Reviewed measurements

Compare the evidence.

Two reports. The same workloads. Benchmark medians and separate network trials in view.

Selected plans

Vultrvoc-c-2c-4gb-50s-amd

Matching observed resources: x86_64 · 2 vCPU · 3.8 GiB RAMSource location: London, UK · same agent build and methodology.

CPU allocation

2 vCPU

Observed memory

3.8 GiB

Processor

AMD EPYC-Milan

Intel Xeon Platinum 8168 @ 2.70GHz

Location

London, UKLHR

London, UKlon1

List price

$40.00 / month$0.055 / hour · before tax

$47.00 / month$0.0699 / hour · before tax

OS / architecture

Ubuntu 24.04.5 LTS / x86_64

Sampling

1 deployment · 24h run skipped

Report date

17 Oct 2026

Methodology

pilot-1 · AI-assisted review

Report

View benchmark report

Each column is one deployment. Matching context does not establish equivalent physical hosts.Benchmark medians · network trials linked separately

PerformanceReviewed data

Vultr2 vCPU · 3.8 GiB

DigitalOcean2 vCPU · 3.8 GiB

01 /

CPU performance

sysbench prime events · max prime 20,000 · gzip level 6 · higher is better

Single thread

1,672events/s84% of 2,000

446.0events/s22% of 2,000

All 2 vCPUs

3,335events/s83% of 4,000

698.6events/s17% of 4,000

gzip compression

242.0MiB/s97% of 250

207.1MiB/s83% of 250

Identical thread counts and test settings. Bars use a common scale within each row.

02 /

Memory bandwidth

sysbench · 256 MiB block · one thread · higher is better

Read

36,666MiB/s92% of 40,000

10,483MiB/s26% of 40,000

Write

25,737MiB/s64% of 40,000

9,336MiB/s23% of 40,000

Bars share one scale across both rows. A memory copy workload is not part of this profile.

PerformanceReviewed data

VultrLondon, UK

DigitalOceanLondon, UK

03 /

Disk performance

fio · direct I/O · 1 MiB / QD32 sequential · 4 KiB / QD32 random

Sequential read

4,444MB/s

2,406MB/s

Sequential write

2,433MB/s

535.7MB/s

Random read

118,094IOPS

90,832IOPS

Random write

99,865IOPS

26,429IOPS

Random read p95

354.3µs

460.8µs

Random write p95

452.6µs

3,719µs

Throughput and IOPS: higher is better. Latency: lower is better. Mixed 70/30 runs and p50/p99 latency are in each full report.

04 /

Network performance

LibreSpeed CLI · HTTPS / IPv4 · three streams · 10 seconds per direction

Frankfurt · source

Network reportSeparate VM test · 17 Oct 2026, 17:39 UTCLondon, UK · Frankfurt, Germany (Clouvider)

Network reportSeparate VM test · 17 Oct 2026, 18:05 UTCLondon, UK · Frankfurt, Germany (Clouvider)

Frankfurt · Download

2,209Mbit/s

648.2Mbit/s

Frankfurt · Upload

962.2Mbit/s

1,056Mbit/s

Frankfurt · HTTP latency

14.4ms

17.1ms

Frankfurt · HTTP jitter

0.13ms

0.05ms

New York · source

Network reportSeparate VM test · 17 Oct 2026, 17:39 UTCLondon, UK · New York, United States (2) (Clouvider)

Network reportSeparate VM test · 17 Oct 2026, 18:05 UTCLondon, UK · New York, United States (2) (Clouvider)

New York · Download

1,202Mbit/s

912.1Mbit/s

New York · Upload

226.3Mbit/s

308.5Mbit/s

New York · HTTP latency

71.9ms

71.6ms

New York · HTTP jitter

0.10ms

0.08ms

Each network source is a linked same-VM supplement when available, otherwise a later test on a separate VM matching provider, plan, recorded region and resources. Frankfurt and New York are separate paths, with one trial per direction. HTTP latency and jitter are recorded; packet loss is not measured. Routes and times affect results; these trials do not establish maximum link capacity or long-term consistency.

PerformanceReviewed data

Vultr2 vCPU · 3.8 GiB

DigitalOcean2 vCPU · 3.8 GiB

05 /

PostgreSQL performance

PostgreSQL 16 · pgbench read/write · 4 clients · durability enabled

In-memory throughput

4,171transactions/s

1,612transactions/s

In-memory average latency

0.96ms

2.48ms

Storage-bound throughput

3,361transactions/s

1,005transactions/s

Storage-bound average latency

1.19ms

3.98ms

Both dataset profiles shown. This profile records average latency, not percentiles; read throughput and latency together.

06 /

Consistency over 24 hours

Separate 24-hour profile · hourly CPU, memory read and random disk read

Hourly repeat runsCPU · memory · disk

Vultr

Not measured24-hour run skipped

DigitalOcean

Not measured24-hour run skipped

No hourly samples exist for these reports, so no chart is drawn.Skipped runs are not scores.

No 24-hour consistency or long-term reliability is established.

07 /

Inspect the source reports.

Full reviewed results and limits for each report.

Vultr report

Report JSONEvidence and limits

DigitalOcean report

Report JSONEvidence and limits

Different workloads reveal different trade-offs. Every number links back to its test.