  • Two to four nodes, minimum budget. A storage system with direct SAS attachment: HPE MSA 2060 SAS 24SFF without drives (price on request) connects to four nodes directly, without SAN switches. This is the cheapest way to get vMotion and HA.
  • A cluster with growth. iSCSI at 25 Gbps — Dell PowerVault ME5012 iSCSI 25GbE (price on request) — or Fibre Channel: HPE MSA 2062 FC 24SFF for 126,800,000 UZS, HPE MSA 2060 FC 12LFF for 88,277,000 UZS, with a QLogic 16G HBA at 10,782,352 UZS in each node. A breakdown of models and shelves is in the selections HPE MSA and PowerVault ME5.
  • Ceph or vSAN. Three nodes or more, NVMe in each, an HBA in direct access mode instead of hardware RAID, a separate 25 Gbps network for replication. Cheaper than a storage system in hardware on a large cluster, but with two or three nodes it loses on both price and simplicity.

Which RAID level and why not RAID 5 on large drives — in the guide RAID 5, 6 or 10; drive classes — in the selection drives and RAID controllers.

Network: 10 Gbps minimum

A gigabit port for ten machines becomes a bottleneck before the processor does: backup, migration and file traffic share one link. The norm for a virtualization host:

  • Two 10 Gbps ports for machines and management: HPE 530SFP+ or Intel X520-DA2 — 4,097,294 UZS each; connection to the switch with a DAC SFP+ 1 m cable for 249,000 UZS without modules.
  • A separate pair of ports for storage and migration if the storage system uses iSCSI: 25 Gbps — Dell 57414 SFP28 OCP3 for 5,031,764 UZS, HPE 640SFP28 for 5,750,588 UZS, cables DAC SFP28 1 m at 399,000 UZS each.
  • Three networks in separate VLANs: management, machines, storage. A single physical port for everything means losing host management when machine traffic overloads it.

Compatibility of ports, modules and cables is covered in the selection network cards, transceivers and DAC; switches with 10G ports for the server room — in the selection Aruba Instant On.

Cluster: two nodes or three

One server is a single point of failure for all machines at once. As soon as the host runs 1C, a DBMS and mail, the question of a second node becomes a question of downtime.

  • Two nodes and shared storage — the minimal setup with HA: if a node fails, the machines restart on the second one within minutes. Each node must be able to carry the entire load alone, so resources are sized with 50% headroom. Proxmox with two nodes requires a third quorum vote — it can be any small machine; vSphere handles quorum through vCenter and the storage system.
  • Three nodes — the standard for Ceph and vSAN and for workloads that need not only availability but also room for maintenance: one node is taken out for maintenance, two carry the load.
  • One node and backups — an honest option for 3–⁠6 machines: downtime of a few hours is acceptable, and restoring from a backup to a new server is a well-practiced procedure. A second node here costs more than the business loses in an hour of downtime.

Ready-made kits with prices

The amounts do not include hypervisor and guest OS licenses, a UPS or switches. UPS sizing by node power consumption — UPS calculator, usable array capacity — RAID calculator.

What to clarify before ordering

  • A list of machines with vCPU, memory and disk, noting separately which of them are 1C, DBMS and terminal servers.
  • Hypervisor: VMware, Proxmox or Hyper-V. This determines the number of cores per processor and whether a Windows Server Datacenter license is needed.
  • Acceptable downtime: hours (one node and backups) or minutes (two nodes and a storage system).
  • Whether you already have a storage system, 10G switches and backups — the kit is built around them.
  • Growth over two to three years: processors and memory are cheaper to plan in from the start than to replace the drive cage and platform later.

Send us the list of machines — we will calculate the nodes, drives and licenses for three years and send a quote including VAT: contacts. Platforms are in the section Servers, ready-made builds with prices are in the same section.

Sources

Other Servix collections

Installation and support — KP Solutions

Installation, setup and maintenance of equipment — from our integrator partner.

kpsolutions.uz