Startup winner
Brave · 0.51s
Brave-4% on cold launch with a real first tab loading.
Independent controlled test · real websites
Real websites,
measured head to head.
Helium 0.17.0.1 and Brave 1.95.101, same Chromium 153 engine, pointed at the real internet: YouTube, reddit, CNN, Amazon and Google Maps, side by side on the same machine. Measured data, not review claims.
CPU winner, real web
Helium · 3.71 CPU-s
Helium -22% over the same 100-second five-tab workload.
Peak RAM winner
Brave · 769MB
Brave-8% peak PSS while loading the five sites.
The resource picture
Every number below is the median of three runs against live sites, with the full range in brackets. Both browsers' ad and tracker blocking was left on, as shipped.
Google Maps
Amazon.com
cnn.com
reddit.com
YouTube
5-tab PSS over time (median run)
Helium Brave
Cumulative CPU, same workload
Helium Brave
| Metric | Helium 0.17.0.1 | Brave 1.95.101 | Result |
|---|---|---|---|
| Cold launch | 0.53 [0.51-0.53]s | 0.51 [0.5-0.55]s | Brave-4% |
| Idle PSS, steady | 377.0 [376.0-377.0]MB | 346.0 [346.0-349.0]MB | Brave-9% |
| Real-web PSS, steady | 736.0 [730.0-741.0]MB | 732.0 [731.0-736.0]MB | Near tie |
| Real-web PSS, peak | 828.0 [825.0-832.0]MB | 769.0 [767.0-786.0]MB | Brave-8% |
| Real-web RSS, steady | 2038.0 [2029.0-2039.0]MB | 1966.0 [1952.0-1973.0]MB | Brave-4% |
| Real-web CPU / ~100s | 3.7 [3.5-4.0] CPU-s | 4.7 [4.2-4.9] CPU-s | Helium -22% |
| Processes at rest | 9 | 8 |
Controlled by design
Same machine, same engine major version, identical flags, fresh profile per run, browser order alternated every round to cancel drift.
Fresh profile, about:blank, launch timed to first CDP response, then 45 seconds sampled at 1Hz.
YouTube, reddit, CNN, Amazon and Google Maps opened via CDP, 3 seconds apart, sampled for 100 seconds. Ads, trackers and each browser's blocking were fully live - that asymmetry is part of real-world use.
All runs hit the live internet minutes apart in alternated order, so network conditions averaged out across both browsers. Per-run tab titles are recorded to prove what actually loaded.
Built to be checked
The download includes the runner and sampler scripts, every raw 1Hz CSV, per-run metadata (launch time, versions, tab titles), browser logs, environment capture and the aggregation script.
PSS = Σ process-tree /proc/<pid>/smaps_rollup
RSS = Σ process-tree /proc/<pid>/status VmRSS
CPU = Δ(utime + stime) / 100 jiffies per second
Launch = Popen → first successful CDP /json/version response
RSS = Σ process-tree /proc/<pid>/status VmRSS
CPU = Δ(utime + stime) / 100 jiffies per second
Launch = Popen → first successful CDP /json/version response
The local-page run, for reference
An earlier round measured the same browsers against locally served pages (no network, no ads). It isolated engine overhead; the real-web numbers above are the headline.
| Metric (local pages) | Helium | Brave | Result |
|---|---|---|---|
| Cold launch | 0.43s | 0.50s | Helium −14% |
| Idle PSS, steady | 368MB | 344MB | Brave −6.3% |
| 5-tab PSS, peak | 520MB | 382MB | Brave −27% |
| 5-tab CPU / 89s | 1.8 CPU-s | 0.8 CPU-s | Brave −56% |
Read the boundaries
These numbers describe this controlled environment. They are not a universal desktop ranking.
Headless sandbox
Ubuntu 22.04 container, Intel Xeon 2.60GHz, 2 cores, about 2GB RAM, no swap, GPU disabled. Relative ordering is the signal.
Live network is a variable
Ads and page weight change between loads. Alternated order plus repeated runs average this out; ranges are published next to every median.
Blockers were live
Brave Shields and Helium's default filtering both ran. Different amounts of blocked ad/tracker work are a real product difference, now included rather than held constant.
Datacenter IP
Some sites serve lighter pages or challenges to datacenter IPs. Per-run tab titles in the metadata show exactly what loaded in every run.
Small sample
Three runs per phase per browser. Enough for repeatable medians here, not fine-grained statistics.
Matched engines
Helium Chromium 153.0.8010.36 vs Brave Chromium 153.0.8010.37, both on defaults, muted audio, identical window size.