Here is a story we hear every week. Someone upgrades to gigabit fiber, runs a speed test, sees 940 Mbps down, screenshots it — and then their video call freezes an hour later while a cloud backup runs in the background. The speed test said excellent. The experience said terrible. Both were telling the truth about different things.

The problem is not your connection. The problem is that a download number alone is a capacity metric, not a quality metric. Today we are shipping the biggest update in SpeedPulse’s history, built around one idea: a speed test should explain your experience, not just flatter your line.

What’s new — the 60-second version

  • Jitter measurement — 8 timed pings, and the variation between them. The metric that makes gaming and calls feel laggy even when speed looks fine.
  • Bufferbloat grade (A+ to F) — we keep pinging while your line is saturated, then grade how much extra latency appears under load.
  • Connection info card — your IP, ISP, ASN and test-server location, resolved via Cloudflare with dual fallbacks.
  • Real-world activity verdicts — plain-English scores for 4K streaming, gaming, video calls and browsing, each one explaining why.
  • Mbps ⇄ MB/s toggle — switch units on any result, because “94 MB/s” is what your file manager actually shows.
  • One-click CSV export — your full test history, no account required.
  • Multi-node latency panel — after every test, we probe 8 independent CDN edges to show whether your ISP treats speed-test traffic specially.

Let’s dig into the two headline metrics, because they change how you should read every speed test you ever run again.

Bufferbloat: the number that explains “fast but laggy”

Bufferbloat is the latency that appears when your link gets busy. Every router has buffers. When a download saturates your line, packets queue in those buffers — and every queued packet waits its turn. A ping that took 12 ms at idle can take 400 ms or more mid-download. Your interactive traffic (game packets, call audio, DNS) sits in the same queue behind a wall of bulk data.

That is why “the internet feels stuck when someone starts a backup” is a real, measurable phenomenon — and why most speed tests never show it. They measure your ping before the load starts, report a nice low number, and move on.

SpeedPulse now runs pings concurrently with the download and upload phases, compares them against your idle latency, and grades the increase on the scale popularized by the bufferbloat.org community and Waveform:

Latency under load: an A+ connection stays flat while an F connection spikes above 600 milliseconds during download and upload saturation Line chart comparing two connections over 25 seconds. During the download and upload saturation phases, the A+ connection's latency stays near 12 milliseconds while the F connection climbs above 600 milliseconds, then recovers when load ends. DOWNLOAD SATURATION UPLOAD SATURATION 0 100 200 300 400 500 600 700 LATENCY (MS) 0s 5s 12s 19s 25s TEST TIMELINE F · +600 MS EXTRA packets trapped in a bloated buffer A+ · <5 MS EXTRA managed queue, latency stays flat LATENCY UNDER LOAD SPEEDPULSE · GRADE A+ VS F
The same speed test, two different connections. Both show ~940 Mbps download. The A+ line barely moves under load; the F line's latency multiplies 50×. Only one of these lines shows up in a traditional speed test result.

The grading scale

We grade the increase in median latency under load, on the scale the industry has converged on:

Extra latency under loadGradeWhat it feels like
< 5 msA+Latency stays flat — gaming and calls coexist with downloads
< 30 msABarely noticeable when the line is busy
< 60 msBMinor added delay under load
< 200 msCCalls and games may lag while backups run
< 400 msDClearly noticeable lag during transfers
≥ 400 msFThe connection can feel stuck when busy

If you want the full story — why big router buffers backfire, and how to fix a bad grade with SQM or fq_codel — we wrote a deep technical guide to bufferbloat.

Jitter: the metric that hides inside your ping

Your ping is an average. Applications don’t experience averages — they experience every individual packet. Jitter measures how unevenly those packets arrive, and we now measure it with 8 precisely-timed samples, scoring the mean absolute difference between consecutive round trips.

Two ping trains of 8 samples: a smooth connection with 2 ms jitter and an erratic one with 45 ms jitter, despite a similar average Top panel: eight ping dots hovering between 24 and 27 milliseconds. Bottom panel: eight ping dots swinging wildly between 8 and 140 milliseconds. Same average, very different experience. PING TRAIN × 8 — THE JITTER PICTURE SPEEDPULSE · JITTER SMOOTH — JITTER ≈ 2 MS 25 ms avg 2426 2524 2725 2624 ERRATIC — JITTER ≈ 45 MS 2085 15120 308 14025 SAME AVERAGE · WILDLY DIFFERENT GAME AND CALL EXPERIENCE — VALUES IN MS
Averages hide chaos. Both trains average roughly 25 ms. The top one delivers a smooth call; the bottom one delivers stutter, rubber-banding and "can you hear me now?"

As a rule of thumb: under 5 ms is excellent, 5–15 ms is fine for almost everyone, and above 30 ms real-time applications start to suffer. We have a full guide to jitter if you want the complete picture.

Verdicts: numbers translated into English

Here is the feature we’re proudest of. After every test, we score your connection for four activities — 4K streaming, online gaming, video calls and web browsing — and, crucially, every score explains itself. Not opaque dots. Not a mystery percentage. A sentence that tells you which metric limited your score and what would fix it.

Example activity verdicts: streaming 5 out of 5, gaming 4 out of 5, video calls 3 out of 5, browsing 5 out of 5, each with a plain-language reason A sample result scorecard with four activity rows, dot ratings from one to five, and short explanatory sentences for each. ACTIVITY VERDICTS — SAMPLE RESULT EVERY SCORE EXPLAINS ITSELF 4K STREAMING Enough download for 8 simultaneous 4K streams (~25 Mbps each). 5/5 ONLINE GAMING Fast (14 ms) and steady (3 ms jitter) — great for competitive play. 4/5 VIDEO CALLS Calls work, but expect freeze-ups when the line is otherwise busy (bufferbloat 180 ms). 3/5 WEB BROWSING Pages and web apps load snappily on this connection. 5/5 SAMPLE VERDICT FOR A 940 MBPS LINE WITH C-GRADE BUFFERBLOAT — SPEED, STABILITY AND LOAD COMBINED
One line, three stories. Gigabit bandwidth earns streaming and browsing a perfect score, but bufferbloat drags video calls down to 3/5 — the exact result our opening story was about.

The verdicts combine all five metrics: download and upload for bandwidth-hungry activities, ping and jitter for responsiveness, and bufferbloat as a penalty for everything real-time. Gaming, for instance, loses points when idle ping exceeds 20 ms, when jitter climbs past 5 ms, or when bufferbloat exceeds 60 ms — because each of those shows up mid-match as input lag.

Also shipping today

Connection info card. Every test now shows your IP, ISP and ASN, city-level location, and which test server you hit — resolved through Cloudflare with two independent fallbacks, so the card never comes up empty.

Mbps ⇄ MB/s. ISPs sell megabits; your file manager shows megabytes. One toggle now switches every speed on the result screen and in your history between Mbps and MB/s (940 Mbps = 117.5 MB/s). Small feature; surprisingly, no major speed test offers it.

CSV export, no account. Your entire test history — download, upload, ping, jitter, bufferbloat, timestamps — exports to CSV in one click. Track your line’s health over time, hand the file to your ISP’s support desk, or feed it into a spreadsheet. No signup wall.

Multi-node latency panel. After each test we ping 8 independent CDN edges (Cloudflare, Fastly, jsDelivr, Bunny, UNPKG, GitHub, Google, AWS) and chart them side by side. If the CDN hosting your speed test is dramatically faster than everything else, that’s a classic sign of speed-test traffic being treated specially by the network. We make that visible. More on the methodology in how the measurement engine works.

What to do with your results

  1. Run the test with your network idle. Note the bufferbloat grade — this is your headline quality number.
  2. Run it again during heavy load (start a big download or a cloud backup first). Watch the loaded latency climb. That delta is bufferbloat.
  3. Check the verdict reasons. They name the limiting metric. If bufferbloat or jitter is the culprit on an otherwise fast line, the fixes are cheap: enable SQM/fq_codel on your router, or cap your upload slightly below its maximum. Our bufferbloat guide walks through every option.

Join the discussion

We built this update around a conviction: the industry over-indexes on bandwidth and under-indexes on latency under load. Push back on that — or bring data:

  • What’s your bufferbloat grade, and what router/ISP combo produced it? (Mine: A on stock pfSense, A+ after enabling fq_codel.)
  • Should regulators require ISPs to publish bufferbloat grades, the way they advertise download tiers?
  • Have you ever caught a speed-test CDN being prioritized? The multi-node panel is designed exactly for that — we’d love to see your screenshots.

Share your results card, argue with our grading thresholds, or ask anything about the methodology — links to the technical deep-dive and the competitor comparison are right there in the share bar.