
A/B Testing CPU Requirement (Per-User) Calculator
Estimate the CPU cores needed to run an A/B test from user traffic, request frequency, processing time, experiment overhead, and target utilization.
Overview
Use this A/B testing CPU requirement calculator to estimate the CPU capacity needed for active users during an experiment. Enter your concurrent user count, request frequency, baseline CPU time, experiment overhead, number of variants, and preferred utilization target.
How it works
The calculator converts user activity into requests per second, then multiplies that request rate by the combined CPU time for each request. This gives the CPU work needed each second. It divides that workload by your target utilization, so capacity includes headroom rather than planning to run at 100% CPU. The per-user result divides total planned cores by concurrent users. Variant count is used to show the expected even traffic split; it does not automatically increase total traffic or CPU workload.
How to use this calculator
- 1Enter the expected number of concurrent users during the busiest part of the test.
- 2Add the average requests each active user makes per minute.
- 3Enter your measured or estimated baseline CPU time per request.
- 4Add the CPU overhead created by experiment assignment, logging, and variant logic.
- 5Set the number of variants and your preferred CPU utilization target.
- 6Review total cores, per-user capacity, request rate, and users per variant.
Example Calculation
Concurrent users
10000
Requests per user per minute
6
Baseline CPU time per request
20
Experiment overhead per request
5
Number of variants
2
Target CPU utilization
70%
Estimated CPU cores required
35.71 cores
This workload produces about 1,000 requests per second. At 25 ms of CPU time per request and a 70% utilization target, it requires approximately 35.71 CPU cores, or 0.003571 cores per concurrent user. Each variant receives about 5,000 concurrent users with an even split.
Frequently asked questions
What does CPU cores per user mean in this calculator?
It is the planned total CPU capacity divided by the number of concurrent users. It helps compare the relative infrastructure cost of different traffic or experiment designs.
Does adding more A/B test variants require more CPU?
Not necessarily. If total traffic and per-request processing stay the same, splitting users among more variants does not by itself increase total CPU demand. Extra variant logic, assets, or analytics can increase the per-request overhead.
What CPU utilization target should I use?
A lower target leaves more headroom for traffic spikes, uneven load, and slower requests. The suitable target depends on your reliability goals, autoscaling speed, and workload variability.
How do I measure CPU time per request?
Use application performance monitoring, load tests, or host and container metrics during representative traffic. Include both normal request processing and experiment-related work where possible.
Does this estimate include database or network capacity?
No. It estimates application CPU capacity only. Database connections, memory, network bandwidth, storage, and third-party service limits should be planned separately.
Why should I load test an A/B experiment?
Load testing can reveal overhead from feature flags, analytics events, cache behavior, database queries, and unequal traffic patterns that a simple average estimate may not capture.
Explore Related Calculators
Assumptions and warnings
Assumptions
- Traffic is distributed evenly across the specified variants.
- Each active user generates requests at a steady average rate.
- The CPU time entered is representative of the application workload under test.
- One CPU core is treated as one CPU second of processing capacity per second.
- The result is a planning estimate and does not include a separate allowance for host operating systems, background jobs, or outages.
Warnings
- This calculator provides an infrastructure planning estimate only; benchmark your application under realistic load before making production capacity decisions.
- Actual CPU use can vary with request mix, caching, database latency, autoscaling behavior, and uneven variant traffic.