Report #19 benchmark
Attested, locked device valid
- GPU
- Adreno (TM) 660
- Driver
- turnip_faux123_v1.1.9-rc13 Mesa 26.3.0-devel (git-dd45fd0baa)
- Device
- Samsung Galaxy S21+ 5G
- Chip
- SM8350
- CPU
- 4 x 1.80 GHz + 3 x 2.42 GHz + 1 x 2.84 GHz, arm64-v8a
- Memory
- 7.2 GB
- Reported as
- samsung / SM-G996U1
- Received
- Android
- 14
- Library SHA-256
- 49de278a13c27a6e1ead2293e4e4b2c4475b5289ee67b28ee81c18781023baf3
Conditions
- seconds_per_scene
- 6
- warmup_frames
- 30
- render_width
- 1920
- render_height
- 1080
- heavy_width
- 3840
- heavy_height
- 2160
- heavy_below_ms
- 8
- heavy_particles
- 2
- display_width
- 1080
- display_height
- 2400
- display_refresh_hz
- 120
- rounds
- 3
- scene_set
- 3
- vsync
- OFF
Every scene draws at 1920 by 1080, whatever the screen. A scene under 8.0 ms there is too light for the chip and runs again at a heavier load: at 3840 by 2160, as the row named with _2160p, or, for compute_nbody, with 2 times the particles, as the row named with _2x. Compare drivers on one device; across devices, compare only rows that kept the GPU busy.
Graphics chip below 90% of its top speed (GPU clock) during: compute_nbody, pipeline_barriers_2160p. The device was holding the graphics chip back (heat or a power-saving mode), or the scene is too light to need more. Compare these numbers only with a run at a similar clock.
Scenes
| Scene | Median ms | Average ms | p95 ms | Rounds | Started hot | GPU clock |
|---|---|---|---|---|---|---|
| oit_linked_lists | 30.99 | 30.96 | 31.14 | 3 | 0 | 840 of 840 MHz |
| subpasses | 6.10 | 6.20 | 7.26 | 3 | 0 | 809 of 840 MHz |
| compute_nbody | 13.10 | 12.93 | 13.97 | 3 | 0 | 693 of 840 MHz |
| pipeline_barriers | 3.78 | 3.79 | 4.28 | 3 | 0 | 792 of 840 MHz |
| subpasses_2160p | 23.38 | 23.12 | 24.49 | 3 | 0 | 799 of 840 MHz |
| pipeline_barriers_2160p | 14.32 | 14.21 | 15.32 | 3 | 0 | 755 of 840 MHz |
Individual runs (18)
| Scene | Round | State | Median ms | Frames |
|---|---|---|---|---|
| oit_linked_lists | 0 | ok | 30.99 | 161 |
| subpasses | 0 | ok | 6.10 | 937 |
| compute_nbody | 0 | ok | 13.10 | 433 |
| pipeline_barriers | 0 | ok | 3.78 | 1507 |
| oit_linked_lists | 1 | ok | 31.05 | 161 |
| subpasses | 1 | ok | 6.46 | 899 |
| compute_nbody | 1 | ok | 13.52 | 422 |
| pipeline_barriers | 1 | ok | 3.88 | 1475 |
| oit_linked_lists | 2 | ok | 31.09 | 161 |
| subpasses | 2 | ok | 6.70 | 882 |
| compute_nbody | 2 | ok | 13.15 | 434 |
| pipeline_barriers | 2 | ok | 3.94 | 1466 |
| subpasses_2160p | 0 | ok | 23.38 | 229 |
| pipeline_barriers_2160p | 0 | ok | 14.32 | 372 |
| subpasses_2160p | 1 | ok | 25.22 | 205 |
| pipeline_barriers_2160p | 1 | ok | 17.13 | 302 |
| subpasses_2160p | 2 | ok | 26.98 | 192 |
| pipeline_barriers_2160p | 2 | ok | 15.46 | 346 |