Multi-pass GPU kernel test
More passes and a longer timed window than the primary WebGL 2 kernel. Use it to see how a GPU recovers after several minutes of load.
Rated 4.4 out of 5 based on 198 reviews
What this test measures
This page adds pass depth and session length on top of the primary kernel test. Highest pass counts are meant for high-end desktop GPUs. On a phone, start near the default and double until FPS halves.
How to scale the load
- Start near default passes, then double until FPS halves.
- Match timer length to how long you need stability data.
- Desktop cards with cooling headroom tolerate the top of the slider longer.
Graphics adapter
- Renderer
- -
- Vendor
- -
How to run the multi-pass kernel test
- Run the primary WebGL 2 kernel test first so you know a baseline.
- Set a moderate pass count and a 60 second window.
- Press Start and watch FPS over the full timer.
- Increase passes only if the baseline stayed smooth.
- Stop if the device overheats.
Multi-pass kernel
Timed stacked WebGL fragment work
| Compared with | Primary kernel test at /tests/compute/ |
|---|---|
| What you measure | Sustained FPS and stability across extra passes |
When to use extra passes
One kernel draw may still leave headroom. Extra passes keep the GPU busy longer so you can see throttling that a short run hides. This is still numeric fragment work, not a volume shader BM scene.
Heat and timed windows
A longer timer is the point of this page. Use AC power when you can, and stop if the laptop or phone is uncomfortably hot. Wait between runs so you are not comparing a hot chip to a cold one.
Multi-pass questions
Should I skip the primary kernel test?
Can a phone run the top slider?
Is this a volume shader BM test?
Will this damage my hardware?
Results
--
Summary
- -
- -
Details
- -