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

DL360 Gen11 or DL380 Gen11: how to choose between 1U and 2U

Servix technical team · Updated

In brief. The processors, memory and chipset of the DL360 Gen11 and DL380 Gen11 are the same — the only difference is the chassis height. 1U is chosen when data resides on an external storage system and rack space is expensive: up to 10 SFF bays, low-profile cards, 40,000,000 UZS for the base platform. 2U is chosen when local drives (up to 24 SFF), full-size GPUs or spare slots are needed: 49,000,000 UZS for the platform, price of ready-made builds on request. The rule is simple: if at least one growth scenario requires drives or slots, take 2U.

The two most in-demand servers in the HPE ProLiant lineup are the eleventh-generation DL360 and DL380. Their processors, memory and chipset are the same; the only difference is the chassis height: 1U versus 2U. But it is exactly this that determines which server will suit your task and which will become a source of problems after a year of operation.

What they have in common

Both platforms are built on the same foundation: 4th and 5th Gen Intel Xeon Scalable processors, DDR5 memory with eight channels per socket, PCIe Gen5 bus, management via iLO 6, the same families of RAID controllers and network adapters. Therefore, in terms of “bare” computing performance with the same processors there is no difference between them — and hypervisor licenses are calculated the same way, per core.

Where the line is drawn

Criterion

DL360 Gen11 (1U)

DL380 Gen11 (2U)

Rack space

1U — twice as dense

2U

Drive bays

up to 10 SFF

up to 24 SFF or 8 LFF

PCIe expansion slots

limited number, low-profile

more slots, including full-size

Graphics accelerators

compact models only

full-size double-width GPUs

Cooling

denser, fans are louder and run at higher speeds

airflow headroom

High-end processors

TDP limits in a number of configurations

supports hotter models

Platform price in the catalog

40,000,000 UZS

49,000,000 UZS, ready-made builds — price on request

When it is definitely the DL360 Gen11

  • Rack space is expensive and units are few. The classic case for renting space in a commercial data center: twice the computing power in the same rack.
  • Data resides on a storage system, not in the server. If a couple of drives are enough for the OS and logs, and everything else comes over FC or iSCSI, the 24 bays of the DL380 are not needed. How to connect an array is covered in the selections HPE MSA 2060 and 2062 and Dell PowerVault ME5.
  • Virtualization cluster nodes with external storage. A typical scenario: 1U nodes, all capacity on the array.
  • Edge and network services. Domain controllers, DNS, proxies, application servers.

When you need the DL380 Gen11

  • Local drives are the core of the solution. Software-defined storage, hyperconverged clusters, backup servers. Drive and controller selection is in the selection drives and RAID controllers.
  • A GPU is required. Model inference, VDI with graphics acceleration, video processing — a full-size card will not fit in 1U.
  • Many expansion cards. Two FC adapters, two network adapters, a RAID controller — in 1U the slots will run out faster than you expect.
  • Growth is planned. 2U forgives planning mistakes: drives and cards can be added later. In 1U there is no room to expand.
  • The server is not in a data center but in an office server room. A 1U server at high fan speeds is noticeably loud — this is a separate item on the server room checklist.

What it costs in practice

The chassis price difference between 1U and 2U is small compared with the rest of the configuration: 40,000,000 UZS versus 49,000,000 UZS for the base platforms. The main money goes to processors, memory and drives. So the decision is made not by the chassis price but by where the data will reside and how many expansion cards will be needed a year from now. The price of ready-made DL380 Gen11 builds in the catalog (8LFF, 8SFF, 16SFF and 24SFF) is on request; the full list is in the selection HPE ProLiant Gen11, and what makes up the final price of a server is explained here.

A common mistake when choosing

Saving one unit at purchase turns into replacing the server a year later, when it turns out that another controller or local capacity for backups is needed. The rule is simple: if at least one development scenario requires drives or slots, take 2U. The chassis price difference is not comparable with the cost of an unplanned platform replacement.

The opposite mistake also happens: a 2U server with two drives and one network adapter in an expensive rack is space that is paid for and not used.

If you need an even more compact option

If the task is single-processor — a 1C server for a small company, a file server, a monitoring node — look at the DL320 Gen11 for 41,000,000 UZS. It is a 1U single-socket server: the same Gen11 architecture, but noticeably lower cost and power consumption. A 1C calculation by number of users is in the selection server for 1C.

All three platforms are listed in the catalog in new and refurbished versions. Write to us with a description of your task — we will select a configuration and price out both options.

Questions and answers

Does the performance of the DL360 and DL380 differ with the same processors?

Not on typical workloads: the platform, memory and bus are the same. A difference appears under sustained full load on high-end processors, when cooling in 1U runs at its limit while 2U has airflow headroom.

Can a GPU be installed in the DL360 Gen11?

Only compact single-width cards with low TDP. Full-size double-width accelerators go into the DL380 Gen11 — it has the slots, power and cooling for them.

What should I choose for a three-node virtualization cluster?

If capacity is on an external storage system, three DL360 Gen11 servers are denser and cheaper in terms of space. If a hyperconverged cluster is being built on local drives, only the DL380 Gen11 with an SFF drive cage and a controller in HBA mode.

How much memory should go into each node?

Memory almost always runs out before the processor: plan for at least 30% headroom and a reserve for node failure in the cluster. How modules are distributed across channels matters more than their number — see the breakdown in the guide server RAM.

Related reading