Price Per Core on Budget VPS Plans: Normalising vCPU Counts for Oversubscription

Price per core is the fastest way to compare budget VPS plans — and the easiest number to get wrong. A $6 plan advertising “4 vCPU” and a $12 plan advertising “2 vCPU” are not a 2x gap in value. On the budget tier, advertised cores are almost always shared cores, and what you pay per usable core depends on how heavily the provider oversubscribes the physical host.

This is the normalisation method we use when building our VPS comparison table: convert every plan to a per-usable-core figure before you look at the monthly price. It takes about twenty minutes per provider and it routinely changes which plan is actually cheapest.

Step 1: Separate dedicated cores from burst cores

Read the spec sheet for the word “dedicated.” On budget plans you will usually see one of three arrangements:

• Shared vCPU with a fair-use cap. The provider gives you a slice of a shared vCPU pool. Typical oversubscription is 4:1 to 12:1 per physical core.
• Burst credits. A single “vCPU” that can spike to 100% briefly but is throttled to a fraction of a core under sustained load.
• Dedicated core (rare under $15). One physical core pinned to your VM. Usually only appears on plans above $20 or on providers that explicitly sell “dedicated CPU” tiers.

Only the third category lets you use the advertised count directly. For the first two, you need a derating factor.

Step 2: Measure sustained throughput, not peak

Benchmarks that run for ten seconds measure burst. Budget hosts are designed to look good in short bursts and throttle in minute five. Run a single-threaded loop for at least fifteen minutes and record the median throughput in the second half of the run:

# sustained single-core throughput (run 15 min, note the last 5 min)
sysbench cpu --cpu-max-prime=20000 --threads=1 --time=900 run

# concurrent check: how many threads before throttling
sysbench cpu --cpu-max-prime=20000 --threads=4 --time=300 run

Divide the sustained result by the provider’s reference single-core score (any published CPU benchmark for that generation). The ratio is your effective core count. A plan sold as 4 vCPU that returns 0.6 cores’ worth of sustained work is effectively a 0.6-core plan, and its real price per core is roughly 6.7x the headline math.

Step 3: Apply the derating factor to price

Effective price per core = monthly price ÷ (advertised cores × derating factor). Two worked examples from plans we tested:

Plan	Price	Advertised vCPU	Sustained score (ref = 100)	Derated cores	Effective $/core
A	$6.00	4	38	1.52	$3.95
B	$12.00	2	88	1.76	$6.82
C	$4.50	2	52	1.04	$4.33

Plan B looks expensive per core here — but it is the only one that held 88% of its advertised throughput under a four-thread load. If your workload is single-threaded (a small web app, a database, a game server), Plan B is the cheapest way to buy reliable latency. If you are running many light background workers, Plan C wins on raw throughput per dollar.

Step 4: Check what else scales with cores

Cores do not exist in isolation. Before you declare a winner, confirm that RAM and disk scale proportionally. A plan with 4 cheap cores and 2 GB of RAM will swap under any real multi-threaded load, which wipes out the core advantage entirely. Pair the core measurement with a memory-bandwidth check and a quick disk test:

# memory bandwidth and latency under load
sysbench memory --memory-block-size=1M --memory-total-size=10G run

# disk: random 4k read IOPS (the number that decides database speed)
fio --name=randread --ioengine=libaio --direct=1 --rw=randread \
    --bs=4k --size=1G --numjobs=4 --iodepth=16 --runtime=120 --group_reporting

If the memory test collapses once the CPU test is running simultaneously, you have found an oversubscribed host before you paid for it. Run both at once — that combination is what your production server actually experiences.

Step 5: Re-run the numbers at renewal

Budget providers commonly change the underlying host generation between your first and second year without changing the plan name. A plan that derated at 0.75 in year one can derate at 0.45 in year two on the same invoice. Re-test sustained throughput ninety days before renewal and recompute the effective price per core. If it has fallen more than 20%, downgrading or migrating is usually worth more than any loyalty discount.

One more consideration that budget buyers routinely miss: a provider’s stated core count is a marketing figure, not a capacity guarantee. Ask what happens when the physical host is full. Good budget hosts document a fair-use policy and publish it; weak ones throttle silently and leave you to discover it. Read that document before you pay, because no amount of benchmarking after the fact will recover a plan whose terms allow the provider to halve your throughput at will.

Take the cheapest plan that derates cleanly against your workload shape. If you are running many light workers, higher advertised core counts win even at low derating; if you need consistent single-thread latency, fewer but better-kept cores win. Matching the measurement to the workload is the whole exercise — our VPS comparison table lists vCPU counts alongside RAM and disk so you can apply the derating factor to every candidate at once.