Performance Profiles

how-to · Reviewed 2026-11-03 GA

Run the same workload across different hardware, and read the answer as a graph of cost against time to solve. Performance Profiles replaces gut instinct and ad-hoc spreadsheet benchmarking with a self-service comparison.

A Profile represents one workload held constant while the hardware underneath it changes. Fixing everything except the architecture is the point: it is what makes the comparison a measurement rather than an anecdote.

Access is controlled by your organization administrator, so if you do not see Performance in the platform, ask them to enable it.

Where it fits

Rescale recommends three steps, and Performance Profiles is the middle one:

1. Measure. Find the workloads worth optimizing, using the economics dashboard.
2. Optimize. Benchmark those workloads across hardware here.
3. Implement. Fix the winning configuration in a template so the rest of the organization inherits it.

Performance Home

The Performance home page lists your Profiles and is where you create new ones.

Classic UIEach row is one workload benchmarked against several architectures.

Filter by All Profiles, My Profiles, Shared Profiles (shared with you) or Archived Profiles. Six columns are available:

Column	What it shows
Name	The profile name
Runs	How many hardware benchmarks have run against this profile
Last Updated	When a benchmark last ran in it
Description	The description you gave it
Profile Configuration	The template or job the profile was built from
Shared	Who it is shared with, or who shared it with you

Use the … icon to customize which columns and filters appear.

Create a Profile

1. Click New Profile at the top left of Performance Home.

2. Name it. Click Untitled Performance Profile, type a descriptive name, and confirm with the green checkmark.

3. Select the Profile Configuration. Click Select Configuration and pick either a Template or a completed Job. This fixes the software, version, command and input files for every benchmark in the profile.

   The configuration locks after the first run

   You cannot change a profile's configuration once its first benchmark has run. If you need a different configuration, create a new profile.

4. Add the hardware. Click Add Infrastructure, or Add in the table, then select each coretype you want and its core count.

   Classic UIThe drawer links to Coretype Explorer, so you can check a specification without leaving the flow.

   Click Add. To benchmark the same coretype at several core counts, use Add again and select it at each count.

5. Review and run. Check the hardware and core counts, then click Run to start every pending benchmark.

Read the results

The benchmark table

The table lists every benchmark run with its status, run time and cost.

Classic UIRun time defaults to execution time only. Cost covers hardware and software together.

Both numbers need a caveat before you compare them:

• Run time is execution time by default, which excludes time spent in preconditions, starting and stopping. You can change this: Policies > Platform Settings > Performance > Run Time Measurement switches it to total run time.
• Cost includes both hardware and software.

Select a row's checkbox and click Delete to drop an outlier from the table and the graph, or use the eye icon to hide it from the graph while keeping it in the table. Adding more hardware later and clicking Run runs only the new benchmarks, not the existing ones.

The graph

Results plot as Cost against Time to Solve by default, and either axis can be changed.

Classic UILines join the same coretype at different core counts. Cheaper and faster is toward the bottom left.

Axis option	Definition
Cost	The cost to run the job, in USD by default
Time to Solve	Run time in minutes, execution time only unless you changed the measurement
Number of Cores	The core count for that benchmark

Balanced hardware sits toward the bottom left, cheaper and faster. For a comparison with several competing options, look at the points on the Pareto front: the set where nothing else is both cheaper and faster, which is where the genuinely efficient choices are.

Download the results as CSV

Download CSV in the top right exports the benchmarks for your own analysis. The columns are:

Column	Definition
benchmark_name	Name of the profiled job
template_name	Name of the profile configuration used
software_name	Software and version, comma-separated
run_number	Benchmark run number
coretype_name	Name of the hardware
unit_price	Price per unit for that coretype
core_count	Cores used for the benchmark
processor	Processor name
start_timestamp	When the benchmark started
end_timestamp	When the benchmark ended
unit_hours	(run_time_minutes / 60) * core_count

Options on a Profile

• Description. Click the + to add one. Worth using to record which hardware won and why the profile exists.
• Run strategy. Run benchmarks in parallel or serial. Serial matters when software licenses are the constraint and you do not want every benchmark holding one at once.
• Run time metric. Switch between execution time and total run time under Policies > Platform Settings > Performance > Run Time Measurement.

Share a Profile

Sharing a Profile lets other users view its benchmarks.

Option	Required	What it allows
Include both input files and results	Required	Full access to the benchmarks and graphs
Allow in-browser SSH access	Optional	SSH into a running job
Allow adding and removing benchmarks	Optional	Shared users can add and remove benchmarks

FAQs

Is High-Performance Storage supported? No. HPS is available in a single region, which makes it unreliable as a constant across a set of benchmarks.

Can I tweak one benchmark's configuration? Not through the profile, by design: the configuration is locked so hardware is the only variable. But before running a benchmark you can click the hardware name in the Hardware Benchmark Runs table to open that individual job in a new tab, where every setting can be changed. Use this when a particular architecture needs a different command, or when you deliberately want to compare software versions.

Related

• Coretype Explorer to compare specifications before choosing what to benchmark
• Coretypes for what each coretype is built for
• Templates for applying a winning configuration to future jobs
• Economics dashboard for finding the workloads worth profiling
• Hardware for the platform's own recommendations, without running benchmarks