CPU and RAM comparison with sysbench: Start-like versus Power-like

Topics about Proxmox 9, KVM virtualisation, host layout and practical notes for VPS environments.
rychy
Site Admin
Posts: 132
Joined: Tue Oct 20, 2026 10:45 am

CPU and RAM comparison with sysbench: Start-like versus Power-like

Post by rychy »

🇧🇬 Прочети на български

After the disk benchmarks we also ran a separate CPU / RAM test with sysbench to see more clearly the difference between a lighter and a stronger VPS profile. The test was run in two temporary LXC clones made from the plan templates, not directly on the Proxmox host.

Test environment

  • Start-like: 2 vCPU, 4 GB RAM
  • Power-like: 4 vCPU, 8 GB RAM
  • sysbench was used for the CPU and memory tests
  • this is not a synthetic test of the host, but a measurement inside the temporary instances themselves

First results

Code: Select all

Metric                     Start-like        Power-like
-------------------------------------------------------
CPU single-thread          474.09 eps        472.63 eps
CPU all cores              925.28 eps       1841.70 eps
Memory read            46927.98 MiB/s    46962.82 MiB/s
Memory write           28681.49 MiB/s    11653.67 MiB/s

Second memory pass

Code: Select all

Metric                              Start-like        Power-like
----------------------------------------------------------------
4K read, all cores               22135.75 MiB/s    35270.54 MiB/s
4K write, all cores               1836.54 MiB/s     1933.31 MiB/s
1M read, 1 thread                25636.51 MiB/s    25607.77 MiB/s
1M write, 1 thread               21209.21 MiB/s    19720.22 MiB/s

What we see here

  • single-thread CPU is almost the same, which is normal on the same host processor
  • with multi-core CPU Power-like is almost 2x above Start-like, which matches 4 versus 2 vCPU
  • for memory read Power-like shows a real lead in the multi-core 4K profile
  • for memory write there is no strong, clear signal of a big difference between the two profiles
  • the first, lower write result for Power-like should not be read on its own as a marketing conclusion

Practical conclusion

  • if your workload is more sensitive to a single core, the difference will not be dramatic
  • if the workload uses more threads, the Power-like profile has a clear lead
  • a disk benchmark alone is not enough to judge a VPS plan, which is why the CPU and RAM tests are the important second half

For the storage comparison see also the topics Comparing the storage pools in Proxmox and Comparison table: Start vs Power vs Plus vs Micron.