The core count stopped meaning anything
A VPS plan is advertised by cores and RAM, but the number that decides throughput is the silicon underneath. AMD's EPYC 9005 series (Turin) and 9004 series (Genoa) differ in IPC, memory generation, and instruction support. An eight-core plan from 2019 and an eight-core plan from 2026 share a description and very little else.
European providers have started pricing that difference explicitly. Here is what each one publicly commits to, and where the gaps are.
Hetzner: generation is now a product tier
Hetzner is the most transparent about this, even though it does not name CPU models per plan. Its own documentation splits the cloud line into three tiers and states the hardware generation directly:
- Cost-optimised (CX, CAX): "older hardware gen." This is the €5.49 to €29.49 range.
- Regular performance (CPX): "most recent hardware gen."
- Dedicated (CCX): "most recent hardware gen," with dedicated cores.
That is unusual honesty. The practical consequence is visible in the pricing: after the June 2026 adjustment, CX33 (4 vCPU, 8 GB) is €8.49 while CPX32 (4 vCPU, 8 GB) is €35.49. You are paying roughly four times as much for the same nominal specification on current hardware.
If your workload is latency-sensitive, that premium is the product. If it is a blog or a batch job, the older generation is the better buy.
Netcup: newest silicon on root servers, previous generation on VPS
Netcup's Generation 12.5 refresh splits the same way, one step further along.
The Root Server (RS) G12.5 line runs on AMD EPYC 9645 processors from the 9005 series, with dedicated core pinning. The VPS G12.5 line, by contrast, is widely reported to run on EPYC 9004-series hardware; Netcup does not publish a fixed CPU model for the VPS tier, so treat that as an inference from delivery data rather than a contract term.
The gap matters if you are choosing between a €14.50 VPS and a root server at a similar price: the root server is where the current generation lives. Both include DDR5 ECC memory and NVMe in RAID 10, so the memory and storage subsystems are current even where the CPU is one generation back.
UpCloud: current AMD, stated plainly
UpCloud describes its cloud servers as running on premium AMD CPUs, and its Private Cloud product specifies AMD EPYC 9575F (9005 series, Zen 5) with 64 cores and 128 threads per node. The Starter tier is explicitly labelled as using previous-generation AMD CPUs, which is how it reaches €3/month.
That is a clean split: pay €3 and get older cores, pay standard rates and get current ones. The 99.99% SLA applies to both.
Contabo: the clearest old-versus-new split
Contabo operates two product lines that differ mainly by hardware generation:
- Core VPS: described as proven, older hardware for lighter workloads, from €4.40/month on a 24-month term.
- Performance VPS: current AMD EPYC CPUs with NVMe, aimed at compute-heavy workloads.
Contabo's own FAQ says that paying for the latest CPU generation is unnecessary for workloads that run fine on proven hardware, which is an unusually direct statement of the trade-off. If you need current silicon from Contabo, the Performance line is the only route.
OVHcloud: current infrastructure, unpublished CPU
OVH's VPS 2027 range is new hardware with NVMe storage and daily backups, but the public VPS page does not name a CPU model or generation. OVH's dedicated and public cloud ranges do publish EPYC generations, and its recent server launches emphasise high-clock EPYC parts, but that does not transfer to the VPS tier automatically.
The honest position: OVH sells current-generation infrastructure and does not make the VPS CPU model part of the offer. If the CPU generation is a hard requirement, ask support before ordering, or benchmark after.
IONOS and Scaleway
IONOS states that it uses "the latest AMD and Intel CPUs" alongside Samsung NVMe and Dell servers, without tying specific generations to specific plans. Scaleway states that its instances run on AMD EPYC for x86 and offers ARM as an alternative, with dedicated or shared vCPU options per family.
Both are current-platform providers without a per-plan CPU commitment. That is normal for the industry and the reason the next section exists.
How to check what you actually got
You do not have to trust the marketing. On any Linux VPS:
lscpushows the model name, base and boost clock, cache sizes, and flags such asavx512f.cat /proc/cpuinfo | grep "model name"gives the marketing string for the host CPU.dmidecode -t processoron bare metal, or the hypervisor's own metadata, confirms the platform.- A short
sysbench cpuoropenssl speedrun gives you a single-core number you can compare across candidates.
Two cautions. First, virtualised guests often report a generic model string, so the absence of a model name is not proof of old hardware. Second, the same plan can land on different hosts over time, so benchmark the instance you have, not the plan name.
Does the newest CPU matter for you
It matters when:
- You run latency-sensitive request serving where single-core IPC dominates.
- You use AVX-512 for numerical work, which the 9004 and 9005 series support.
- You are memory-bandwidth bound and DDR5 helps.
- You run databases where per-core performance sets the ceiling.
It matters much less when:
- The workload is I/O bound and NVMe is the bottleneck.
- You are serving static content through a CDN.
- The process is idle most of the day.
- You would rather spend the difference on more RAM.
The practical ranking
If "newest CPU" is the deciding criterion: Netcup's root servers and UpCloud's standard and private tiers are the clearest current-generation offers, with Hetzner's CPX and CCX lines close behind and explicitly labelled as current. Contabo's Performance line and Scaleway's premium families are next. Hetzner's CX and CAX lines, Contabo's Core line, and UpCloud's Starter tier are deliberately older hardware sold cheaply, which is a legitimate choice rather than a defect.
Prices are October 2026 list prices, excluding VAT. CPU policy changes without notice; verify before committing to a long term.




No comments yet