Hetzner CAX21

Falkenstein DC Park 1

Benchmark complete

Finished Oct 15, 2026

Geekbench 7 single-core

963

Faster than 33% of 4 vCPU plans

Geekbench 7 multi-core

3,424

Faster than 44% of 4 vCPU plans

Disk 4K IOPS

58,137

fio random read and write, 50/50

Monthly price at test time

$12.49

274 multi-core points per dollar

Same price today

Server

Hetzner CAX21

Provider

Hetzner
Plan

CAX21
Location

Falkenstein DC Park 1
vCPU

4
RAM

8 GB
Storage

80 GB NVMe
Network

Unknown
OS image

ubuntu-24.04
Trigger

Scheduled
Queued at

Oct 15, 2026, 11:54:00 UTC

Visit Hetzner

CPU

Version	Single-core	Multi-core	Result link
Geekbench 7	963	3,424	Geekbench Browser

Scores from different Geekbench versions aren't comparable with each other. Charts only use Geekbench 7.

Memory, TLS and scheduler

Memory bandwidth (STREAM)

Test	Result	Runs	Spread
Copy higher is better	73.4 GB/s	3	2.06%
Scale higher is better	73.2 GB/s	3	0.941%
Add higher is better	69.2 GB/s	3	1.36%
Triad higher is better	69.8 GB/s	3	2.54%

Average of 3 runs. Spread is the standard deviation as a share of the average.

TLS and crypto (OpenSSL)

Test	Result	Runs	Spread
RSA 4096 signs higher is better	194/s	3	0.555%
RSA 4096 verifies higher is better	15,893/s	3	0.162%
SHA-256 higher is better	2.9 GB/s	3	0.904%
AES-256-GCM higher is better	9.39 GB/s	3	0.0515%
ChaCha20-Poly1305 higher is better	3.45 GB/s	3	0.0639%

Average of 3 runs. Spread is the standard deviation as a share of the average.

Scheduler latency (schbench)

Test	Result	Runs	Spread
Request latency, median lower is better	15 ms	3	0.123%
Requests per second, 90th percentile higher is better	263/s	3	0.219%
Request latency, 99.9th percentile lower is better	24.4 ms	3	11.1%
Requests per second, average higher is better	262/s	3	0.046%

Average of 3 runs. Spread is the standard deviation as a share of the average.

Disk

Block size	Read	Write	Total
4K	113.6 MB/s (29,078 IOPS)	113.5 MB/s (29,059 IOPS)	227.1 MB/s (58,137 IOPS)
64K	946.5 MB/s (15,144 IOPS)	974.7 MB/s (15,594 IOPS)	1.88 GB/s (30,738 IOPS)
512K	1.08 GB/s (2,221 IOPS)	1.18 GB/s (2,411 IOPS)	2.26 GB/s (4,632 IOPS)
1M	1.29 GB/s (1,318 IOPS)	1.44 GB/s (1,471 IOPS)	2.72 GB/s (2,789 IOPS)

fio random read and write, 50/50, run by YABS against the boot disk.

Network

IPv4 transfer speed

Target	Protocol	Send	Receive
Clouvider, London, UK (10G)	IPv4	4.77 Gbps	4.51 Gbps
Eranium, Amsterdam, NL (100G)	IPv4	5.99 Gbps	5.43 Gbps
Leaseweb, Singapore, SG (10G)	IPv4	1.07 Gbps	1.35 Gbps
Clouvider, Los Angeles, CA, US (10G)	IPv4	830 Mbps	1.23 Gbps
Leaseweb, NYC, NY, US (10G)	IPv4	1.79 Gbps	2.61 Gbps

Public iperf3 servers vary in capacity and availability. Compare a target only with itself over time.

Datastores

PostgreSQL

fsync speed (pg_test_fsync)

Method	Operations per second	Time per operation
One 8 kB write
open_datasync	3,362	0.297 ms
fdatasync	3,139	0.319 ms
fsync	1,210	0.827 ms
fsync_writethrough	Not supported here	n/a
open_sync	1,204	0.831 ms
Two 8 kB writes
open_datasync	1,569	0.637 ms
fdatasync	2,807	0.356 ms
fsync	1,155	0.866 ms
fsync_writethrough	Not supported here	n/a
open_sync	623	1.61 ms
open_sync at different write sizes
1 * 16kB open_sync write	1,189	0.841 ms
2 * 8kB open_sync writes	635	1.57 ms
4 * 4kB open_sync writes	304	3.29 ms
8 * 2kB open_sync writes	165	6.05 ms
16 * 1kB open_sync writes	78.3	12.8 ms
fsync on a non-write descriptor
write, fsync, close	1,124	0.89 ms
write, close, fsync	1,138	0.879 ms
Non-synced 8 kB writes
write	1,016,768	0.001 ms

pg_test_fsync against the Docker volume Postgres keeps its data on. fdatasync is Linux's default wal_sync_method.

pgbench at scale 10

Workload	Clients	Transactions/s	Median	p95	p99	p99.9
Read/write (TPC-B like)	1	1,075	0.854 ms	1.35 ms	1.88 ms	3.45 ms
Read/write (TPC-B like)	8	4,240	1.62 ms	3.42 ms	5.02 ms	8.12 ms
Read/write (TPC-B like)	32	4,438	5.44 ms	18.2 ms	30.5 ms	46.7 ms
Read only	1	18,718	0.046 ms	0.083 ms	0.122 ms	0.22 ms
Read only	8	63,214	0.115 ms	0.221 ms	0.282 ms	0.486 ms
Read only	32	63,397	0.461 ms	0.772 ms	0.976 ms	1.56 ms

Scale 10 (1,000,000 accounts), loaded in 1.45 s. 60 s per run, prepared statements. Latency comes from sampled transaction logs, and p99.9 is left out when fewer than 1,000 were sampled.

pgbench at scale 100

Workload	Clients	Transactions/s	Median	p95	p99	p99.9
Read/write (TPC-B like)	1	949	0.96 ms	1.52 ms	2.31 ms	4.08 ms
Read/write (TPC-B like)	8	4,247	1.67 ms	2.97 ms	7.2 ms	11.8 ms
Read/write (TPC-B like)	32	5,257	5.56 ms	10 ms	16.2 ms	29.7 ms
Read only	1	13,920	0.063 ms	0.118 ms	0.167 ms	0.307 ms
Read only	8	48,756	0.149 ms	0.291 ms	0.376 ms	0.988 ms
Read only	32	49,312	0.6 ms	1 ms	1.28 ms	2.89 ms

Scale 100 (10,000,000 accounts), loaded in 13.8 s. 60 s per run, prepared statements. Latency comes from sampled transaction logs, and p99.9 is left out when fewer than 1,000 were sampled.

MySQL

sysbench OLTP

Workload	Threads	Transactions/s	Queries/s	Median	p95	p99
Read/write	1	188	3,761	5 ms	7.17 ms	11.7 ms
Read/write	8	895	17,894	8.43 ms	13.7 ms	20.4 ms
Read/write	32	1,108	22,169	27.7 ms	47.5 ms	57.9 ms
Read only	1	621	9,938	1.5 ms	2.31 ms	3.02 ms
Read only	8	2,389	38,218	3.25 ms	4.82 ms	6.1 ms
Read only	32	2,138	34,211	15 ms	20.4 ms	23.9 ms

10 tables of 100,000 rows, loaded in 6.1 s. 60 s per run, one connection per thread.

Redis

memtier

Mix	Operations/s	Hits/s	Median	p99	p99.9
1:10 set:get	81,964	74,512	0.751 ms	1.5 ms	2.62 ms
1:1 set:get	80,177	40,088	0.767 ms	1.52 ms	2.64 ms

1,000,000 keys of 32 bytes. 64 connections over 4 threads, 60 s per ratio.

Web frameworks

Framework	Fortunes req/s	Single query req/s	JSON req/s	Fortunes p99 at 256 connections
Next.js nextjs	873	2,924	4,348	406 ms
Express express-postgresjs	11,865	14,578	35,463	45.5 ms
Fastify fastify-postgres	11,464	15,256	40,335	46 ms
NestJS nestjs	5,205	6,087	18,795	78.4 ms
Nuxt nuxt	9,600	12,905	32,374	64.2 ms
FastAPI fastapi	4,759	5,873	32,017	196 ms
Django django-postgresql	2,645	2,922	6,849	264 ms
Rails rails	3,398	4,435	16,823	123 ms
Laravel laravel	953	991	1,874	296 ms
Symfony symfony	1,230	1,376	3,708	230 ms

Peak requests per second across the 64 and 256 connection levels, measured with wrk. Each test has to pass the TechEmpower response checks first, or it's marked invalid.

Web apps

WooCommerce

Page mix

Step	Requests/s	Median	p95	p99	Error rate
8 virtual users
Home page	3.27	490 ms	619 ms	643 ms	0%
Shop page	3.27	664 ms	796 ms	840 ms	0%
Category page	3.27	425 ms	682 ms	744 ms	0%
Product page	3.27	459 ms	563 ms	608 ms	0%
Product search	3.27	391 ms	526 ms	586 ms	0%
32 virtual users
Home page	3.14	1.92 s	2.21 s	2.3 s	0%
Shop page	3.14	2.48 s	2.89 s	2.97 s	0%
Category page	3.14	1.94 s	2.59 s	2.76 s	0%
Product page	3.14	1.84 s	2.1 s	2.31 s	0%
Product search	3.14	1.69 s	1.92 s	1.96 s	0%

Cart and checkout

Step	Requests/s	Median	p95	p99	Error rate
8 virtual users
Add to cart	6.37	507 ms	663 ms	703 ms	0%
Cart	6.37	414 ms	546 ms	576 ms	0%
Checkout	6.37	330 ms	431 ms	463 ms	0%
32 virtual users
Add to cart	5.99	1.97 s	2.36 s	2.43 s	0%
Cart	5.99	1.86 s	2.1 s	2.19 s	0%
Checkout	5.99	1.46 s	1.69 s	1.77 s	0%

k6 with a fixed number of virtual users, 61 to 68 s per level. The error rate counts responses that weren't a 200.

Discourse

Signed in

Step	Requests/s	Median	p95	p99	Error rate
1 virtual user
Topic page	1.4	317 ms	424 ms	741 ms	0%
Topic JSON	0.71	159 ms	181 ms	191 ms	0%
Home page	0.71	285 ms	382 ms	477 ms	0%
Categories page	0.347	328 ms	352 ms	442 ms	0%
User activity page	0.347	199 ms	292 ms	452 ms	0%
4 virtual users
Topic page	5.82	306 ms	356 ms	660 ms	0%
Topic JSON	2.88	159 ms	195 ms	245 ms	0%
Home page	2.88	272 ms	421 ms	750 ms	0%
Categories page	1.47	319 ms	358 ms	581 ms	0%
User activity page	1.44	186 ms	223 ms	375 ms	0%
16 virtual users
Topic page	5.37	1.24 s	1.64 s	2.17 s	0%
Topic JSON	2.7	925 ms	1.16 s	1.51 s	0%
Home page	2.68	1.16 s	1.39 s	2 s	0%
Categories page	1.37	1.24 s	1.54 s	2.37 s	0%
User activity page	1.33	966 ms	1.22 s	1.41 s	0%

Anonymous

Step	Requests/s	Median	p95	p99	Error rate
16 virtual users
Topic page	115	34.7 ms	66.6 ms	97.1 ms	0%
Topic JSON	57.8	33.2 ms	64.4 ms	100 ms	0%
Home page	57.8	35.2 ms	67 ms	100 ms	0%
Categories page	28.9	34.4 ms	65.9 ms	95.3 ms	0%
User activity page	28.9	198 ms	259 ms	455 ms	0%

k6 with a fixed number of virtual users, 60 to 61 s per level, after a 21 s warmup. The error rate counts responses that weren't a 200. Discourse ran with 8 app workers and 2048 MB of Postgres shared buffers, what its installer picks for a 8 GB box. Its installer adds no swap on a box this big. The container peaked at 2,848 MB.

Ghost

Readers

Step	Requests/s	Median	p95	p99	Error rate
1 virtual user
Home page	6.26	46 ms	50.4 ms	54.6 ms	0%
Post page	15.6	25 ms	29 ms	36.7 ms	0%
Tag page	3.13	30.9 ms	35.2 ms	40.9 ms	0%
Author page	3.13	31.1 ms	34.8 ms	38.1 ms	0%
RSS feed	3.13	33.9 ms	43.2 ms	49.5 ms	0%
4 virtual users
Home page	7.97	132 ms	158 ms	169 ms	0%
Post page	19.9	93.7 ms	125 ms	142 ms	0%
Tag page	3.98	85.2 ms	111 ms	122 ms	0%
Author page	4	87.1 ms	112 ms	121 ms	0%
RSS feed	3.98	90.4 ms	122 ms	129 ms	0%
16 virtual users
Home page	8.03	506 ms	570 ms	600 ms	0%
Post page	20	425 ms	485 ms	512 ms	0%
Tag page	3.98	282 ms	336 ms	364 ms	0%
Author page	3.98	280 ms	331 ms	348 ms	0%
RSS feed	4.02	282 ms	338 ms	361 ms	0%

Staff in Ghost Admin

Step	Requests/s	Median	p95	p99	Error rate
1 virtual user
Post list	3.15	43.5 ms	50.6 ms	63.2 ms	0%
Post editor	9.48	18.1 ms	21.7 ms	24.3 ms	0%
Saving a draft	9.48	49.4 ms	57.5 ms	65.9 ms	0%
Member list	3.15	57.8 ms	67.2 ms	74.2 ms	0%
4 virtual users
Post list	4.5	57.8 ms	65.3 ms	71.9 ms	0%
Post editor	13.5	28.9 ms	35.5 ms	39.4 ms	0%
Saving a draft	13.6	214 ms	271 ms	286 ms	0%
Member list	4.51	67.7 ms	77 ms	83 ms	0%

k6 with a fixed number of virtual users, 60 s per level, after a 20 s warmup. The error rate counts responses that weren't a 200. Ghost ran the way its ghost-docker setup runs it: one Node process, and MySQL with its stock 128 MB buffer pool. Memory peaked with Ghost at 515 MB and MySQL at 627 MB.

Wagtail

Visitors

Step	Requests/s	Median	p95	p99	Error rate
1 virtual user
Home page	1.11	121 ms	132 ms	137 ms	0%
Blog index	1.1	412 ms	455 ms	464 ms	0%
Blog post	3.31	38.4 ms	60.1 ms	63.4 ms	0%
Bread page	3.31	42.2 ms	47.4 ms	49.8 ms	0%
Location page	1.11	39.6 ms	46.4 ms	47.4 ms	0%
Site search	1.1	67.2 ms	122 ms	132 ms	0%
4 virtual users
Home page	1.87	292 ms	495 ms	645 ms	0%
Blog index	1.9	1.11 s	1.55 s	1.69 s	0%
Blog post	5.61	75.8 ms	147 ms	207 ms	0%
Bread page	5.66	87.2 ms	154 ms	171 ms	0%
Location page	1.92	78.5 ms	151 ms	177 ms	0%
Site search	1.85	110 ms	292 ms	343 ms	0%
16 virtual users
Home page	2.34	768 ms	993 ms	1.07 s	0%
Blog index	2.46	2.11 s	2.41 s	2.6 s	0%
Blog post	7.03	432 ms	669 ms	776 ms	0%
Bread page	7.13	438 ms	678 ms	824 ms	0%
Location page	2.39	419 ms	697 ms	804 ms	0%
Site search	2.34	514 ms	739 ms	863 ms	0%

Editors in the Wagtail admin

Step	Requests/s	Median	p95	p99	Error rate
1 virtual user
Admin dashboard	0.562	116 ms	132 ms	154 ms	0%
Page explorer	1.14	404 ms	437 ms	597 ms	0%
Page editor	1.7	196 ms	244 ms	406 ms	0%
Image library	0.562	81.3 ms	90.9 ms	767 ms	0%
Admin page search	0.562	46.8 ms	99.4 ms	173 ms	0%
4 virtual users
Admin dashboard	0.724	387 ms	731 ms	799 ms	0%
Page explorer	1.43	1.34 s	3.11 s	3.59 s	0%
Page editor	2.16	585 ms	1.09 s	1.31 s	0%
Image library	0.724	277 ms	644 ms	700 ms	0%
Admin page search	0.724	132 ms	440 ms	517 ms	0%

k6 with a fixed number of virtual users, 60 to 61 s per level, after a 21 s warmup. The error rate counts responses that weren't a 200. Wagtail ran the way bakerydemo ships it: uWSGI with 2 workers of 4 threads each, whatever the box, and Postgres and Redis on stock settings. Memory peaked with Wagtail at 293 MB, Postgres at 153 MB, and Redis at 25 MB.

Advertised vs. measured

Spec	Advertised	Measured
vCPU	4	4
CPU	Shared vCPU	0% of busy CPU time stolen
RAM	8 GB	7.5 GiB
Storage	80 GB NVMe	75 GiB
Architecture	arm64	arm64
Virtualization	KVM	KVM

Guests usually report slightly less RAM and disk than advertised because of kernel reservations and filesystem overhead.

Detected hardware

CPU model: Neoverse-N1
CPU cores: 4
CPU frequency: Unknown
AES-NI: Yes
VM-x/AMD-V: No
RAM reported: 7.5 GiB
Swap reported: 0 MiB
Disk reported: 75 GiB
Virtualization reported: KVM
CPU steal: 0% of busy CPU time stolen
Distro: Ubuntu 24.04.4 LTS
Kernel: 6.8.0-138-generic
Architecture: arm64
IPv4 reachable: Yes
IPv6 reachable: Yes

Advertised at test time

Plan: CAX21
Product line: cost_optimized
Location: Falkenstein DC Park 1
vCPU: 4 shared vCPU
Architecture: arm64
RAM: 8 GB
Storage: 80 GB NVMe
Virtualization: KVM
Network: Unknown
Transfer: 21.99 TB
Port 25: On request
Monthly price at test time: $12.49
Hourly price at test time: $0.020
YABS version: v2026-09-20
Ran at: Oct 15, 2026, 11:54:44 UTC
Parser version: 5

Run timeline

Queued: Oct 15, 2026, 11:54:00 UTC
Server created: Oct 15, 2026, 11:54:37 UTC
First update from the server: Oct 15, 2026, 11:54:37 UTC
Completed: Oct 15, 2026, 13:30:22 UTC
Server deleted: Oct 15, 2026, 13:30:22 UTC
Cost of this run: $0.032

TestResultTime

File Compression Not run

Navigation Not run

HTML5 Browser Not run

PDF Viewer Not run

Photo Library Not run

Clang Not run

Text Processing Not run

Asset Compression Not run

HDR Not run

Photo Editor Not run

Ray Tracer Not run

Structure from Motion Not run

Game Physics Not run

Video Encoder Not run

Audio Encoder Not run

Video Player Not run

Watch party 0 watching

This party closed at 14:30 UTC. It's read only now.

Loading the party

RackCensus

Browse

Server

CPU

Memory, TLS and scheduler

Memory bandwidth (STREAM)

TLS and crypto (OpenSSL)

Scheduler latency (schbench)

Disk

Network

IPv4 transfer speed

Datastores

PostgreSQL

fsync speed (pg_test_fsync)

pgbench at scale 10

pgbench at scale 100

MySQL

sysbench OLTP

Redis

memtier

Web frameworks

Web apps

WooCommerce

Page mix

Cart and checkout

Discourse

Signed in

Anonymous

Ghost

Readers

Staff in Ghost Admin

Wagtail

Visitors

Editors in the Wagtail admin

Advertised vs. measured

Detected hardware

Advertised at test time

Run timeline

Watch party