What was tested, and what it is not
These are tests of BroadNet’s own build server, not of any plan we sell. The server is an older dual-socket machine used for builds and image processing, and other jobs may have been running. The results show our method in action and give you commands to compare with any server, including the one in your quote.
| CPU | Intel(R) Xeon(R) CPU E5-2686 v4 @ 2.30GHz × 2 sockets (36 cores, 72 threads) |
|---|---|
| CPU limit for this test | the build account is limited by its control group to 16 CPUs; multi-thread tests use 16 threads |
| Memory | 251 GB installed |
| Storage | hardware RAID logical volume reported as rotational (ext4) |
| Operating system | Ubuntu 24.04.4 LTS, kernel 6.8.0-138-generic |
| Tools | OpenSSL 3.0.13 30 Jan 2024 (Library: OpenSSL 3.0.13 30 Jan 2024); gcc (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0 |
| Run finished | 2026-11-08T10:04:42Z (UTC) |
| Load average before / after | 1.08 4.47 3.71 / 5.93 5.70 4.51 |
Results
Each test ran 3 times; the table shows the median and the range.
| Test | Median | Range | Unit |
|---|---|---|---|
| AES-256-GCM, 1 thread (16 KiB blocks) | 2,673 | 2,666 to 2,727 | MB/s |
| AES-256-GCM, 16 threads (16 KiB blocks) | 38,582 | 37,288 to 38,666 | MB/s |
| SHA-256, 1 thread (16 KiB blocks) | 369 | 364 to 372 | MB/s |
| RSA-2048 signatures, 1 thread | 1,394 | 1,322 to 1,400 | signs/s |
| RSA-2048 signatures, 16 threads | 19,854 | 17,785 to 19,900 | signs/s |
| Memory bandwidth, triad, 1 thread | 10.3 | 10.2 to 10.3 | GB/s |
| Memory bandwidth, triad, 16 threads | 74.6 | 72.5 to 75.3 | GB/s |
| Sequential write, 1 MiB direct I/O, 4 GiB | 438 | 406 to 622 | MB/s |
| Sequential read, 1 MiB direct I/O, 4 GiB | 299 | 166 to 324 | MB/s |
| Synchronous 4 KiB write latency (2,000 writes) | 628 | 303 to 1,255 | µs per write |
Exact commands
Run these on any Linux server to compare. The full script, raw output for every run and the parser are in the site source (scripts/bench/ and reports/bench-raw/).
# AES-256-GCM, 1 thread (16 KiB blocks)
openssl speed -seconds 3 -evp aes-256-gcm
# AES-256-GCM, 16 threads (16 KiB blocks)
openssl speed -seconds 3 -multi 16 -evp aes-256-gcm
# SHA-256, 1 thread (16 KiB blocks)
openssl speed -seconds 3 -evp sha256
# RSA-2048 signatures, 1 thread
openssl speed -seconds 3 rsa2048
# RSA-2048 signatures, 16 threads
openssl speed -seconds 3 -multi 16 rsa2048
# Memory bandwidth, triad, 1 thread
gcc -O3 -march=native -fopenmp triad.c -o triad && OMP_NUM_THREADS=1 ./triad
# Memory bandwidth, triad, 16 threads
OMP_NUM_THREADS=16 OMP_PROC_BIND=spread ./triad
# Sequential write, 1 MiB direct I/O, 4 GiB
dd if=/dev/zero of=ddtest.bin bs=1M count=4096 oflag=direct conv=fsync
# Sequential read, 1 MiB direct I/O, 4 GiB
dd if=ddtest.bin of=/dev/null bs=1M iflag=direct
# Synchronous 4 KiB write latency (2,000 writes)
dd if=/dev/zero of=dsync.bin bs=4k count=2000 oflag=dsyncHow to read these numbers
- OpenSSL AES-256-GCM shows hardware-accelerated encryption throughput (16 KiB blocks), relevant to TLS-heavy services. All-thread figures scale with core count.
- RSA-2048 signs per second approximates TLS handshake capacity for RSA certificates.
- Memory triad is a STREAM-style a = b + s × c loop over three 640 MB arrays (our own short C program, shown in the script; not the official STREAM binary); it measures sustainable memory bandwidth, which bounds many data-heavy workloads.
- Sequential disk uses 1 MiB direct I/O to bypass the page cache. Synchronous 4 KiB writes approximate the commit latency a database sees.
None of these is a complete benchmark of an application. The plan benchmark method adds fio, network and application-level tests.
Frequently asked questions
Why publish tests of a server you do not sell?
To show the method and the level of detail you can expect when we benchmark plans: exact commands, run counts, medians, ranges and raw output, with the limits stated.
Can I run the same tests on my current server?
Yes. The commands need only OpenSSL, GCC and coreutils. Compare medians of several runs, at quiet times, and treat differences under about 10% as noise.