On 12 September 2026, we measured two new VPS instances with matching resources: one from the AMD EPYC/Ryzen group and one from Intel E5. In the single-thread CPU workload described here, AMD achieved a median of 1,577.07 events/s, compared with 314.17 events/s on Intel. The ratio between the medians was 5.02x.
This result applies to these two instances and this test. It is not a performance guarantee for every tier, processor or application.
What we compared
Both VPS instances had 1 vCPU, 1 GB RAM, 10 GB disk and a public IPv4 address. We created them from Ubuntu 24 and let the initial updates finish before testing. After a reboot, both ran Ubuntu 24.04.5 LTS and kernel 6.8.0-139-generic.
- AMD: physical AMD Ryzen 9 7900 12-core processor. The guest reported
AMD EPYC Processor, with theEPYCvirtual CPU profile. - Intel E5: physical Intel Xeon E5-2673 v3 at 2.40 GHz. The guest reported
QEMU Virtual CPU version 2.5+, with thex86-64-v3virtual CPU profile.
The processor name shown inside a virtual machine may differ from the physical CPU. We kept each tier's commercial configuration, without changing host affinity, mitigations or hypervisor settings to favor the test.
On the test date, this configuration cost EUR 6/month on AMD and EUR 5/month on Intel E5, including IPv4. The Manager quoted hourly totals of EUR 0.0083 and EUR 0.0069, respectively. The monthly figure is a 720-hour reference; billing covers the hours the VPS exists.
Reproducible method
We installed sysbench from Ubuntu's signed repository: version 1.0.20, package 1.0.20+ds-6build2. We verified that the executable, loaded libraries and their versions matched. Neither VPS was running a website, an application database or a Matrix intervention.
Each instance completed five 20-second repetitions, using one thread and a prime limit of 20000, with ten seconds of rest between repetitions. The two series ran in parallel on separate physical hosts during the same time window. There was no separate warmup.
sysbench --threads=1 --time=20 --events=0 \
--report-interval=5 --cpu-max-prime=20000 cpu runAn event is one execution of sysbench's prime-number workload. Events/s describes this specific calculation. The official CPU test source shows the operations it performs.
Before measurement, one-second samples showed 99-100% idle CPU on both guests. We retained CPU counters, load, resources and every result. The steal counter records involuntary CPU waiting time; see the Linux documentation for /proc/stat.
All five results
Rates in events/s, in execution order:
- AMD: 1,577.83; 1,585.93; 1,573.94; 1,577.07; 1,568.69.
- Intel E5: 311.86; 314.29; 315.52; 314.17; 310.70.
AMD's median was 1,577.07, with a range of 1,568.69-1,585.93 events/s. Intel E5's median was 314.17, with a range of 310.70-315.52 events/s.
Steal per repetition, in the same order:
- AMD: 0.000%; 0.000%; 0.050%; 0.050%; 0.050%.
- Intel E5: 0.000%; 0.000%; 0.000%; 0.000%; 0.000%.
We publish all five measurements from each VPS instead of selecting the fastest attempt. Dates, method and figures are available in the public benchmark JSON.
Measured performance and price
Dividing each median by the monthly price of the tested configuration gives 262.85 events/s per EUR/month on AMD and 62.83 on Intel E5. The ratio between those two values is 4.18x.
This compares the measured CPU workload with its price. It does not measure business profitability or savings for every application. AMD costs one euro more per month at this size; choosing resources also calls for testing the workload that the VPS will actually run.
Test scope
We measured one VPS per group on one date. Five repetitions describe variation within those instances; they are not five independent servers. Host placement, shared load and virtual CPU configuration can affect future results.
This test does not measure disk, network, memory, WordPress performance, remote desktop responsiveness or Matrix quality. It does not establish the same ratio for every EPYC/Ryzen model or every Intel E5 processor.
For available resources, see the VPS configurator. Windows use is documented separately in the remote desktop guide. For the AI administration workflow, see the WordPress installation case with Matrix.