How hard the TankCoach logger hits your game
Logger 2.28 takes about 1 % of game time — roughly 1 ms out of every 100. At 60 FPS that is about 0.2 ms per frame, and a frame lasts 16.7 ms. The longest single hitch (about 27 ms) happens after the battle, while the results are written; during the battle no logger call came close to one frame.
Before (2.27, 8 battles): 2.8–5.7 % of the time (median 4.1 %), 28–57 ms per second of battle, and a 26–29 ms hitch at the very start, before the countdown ended. On 2.28 the load dropped about fourfold, and the start-of-battle hitch is down to 1–1.5 ms.
Share of game time, battle by battle
How much of the game's main-thread time the logger took over the battle. Orange — 2.27, green — 2.28. One 2.27 battle is excluded (see below).
FPS dips at the start are the map loading
FPS second by second in one battle (AMX 13 57, Airfield, logger 2.28). The first 7 s run at 4–13 FPS while the game finishes loading the map. That happened in 13 of 15 battles; the other two started straight at 60 FPS. After that, average FPS per battle was 47–59. You can open this very battle as an example report.
All battles
| Logger | Tank | Map | Length | % of time | ms per 1 s | Longest hitch | Avg FPS |
|---|---|---|---|---|---|---|---|
| 2.27 | Object 261 | Abbey | 10:34 | 3.37 | 33.7 | 31 ms · battle start, before the countdown | 56 |
| 2.27 | 105 leFH18B2 | Pilsen | 7:36 | 3.61 | 36.1 | 29 ms · battle start, before the countdown | 58 |
| 2.27 | Pz.Kpfw. I Ausf. C | Lakeville | 5:45 | 4.59 | 45.9 | 30 ms · battle start, before the countdown | 58 |
| 2.27 | AMX 13 57 | Overlord | 4:26 | 5.69 | 56.9 | 29 ms · battle start, before the countdown | 56 |
| 2.27 | Vz. 64 Blesk | Mittengard | 8:29 | 3.72 | 37.2 | 32 ms · battle start, before the countdown | 55 |
| 2.27 | AMX 13 57 | Ruinberg | 6:53 | 4.98 | 49.8 | 31 ms · after the battle, results | 52 |
| 2.27 | Progetto M40 mod. 65 | Great Wall | 5:01 | 2.83 | 28.3 | 32 ms · battle start, before the countdown | 56 |
| 2.27 | Progetto M40 mod. 65 | Highway | 7:49 | 4.41 | 44.1 | 31 ms · battle start, before the countdown | 58 |
| 2.27 | Progetto M40 mod. 65 | Highway | 87:37 | 0.11 * | 1.1 | 30 ms · battle start, before the countdown | 39 |
| 2.28 | SU-76I | Mines | 8:34 | 0.82 | 8.2 | 20 ms · after the battle, results | 59 |
| 2.28 | AMX 13 57 | Airfield | 6:40 | 1.20 | 12.0 | 27 ms · after the battle, results | 52 |
| 2.28 | T-34-85 Rudy | Widepark | 4:33 | 1.29 | 12.9 | 28 ms · after the battle, results | 48 |
| 2.28 | Progetto M40 mod. 65 | Ghost Town | 8:55 | 1.05 | 10.5 | 27 ms · after the battle, results | 47 |
| 2.28 | Kranvagn | Overlord | 6:14 | 1.64 | 16.4 | 28 ms · after the battle, results | 54 |
| 2.28 | UDES 15/16 | Westfield | 6:13 | 0.83 | 8.3 | 28 ms · after the battle, results | 57 |
| 2.28 | BZ-68 | Fjords | 5:08 | 1.52 | 15.2 | 27 ms · after the battle, results | 51 |
"Length" is how long the logger was recording. "Longest hitch" is the longest single logger call and when it happened. * The Oct 23, 08:33 battle (2.27, Progetto M40 mod. 65) is left out: the recording clock ran for 88 minutes while the logger actually worked for about 5 (557 ticks instead of ~10,000, and 77 FPS samples) — the game was most likely minimized or paused. Its 0.11 % comes from dividing by 83 extra minutes and says nothing about the load.
How we measured
- The logger times every one of its own handlers (
time.clock) and adds them up per battle. "% of time" = that total / battle time on the same clock. The timing itself costs about 1.5 µs per call — ≈ 0.01–0.02 % of the time, too small to matter. - FPS comes from the game itself:
BigWorld.getFPS(), once a second. - Data: 16 WG battles (World of Tanks NA) on one PC — 9 on logger 2.27 (Oct 22–23) and 7 on 2.28 (Oct 23).
- The PC, per its owner, roughly: Ryzen 5, 32 GB RAM, AMD RX 5600 6 GB. The logger itself does not record the CPU model.
What changed in 2.28
- Tank HP is polled 5 times a second instead of 20, and the exact moment of damage comes from the game's own event (
onHealthChanged). Same battle data. - The poll only walks tanks, not every object in the game (shells, effects and so on) — that used to be about two thirds of the whole load.
- A heavy check at battle start moved to the next tick: the start-of-battle hitch is gone (26–29 ms before, about 1 ms now).
Questions or your own numbers — send them on Telegram. How to install the mod.