Dedicated hosts,
managed by us

Name: Managed dedicated hosts from ENTRONYX CLOUD
Brand: ENTRONYX CLOUD
Availability: InStock

A physical host that belongs exclusively to you — operated by us up to the guest boundary, i.e. the edge where your virtual machines begin. If it fails, a human will respond: in 30 minutes, around the clock, even at night and on public holidays. With booked guest operations, it is 15 minutes.

Configure hosts View response times

All amounts net per month and host, plus 19% VAT., for the reference location Frankfurt; the range goes up to €5,793.18$6,720.09. The host price covers bare metal and platform operations. Licence, datastore and disaster recovery are separate.

Initial response to failure: 30 minutes; included in the host price · 15 minutes from level Platform and guest operations
Availability at night: around the clock; NOC as the first point of contact, every day of the year
Host classes: 5; PC-S32 to PC-X256, from €461.59$535.44
Responsibility: 12; Shifts, each individually assigned to a site

You rent the same machine as under Bare Metal — plus the guarantee that the hypervisor (the virtualisation layer), firmware, fabric (the network between hosts and storage) and storage controllers are maintained by us. The surcharge is stated openly on the invoice because it is a service and not a margin.

Classes: 5; in 4 series
Locations: 4; varies by class
Entry: €461.59$535.44; PC-S32, net per month
Setup: €0.00$0.00; on all classes

PC-S

Entry

One host, local NVMe datastores, no shared storage. For development and acceptance environments where a maintenance window is not an incident.

PC-M

Production

48 Genoa or Granite Rapids cores each, 10 Gbit fabric, no memory overcommit. The fit on which most productive clusters run — optionally AMD or Intel.

PC-L

Density

64 cores in one socket. Halves the licence costs per core in per-socket licence models compared to two smaller hosts.

PC-X

Dual-socket

384 cores and 1 TB RAM per host, connected in the cluster via a 25 Gbit fabric. For clusters that have to calculate per rack unit.

Specifications per class

Processor, core count, clock speed, memory, drives and uplink come from the Bare Metal plan behind the class. They are not maintained twice in this table, but are derived from it — a hardware change takes effect here without any rework.

Specifications per host class. Memory calculated in binary, drive capacity in decimal — just as the drives are specified.
Class	Processor	Cores / threads	Base / boost clock	Memory	Local NVMe	Uplink	Stock
PC-S32Entry point to your own virtualisation — datastores are local to the host.	EPYC 4584PX1 sockets · Zen 4 · 3D V-Cache · 5 nm	16 / 3216 per socket	4.2 GHz5.7 GHz	128 GBDDR5-3600 (On-Die-ECC)	2 × 960 GB NVMe Gen41.92 TB raw	3 Gbit/s	In stock78%
PC-M6448 Genoa cores with 12 memory channels — the most frequently booked production class.	EPYC 9454P1 sockets · Zen 4 Genoa · 5 nm	48 / 9648 per socket	2.75 GHz3.8 GHz	256 GBDDR5 ECC RDIMM	2 × 1.92 TB NVMe (Datacenter Edition)3.84 TB raw	1 Gbit/s	In stock61%
PC-M64iIntel counterpart with AMX — the class for SAP certification and inference on the CPU.	Xeon 6741P1 sockets · Granite Rapids · Intel 3	48 / 9648 per socket	2.5 GHz3.8 GHz	256 GBDDR5 ECC RDIMM	4 × 1.92 TB NVMe (Datacenter Edition)7.68 TB raw	1 Gbit/s	Limited stock44%
PC-L12864 cores in one socket — density without the detour via a second NUMA node.	EPYC 95541 sockets · Zen 4 Genoa · 5 nm	64 / 12864 per socket	3.1 GHz3.75 GHz	512 GBDDR5-4800 ECC RDIMM	2 × 3.84 TB NVMe7.68 TB raw	5 Gbit/s	Limited stock33%
PC-X256Dual EPYC 9965, 1 TB RAM, 25 Gbit fabric. For clusters that do not wait.	2× EPYC 99652 sockets · Zen 5c Turin Dense · 3 nm	384 / 768192 per socket	2.25 GHz3.7 GHz	1 TBDDR5-5600 ECC RDIMM	2 × 1.92 TB NVMe Gen5 (system)3.84 TB raw	5 Gbit/s	Low stock21%

Guarantees and price breakdown

The vCPU limit — the maximum number of virtual processor cores allocated per physical core — is an upper guarantee, not a target value: no placement occurs above this ratio, not even briefly. The monthly price is the sum of the bare metal price for the same configuration and platform operations. Licences for the virtualisation stack are deliberately not included — they are tied to the stack, not the host.

Guarantees per class and the breakdown of the monthly price. All amounts net per month and host for the reference location Frankfurt, plus 19% VAT. Excluding stack licence, excluding datastore.
Class	vCPU per core	Guaranteeable vCPUs	RAM for guests	RAM overcommit	Guests per host	Bare metal	Platform operations	Monthly net
PC-S32FORGE-32	≤ 8 : 1	128	112 GB16 GB hypervisor	25%up to 140 GB	120	€220.16$255.39	+ €149.00$172.84	€461.59$535.44
PC-M64TITAN-96 (Genoa)	≤ 6 : 1	288	232 GB24 GB hypervisor	0%	250	€844.00$979.04	+ €179.00$207.64	€1,023.00$1,186.68
PC-M64iTITAN-96i	≤ 6 : 1	288	232 GB24 GB hypervisor	0%	250	€904.00$1,048.64	+ €189.00$219.24	€1,093.00$1,267.88
PC-L128TITAN-128 (Genoa)	≤ 4 : 1	256	472 GB40 GB hypervisor	0%	400	€660.50$766.18	+ €239.00$277.24	€1,898.65$2,202.43
PC-X256TITAN-768	≤ 4 : 1	1,536	960 GB64 GB hypervisor	0%	640	€2,299.99$2,667.99	+ €389.00$451.24	€5,793.18$6,720.09

Limits per class

Every class has a property that excludes it for certain projects. It is stated here in full and not in the small print, because it decides the selection.

PC-S32: 3 Gbit uplink without a separate storage fabric. Shared NFS or iSCSI datastores are not permitted on this class so that replication and guest traffic do not share the same line. Without a shared datastore, there is no automatic HA restart.
PC-M64: One socket, one NUMA node: VMs over 232 GB RAM no longer fit on this class. For wider single VMs, PC-L128 or PC-X256 is the right fit.
PC-M64i: Live migration is only guaranteed between hosts of the same CPU generation. A mixed cluster of PC-M64 (AMD) and PC-M64i (Intel) does not migrate during operations.
PC-L128: The boost ends at 3.75 GHz — the class wins on width, not on clock speed. Loads that depend on single threads run faster on PC-S32 with 5.7 GHz.
PC-X256: Two sockets mean two NUMA nodes. VMs over 480 GB RAM or over 192 vCPU exceed one node and lose around 12% memory bandwidth — with licences per core, this host also pays for 384 cores.

The boundary follows the guest system — not the contract

Everything below the guest belongs to us: bare metal, firmware, hypervisor, fabric, datastore — the shared storage where the guests reside. Everything inside the guest belongs to you. In between are five layers where both sides act — and this is exactly where it is decided whether an operating model is viable.

Entirely with us

5 layers

ENTRONYX CLOUD

Data centre, power, cooling
Hardware and spare parts
Firmware and BIOS
Hypervisor and cluster services
Storage fabric and datastores

You do not touch these layers, and you do not need on-call duty for them.

Shared

5 layers

Shared

Network segmentation and firewall rules
Capacity planning and placement
Guest operating system
Access management and roles
Backup and recovery test

Both sides act here — in separate places and with separate rights.

Entirely with you

2 layers

Customer

Middleware and application
Application data and classification

We have no access here. Not even on request and not even in the event of a fault.

Which access each side has

Responsibility without access rights is a declaration of intent. The following division is enforced technically, not agreed organisationally: what one side is not entitled to is not unlocked for them.

Only ENTRONYX CLOUD

Below the guest boundary, from a separate management network and fully logged.

Management processor (BMC/IPMI): From a separate management network that is not connected to any customer port. Firmware, boot order and remote console of the bare metal.
Hypervisor shell: Console and SSH of the hypervisor, for maintenance, evacuation and troubleshooting below the guest boundary.
Fabric and storage controllers: Switch configuration, zoning (the mapping of which host sees which storage) and the controllers of the datastores. Usage and latency are visible, not the content of the volumes.
Platform telemetry: Utilisation, temperature, drive error counters, packet loss on the uplink. Basis for alerts and capacity reports.

Only you

From the management tier upwards, regardless of the operations level booked.

Management plane of the stack: vCenter, web console or REST API in your own tenant, with authentication against your directory.
Guest console and guest root: Screen access to every VM and full rights in the guest system, regardless of the booked operations level.
Segmentation in the tenant: VLAN tailoring, rule sets and address allocation within the ranges assigned to you.
Snapshots, clones and templates: Creation and restoration via interface or API, without a change request and without a maintenance window.

Opened 2U server chassis from diagonally above on the workbench: heatsink with copper base over the socket, populated and locked memory banks, fan wall, power supplies with IEC connector at the back, drive bays at the front. — The host price includes exactly 4 layers below the guest: Hypervisor, BIOS and firmware, fabric switches, storage controllers. The firmware level — the version of the control software on the bare metal and controller — is only upgraded during the agreed maintenance window, i.e. in the previously announced period, never spontaneously during operations.

An operating model doesn't fail because of the technology, but because of the line nobody read. That's why the division is shown here in full and side by side with the two neighbouring models. In 2 out of 12 layers, the Private Cloud differs from the Public Cloud, in 8 out of 12 from Bare Metal.

ENTRONYX CLOUD
We operate the layer and are liable for its outcome.
Shared
Both parties act in the same layer in separate areas.
Customer
You operate the layer; we have no access there.

Division of responsibility per layer, compared across the three product lines. “Shared” means: both sides act in the same layer in separate places — what that means for each layer is explained below the table.
Layer	Public Cloud	Private Cloud	Bare Metal
Data centre, power, cooling	ENTRONYX CLOUD	ENTRONYX CLOUD	ENTRONYX CLOUD
Hardware and spare parts	ENTRONYX CLOUD	ENTRONYX CLOUD	ENTRONYX CLOUD
Firmware and BIOS	ENTRONYX CLOUD	ENTRONYX CLOUD	Customer
Hypervisor and cluster services	ENTRONYX CLOUD	ENTRONYX CLOUD	Customer
Storage fabric and datastores	ENTRONYX CLOUD	ENTRONYX CLOUD	Customer
Network segmentation and firewall rules	Shared	Shared	Customer
Capacity planning and placement	ENTRONYX CLOUD	Shared	Customer
Guest operating system	Customer	Shared	Customer
Middleware and application	Customer	Customer	Customer
Application data and classification	Customer	Customer	Customer
Access management and roles	Shared	Shared	Customer
Backup and recovery test	Shared	Shared	Customer

The neighbouring models are for comparison, not as an offer on this page: Public Cloud takes capacity planning off your hands and gives you sole control of the guest system; Bare Metal gives you everything from the firmware upwards and takes nothing off your hands.

The 5 shared layers, spelt out

“Shared” is the only entry in the matrix that decides nothing on its own. Each of these layers has an edge — the sentence that decides who acts in the event of a fault.

Network segmentation and firewall rules

We: We operate the fabric, the uplinks and the edge filters of the location and separate your VLANs — the virtual network segments — from those of other tenants.
You: You tailor the segments within your tenant and maintain the rule sets between them.

We do not change any rules in your tenant, not even in the event of a fault. A rule that resolves an outage comes as a change request from you — otherwise no one would know why it is there afterwards.

Capacity planning and placement

We: We provision the capacity of the location, place guests according to the agreed rules and report before a class runs short.
You: You decide on the number and size of the guests and when a host is added.

The vCPU limit per class is a commitment, not a target value. We do not place above it — not even briefly and not even if a guest is waiting as a result.

Guest operating system

We: On our images, we apply kernel and security patches in the maintenance window, provided an operations level with guest operations is booked.
You: You are responsible for custom-built images, installed agents and any configuration above the package management.

A custom-built image is excluded from the availability commitment. This is not a penalty, but a consequence of the fact that we do not know its patch level.

Access management and roles

We: We manage the roles of the management plane, connect your directory and log every administrative access on our side.
You: You assign the roles within your tenant and revoke them when people leave.

We do not create or delete any accounts in your directory. A departed administrator retains their access until you revoke it.

Backup and recovery test

We: We run the backups, monitor their results and provide the environment for the recovery test.
You: You identify the systems requiring protection, approve the recovery point and sign off on the test.

A successful backup is no evidence of a successful recovery. The two only come together in the test, and the test requires your sign-off.

Who patches which layer — and when

Platform operations included in the host price end at the guest boundary. Everything above that is a separate level with its own price. The matrix shows each layer against each level; a tick means that we keep that layer current and answer for it.

Patch responsibility per layer and operations tier. Tick: maintained by ENTRONYX CLOUD. Minus: your responsibility.
Layer	Platform operationsin the host price	Platform and guest operationsfrom €729.00$845.64	Full operationsfrom €2,579.00$2,991.64
Hypervisor
BIOS and firmware
Fabric switches
Storage controllers
Guest operating system		only ENTRONYX images
Databases and middleware in the operations manual

Platform operations

Remains with you

Guest operating system
Middleware
Application
Application data

Change lead time 48 hours

Platform and guest operations

Remains with you

Middleware
Application
Application data

Change lead time 24 hours

Full operations

Remains with you

In-house developed applications
Application data

Change lead time 8 hours

When patching happens

The maintenance window depends on the virtualisation stack, not the bare metal: a host is patched when the cluster it resides in has its window. Firmware and hypervisor work is done on a rolling basis, host by host. That is why there is a notice period in each row. And that is why a cluster with only one host requires planned downtime instead of an evacuation — moving running guests to another host.

Maintenance window per stack. All times in CET or CEST, announced by email to the registered technical contacts and in the customer area.
Stack	Maintenance window and announcement	Host classes	Hosts per cluster
ENTRONYX KVM PlatformOpen hypervisor on libvirt, licensed per socket instead of per core.	Tue and Thu 02:00–05:00 CET, announcement 10 calendar days in advance	5	1 – 16
ENTRONYX Managed vSpherevCenter-managed cluster with HA, DRS and vMotion — licence per core.	Wed 02:00–06:00 CET, rolling host by host, announcement 14 days in advance	4	2 – 32
ENTRONYX HCIHyperconverged: computing power and storage in the same host, licence per core.	Tue 02:00–06:00 CET, rolling with data redistribution before each restart	4	3 – 24
ENTRONYX SAP CertifiedCertified HANA nodes, licensed per host — without overcommitment.	By arrangement per environment, 21 days lead time, outside your closing periods	4	1 – 8

Two hands over a rack blanking panel: one holds a server blade, the other grasps it — the blade visibly changes hands.

Operations scale with the number of systems, not their size

The lowest level is already included in the host price — for PC-S32, that is the €149.00$172.84 surcharge on the bare metal. If you move operations further up, you pay a base amount per environment — a flat rate, regardless of the number of hosts — and a smaller rate per host.

Platform operations

Included

Included in every managed host. We operate everything below the guest.

in the host price

Initial response P1: 30 minutes
Initial response P2: 4 hours
Change lead time: 48 hours
Coverage: P1 around the clock, P2 and P3 Mon–Fri 08:00–18:00

Limit: Firmware maintenance requires a host evacuation. Without a second host in the cluster, this means downtime of up to 45 minutes per maintenance window.

Platform and guest operations

Plus guest patches, backup verification and alert rules per system.

€690.00$800.40Base per environment, plus €39.00$45.24 per host and month

Initial response P1: 15 minutes
Initial response P2: 2 hours
Change lead time: 24 hours
Coverage: P1 and P2 around the clock, P3 Mon–Sat 08:00–20:00

Limit: Guest patches are only applied to images delivered by ENTRONYX. Custom-built images remain your responsibility and are excluded from the availability commitment.

Full operations

Operations up to the application, with change management and capacity planning.

€2,490.00$2,888.40Base per environment, plus €89.00$103.24 per host and month

Initial response P1: 15 minutes
Initial response P2: 1 hour
Change lead time: 8 hours
Coverage: Around the clock via fixed phone number, named technical contact

Limit: Application operations include the services listed in the operations manual. In-house developments not listed are monitored but not modified.

Operations costs by host count

The base amount applies once per environment, the host rate for each system within it. With a single host, the base amount is the predominant part of the total; from about eight hosts, the ratio flips. The amounts are in addition to the host price, not included in it.

Monthly operations costs per level, net plus 19% VAT., excluding hosts, licences and datastores. Calculated as base per environment plus rate per host.
Tier	1 Host	3 Hosts	8 Hosts	16 Hosts
Platform operationsincluded in the host price	€0.00$0.00	€0.00$0.00	€0.00$0.00	€0.00$0.00
Platform and guest operations€690.00$800.40 + €39.00$45.24 per host	€729.00$845.64	€807.00$936.12	€1,002.00$1,162.32	€1,314.00$1,524.24
Full operations€2,490.00$2,888.40 + €89.00$103.24 per host	€2,579.00$2,991.64	€2,757.00$3,198.12	€3,202.00$3,714.32	€3,914.00$4,540.24

Who you reach when something is down

Every operations level has its own escalation path and initial response time. The fastest commitment in the catalogue is 15 minutes for an outage — around the clock, every day of the year.

Outage

A productive service is unreachable or a layer in our responsibility has failed: host, fabric, datastore. An impending data loss also counts here, regardless of whether something is already down.

Significant impairment

The service is running, but restricted: a host dropped from the cluster, redundancy lost, latency permanently above the commitment. A second failure would lead to P1.

Request and change

Everything without impact on operations: change requests, capacity questions, information from the logs. For changes, the lead time of the operations level applies, not the initial response time.

Night-staffed control room: one person in front of three screens with alarm lists, in the background the glass wall to the server hall. — At night, the same team takes the report as at noon: NOC. The initial response counts from the receipt of the report, not from the start of processing — at ENTRONYX CLOUD, this is 30 minutes in the included level and 15 minutes as soon as guest operations are booked.

Initial response times per operations level. Measured from receipt of the report via the channel agreed for the level.
Tier	Initial response P1	Initial response P2	Coverage	Change lead time
Platform operations	30 minutes	4 hours	P1 around the clock, P2 and P3 Mon–Fri 08:00–18:00	48 hours
Platform and guest operations	15 minutes	2 hours	P1 and P2 around the clock, P3 Mon–Sat 08:00–20:00	24 hours
Full operations	15 minutes	1 hour	Around the clock via fixed phone number, named technical contact	8 hours

The way up

Escalation here means: the next level is actively brought in, not the report passed on. Whoever has reached a level does not lose the previous one. With an increasing operations level, the path becomes longer because it ends further up — not because it gets slower.