Performance proof
Infrastructure benchmarks
Representative Ryzen NVMe results with reproduction commands — verify on your instance, not on marketing slides.
HostStack publishes benchmark snapshots so engineers can validate silicon before annual commits. Numbers below are representative of Ryzen-class production nodes — re-run tests on your provisioned instance because neighbour load and SKU size shift results.
Methodology
- Geekbench 6 — single- and multi-core on idle-adjacent windows; compare against your desktop to sanity-check vCPU allocation.
- fio — 4K random and 1M sequential on local NVMe; warm cache noted in context column.
- YABS — community-standard disk/network script; useful for apples-to-apples forum comparisons.
- Provision timer — automated control panel pipeline median on standard KVM SKUs.
| Suite | Metric | Result | Context |
|---|---|---|---|
| Geekbench 6 | Single-core | 2,180+ | Ryzen 7 class node · X4 reference SKU |
| Geekbench 6 | Multi-core | 8,400+ | 4 vCPU allocation · shared scheduler |
| fio | 4K random read | 180K+ IOPS | Local NVMe · cached warm run |
| fio | Seq. read | 3.2 GB/s | NVMe tier · 1M block |
| YABS | Disk speed | 1.8 GB/s | Single-thread · dd equivalent |
| Deploy | Provision time | < 3 min | Standard KVM SKUs · automated pipeline |
How to reproduce
# Geekbench 6 (download from geekbench.com)
geekbench6
# fio random read
fio --name=randread --ioengine=libaio --rw=randread --bs=4k --numjobs=4 --size=1G --runtime=60 --group_reporting
# YABS (community script)
curl -sL yabs.sh | bashShare your results with presales if you are sizing trading or game workloads — we will recommend VDS dedicated threads when steal time appears in your metrics.