VPS on Intel Xeon: how to tell a recent processor from a ten-year-old chip
The word "Xeon" in a plan description only means that the processor is a server one and most likely works with ECC memory. Intel has been releasing chips under this brand since 1998, and in one catalogue you will find processors separated by more than ten years of development and a two- to threefold difference in core speed. So a Xeon plan should be judged by generation, not by brand.
Which Xeons you will find in VPS plans
- Xeon E5-2600 v2, v3, v4 (2013–2016). The most common processors in budget plans: servers built on them are decommissioned en masse by large data centres and sold on the second-hand market for little money. v2 (Ivy Bridge) works with DDR3 memory and does not support AVX2; v3 (Haswell) and v4 (Broadwell) moved to DDR4 and gained AVX2. Typical models are the E5-2680 v2, E5-2690 v4 and E5-2697 v4.
- Xeon Scalable (since 2017): Bronze, Silver, Gold and Platinum across five generations, from Skylake to Emerald Rapids. This is the backbone of modern clouds; a detailed breakdown is on the VPS on Xeon Gold and Platinum page.
- Xeon 6 (since 2024): a generation without the "metal" tiers, with variants built on performance cores and on efficient cores.
- Xeon E3 and Xeon E-2xxx: single-socket chips based on desktop architectures, with few cores and high clock speeds. They go into budget dedicated servers and plans that focus on single-threaded speed.
Why the Xeon in a plan may be old and what that means
A dual-socket server on E5 v2 with hundreds of gigabytes of DDR3 now costs less than a single modern processor, while the virtual cores on it are sold the same way as on new hardware. That is profitable for the host and not always bad for the customer: for a business-card website, a corporate VPN for a small team or a backup store, the speed is enough. But the differences are real and noticeable:
- Core speed. Per clock cycle, an E5 v2 core does noticeably less work than a Xeon core of the 2023–2024 generations; at a similar clock speed the single-threaded difference reaches twofold.
- Instructions. Without AVX2, some modern builds of programs won't start or run in a slow mode: this affects neural networks on the CPU, some databases and compression libraries.
- Memory. DDR3 with four channels per processor versus DDR5 with eight channels on recent Xeons is a difference in bandwidth of several times.
- Vulnerability mitigations. Old chips close Spectre and Meltdown with software patches that take away additional performance on system calls and I/O.
If single-threaded speed matters, compare plans with a high CPU clock speed; if you need many cores and memory bandwidth, look at the alternative too: VPS on AMD EPYC.
How to find out the real processor after buying
lscpu | grep -E 'Model name|Flags'orgrep -m1 'model name' /proc/cpuinfo. A line like "Intel(R) Xeon(R) CPU E5-2680 v2 @ 2.80GHz" names the model and generation unambiguously.- If instead of a model it says "QEMU Virtual CPU", "Common KVM processor" or "Intel Xeon Processor (Skylake, IBRS)", the host shows the guest a generic model. The last variant names an instruction set, not the real chip.
- Check the instructions:
grep -o -w 'avx2\|avx512f' /proc/cpuinfo | sort -u. No avx2 almost certainly means the E5 v2 generation or older, or the host has trimmed the set of flags. - Run a short single-threaded test,
sysbench cpu --threads=1 --time=30 runoropenssl speed -seconds 5 sha256, and compare the result with published figures for the stated model.
How to read a Xeon model number
On the E5 the generation is shown by the suffix: v2, v3, v4; the first digit after the hyphen is the maximum number of processors in a server (E5-2xxx means up to two). On Scalable the first digit of the number indicates the tier (3 is Bronze, 4 Silver, 5 and 6 Gold, 8 Platinum), and the second the generation: the Gold 6148 dates from 2017, the Gold 6348 from 2021. If a host simply writes "Intel Xeon" without a model, ask support for it before paying: an honest answer takes one line.
The letters at the end of the number carry meaning too. On the E5 the letter L means an energy-efficient version with a lower clock speed, while on Scalable the suffixes indicate special variants, such as the refreshed R series or the single-socket U. The clock speed in a name like "@ 2.20GHz" is the base one: comparing plans by it is pointless, because turbo frequencies differ between generations more than base frequencies do. It is more useful to compare processors by the single-threaded result of the same test run on both machines than by the gigahertz in the plan description.
Finally, old dual-socket E5 nodes usually have less total memory speed per core, and more customers live in one chassis. If your task is sensitive to latency, also watch steal time in top during the first days of the rental: a steady 10% or more is more common on old hardware than on new.