Soft shadows. What a softer edge actually costs.
A real percentage-closer soft shadow test: the same scene with hard shadows and with PCSS, alternated so heat cannot favour either, timed frame by frame on your graphics chip.
Illustrative figures, not a measurement. The test reports your own.
Technique
PCSS, 16 + 32 samples
Shadow map
2048 × 2048
Resolution
Fixed 1600 × 900
Duration
About 60 seconds
Three steps a pixel, every frame.
Percentage-closer soft shadows make an edge sharp where an object meets the ground and soft where it is far above it, the way a real light of some size behaves. That realism is paid for per pixel, and that price is what the test measures.
- Find what blocks the light
- From each pixel on screen, 16 samples of the shadow map, spread over a region scaled to the light's size, find the average depth of whatever sits between that pixel and the light.
- Estimate the penumbra
- The gap between that blocker and the surface, relative to the blocker's distance from the light, sets how wide the soft edge should be. Contact points get a sharp edge; high blockers a wide one.
- Filter at that width
- 32 more samples, rotated per pixel so the pattern does not band, are taken over a region that wide and averaged. This filtering is most of the cost.
- Hard and soft, in a balanced order
- After a 6-second warm-up the scene runs six 9-second periods in the order soft, hard, hard, soft, soft, hard, 3 of each, on one camera path. That order puts each mode at the same average point in the run, so steady warming cancels instead of favouring the mode that ran first. Both render the same shadow map, so the difference between them is the soft-shadow filtering alone.
What it costs, and whether it holds.
Four numbers, from frames rendered off screen and timed until the graphics chip had finished them. The picture you see is refreshed about 30 times a second, outside the timing. A hidden tab pauses the test and its clock. Your graphics chip's frame rate with real-time soft shadows and with hard ones, what the soft edge costs as a share of the hard-shadow rate with its spread, and whether that rate holds. Nothing is uploaded.
PCSS frame rate
Frames per second with soft shadows on: the median of the best five one-second windows in the PCSS periods. If frame times sit at the timer's resolution or a display refresh interval, the result says the figure is a lower bound.
Soft-shadow cost
How much frame rate soft shadows take away: one minus the PCSS rate over the hard-shadow rate, from whole periods. Because the periods are in a balanced order, a machine that heats up during the test slows both modes by the same amount, and the spread between the three pairs is printed beside the cost. If either mode sits at a timer or display-refresh floor, the result says whether the cost is at least or at most the figure shown.
Stability
The share of one-second PCSS windows within 85% of the median window. A card that stutters under soft shadows scores lower than its frame rate alone would suggest.
Decay
The first PCSS period against the last one, thirty-six seconds later. A drop smaller than two and a half times the test's own noise reports as 0%.
Illustrative figures, not a measurement. The test reports your own.
Hardware
What it needs. Three tiers, no fine print.
Minimum
Any GPU with WebGL2
Runs. On weak integrated graphics the PCSS frame rate can be low, and the cost figure is then the useful number.
Recommended
Recent integrated or entry dedicated
Enough frames in every period for stability and decay to be read with confidence.
Optimal
Any modern dedicated GPU
May outrun the timer or the display refresh; the result then says the frame rate is a lower bound.
Questions
Soft shadows, answered
- What is PCSS?
- Percentage-closer soft shadows, a technique from NVIDIA research that makes shadow edges soften with distance from the object casting them. It works on an ordinary shadow map in three steps: find the average depth of whatever blocks the light, estimate the width of the soft edge from it, then filter the shadow map at that width.
- Is this part of the main SystemCheck run?
- It is a stage of the nine-minute Deep run, after the combined CPU and GPU load, and it is reported there without changing the score. The three-minute Standard run does not include it. The test on this page is the same workload, run on its own.
- Why compare against hard shadows at all?
- A frame rate on its own mixes the shadow technique with everything else the card is drawing. Rendering the identical scene with a single hard-shadow lookup isolates what the soft edge costs, which is the question a game's shadow quality setting is really asking.
- Can I compare my result with someone else's?
- Only directionally. The workload is fixed, but browsers time the GPU differently: some expose a GPU timer query and others fall back to a readback sync point, and drivers differ. Comparisons between your own runs in the same browser are the reliable ones.