The RigSight log
How RigSight measures a PC
The numbers on this blog come from RigSight's own history. How it records temperatures, frame rates, load, memory and drives, what it compares with what, and what it does with a reading that can't be right.
The numbers in these posts come from RigSight's own history on my PC. Here's how each kind is recorded, so you can judge them for yourself. RigSight keeps all of it on the PC it runs on.
Temperatures
RigSight reads the CPU and GPU temperatures every 2 seconds by default, and every second while its window or the overlay is showing them. For each minute it keeps the average, the highest and the lowest, so a short peak isn't averaged away. On the GPU it keeps three readings, the core, the hot spot, which is the warmest point on the chip, and the memory. Reading the CPU's temperature needs RigSight to run with admin rights.
When a post says "at rest" or "at idle", it means minutes when the CPU and GPU were both under 10% busy. When it says "gaming" or "heavy", it means minutes with the GPU 80% busy or more, unless the post says otherwise.
Frame rate, 1% lows and stutters
Frame rates come from RivaTuner Statistics Server, which has to be running before the game starts. Since version 0.35.0 RigSight reads every frame's time, so the 1% and 0.1% lows are worked out from every frame, not averaged from short windows. A stutter is a frame at least 2.5 times as long as the middle of the 30 frames before it, and at least 20 ms. Nothing counts while a game runs under 20 FPS overall, so loading screens stay out.
CPU and GPU load
Load is how busy each part was, averaged over the minute. "Full load" on the GPU means 95% busy or more. RigSight also records which app was in front, and each app's own share of the CPU, so it can tell a game's load from something running behind it.
Memory
RigSight records the RAM in use for the whole PC and for each app every minute. It also records Windows' committed memory, which is what apps have asked for, against its limit of RAM plus the page file. When committed memory gets close to that limit, Windows is nearly out of memory, however much "in use" shows.
For graphics memory it records the whole card's use, and the game's own when Windows reports it.
Drives
Drive health comes from each drive's own health check, called SMART, which RigSight reads with admin rights. It records each drive's health, the life left an SSD reports, its hours on and its temperature, once a day since version 1.12.0.
Comparing like with like
A temperature only means something next to the load behind it. When RigSight says a part runs hotter than its usual, it compares minutes of the same kind, the same game or the same load and power, and leaves the rest out. When there isn't enough history of that kind yet, it says "Too soon to say" and what it still needs.
Readings that can't be right
Sensors and Windows sometimes report numbers that can't be true. Windows once reported a game using 146 GB of graphics memory on a 12 GB card. Since version 1.10.0, a game's own figure above the whole card isn't shown as the game's. In the posts, a number that looks impossible stays out until I can explain it. My PC reported 146 GB of VRAM tells that story.
Who checks all this is on the author page, and RigSight's code is on GitHub.