pop-os / cosmic-comp

Compositor for the COSMIC desktop environment
GNU General Public License v3.0
443 stars 64 forks source link

Game Performance on Cosmic/Gnome #399

Open XV-02 opened 4 months ago

XV-02 commented 4 months ago

I've been benchmarking gaming performance in both Cosmic DE and Gnome recently.

While system configurations vary notably between individual systems - Discrete AMD desktop, compared to Nvidia discrete desktop, compared to Intel discrete desktop, compared to a hybrid graphics laptop - the run-to-run hardware configuration was the same within any given test series.

Between test-series, the ram configuration may be changed - as QA certifies desktop firmware upgrades for release, some ram configurations needs to be explicitly tested against.

XV-02 commented 4 months ago

This first set of results shows Cyberpunk 2077 and Unigine Superposition benchmarks on a Thelio Spark B1 with an Nvidia RTX 3060Ti.

Also benchmarked in this battery were Xonotic and GLMark2, however, these tests did not seem to execute correctly in Cosmic, having results many times that of Gnome, and are being discarded as invalid at this moment. All benchmarks were run at 1920*1080, and were run at all available fidelity levels provided by the relevant phoronix-test-suite benchmark.

In these tests, the valid benchmarks are all within one percent of each other - and often within a single FPS of each other.

NvidiaBenchmark01

ids1024 commented 4 months ago

Are these benchmarks running with or without vsync? I guess the one that's in the 200s at least must be without vsync.

Vsync is where certain issues may be more apparent. Also latency, but that requires some equipment to measure.

XV-02 commented 4 months ago

This set of results shows Cyberpunk 2077 and Unigine Superposition benchmarks on a Thelio R3 with an AMD RX 6600 XT. I've formatted results as a table due to difficulties getting the phoromatic server to correctly compare test results.

Also benchmarked in this battery were Xonotic and GLMark2, however, these tests did not seem to execute correctly in Cosmic, having results several times that of Gnome, and are being discarded as invalid at this moment. All benchmarks were run at 1920*1080, and were run at all available fidelity levels provided by the relevant phoronix-test-suite benchmark.

In these tests, the valid benchmarks are effectively identical,

Test Gnome Cosmic Performance% (Cosmic/Gnome)
Unigine Super - Low 274.9 273.8 99.6%
Unigine Super - Medium 134.2 133.7 99.6%
Unigine Super - High 95.1 95.0 99.9%
Unigine Super - Ultra 39.3 39.3 100%
CP2077 - Low 129.99 130.37 100%
CP2077 - Medium 100.16 100.47 100%
CP2077 - High 83.52 83.41 99.9%
CP2077 - Ultra 73.69 73.87 100%
XV-02 commented 4 months ago

Okay, another round, with a few more titles. Right now, dedicated Arc graphics test are in the works. For these tables, I'm using a % +/- - positive values indicate better Cosmic performance, negative values indicate that Gnome has an edge.

As a summary: The Hybrid platform test were comparable, as were the dedicated AMD graphics tests. The dedicated Nvidia graphics results showed server regressions in the Unigine benchmarks compared with the last round of tests, though the actual game benchmarks showed a generally consistent results - slight leads in Cyberpunk 2077, slight trailing in Hitman 3 and DiRT 2. Of particular note to me is the variance in the Unigine benchmarks from test to test, as that wasn't exhibited in the game-specific benchmarks.

AMD

Benchmark Cosmic DE Gnome DE % +/-
CP2077 - Low 130.26 129.51 +<1%
CP2077 - Medium 100.30 99.98 +<1%
CP2077 - High 83.42 83.06 +<1%
CP2077 - Ultra 73.59 73.20 +<1%
DiRT 2 - Low 190.99 191.25 -<1%
DiRT 2 - High 190.83 191.14 -<1%
DiRT 2 - Ultra 190.76 191.15 -<1%
Hitman 3 - Low 208.30 209.18 -<1%
Hitman 3 - Medium 205.05 206.09 -<1%
Hitman 3 - Ultra 196.51 197.42 -<1%
Unigine Super - Low 273.2 273.9 -<1%
Unigine Super - Medium 133.6 133.8 -<1%
Unigine Super - High 94.6 94.8 -<1%
Unigine Super - Ultra 39.2 39.2 0%
Unigine Valley 231.65 231.50 +<1%

Nvidia

The Nvidia system was not explicitly leveraging Vsync on any of the benchmarks. DiRT 2 - Low was excerpted due to crashing on Gnome.

Benchmark Cosmic DE Gnome DE % +/-
CP2077 - Low 115.15 114.01 +<1%
CP2077 - Medium 94.81 94.32 +<1%
CP2077 - High 81.26 80.02 +1%
CP2077 - Ultra 71.42 70.89 +<1%
DiRT 2 - Low excerpted N/A
DiRT 2 - High 110.67 114.44 -3%
DiRT 2 - Ultra 110.70 114.58 -3%
Hitman 3 - Low 169.46 176.69 -4%
Hitman 3 - Medium 168.78 175.17 -3%
Hitman 3 - Ultra 162.46 168.60 -3%
Unigine Super - Low 266.2 271.7 -2%
Unigine Super - Medium 139.1 294.0 -111%
Unigine Super - High 103.4 128.5 -24%
Unigine Super - Ultra 45.8 93.2 -103%
Unigine Valley 217.37 263.30 -21%

Hybrid Graphics Laptop

A Serw13 with Nvidia 4070 RTX and Intel i9-13900HX was used.

Benchmark Cosmic DE Gnome DE % +/-
CP2077 - Low 80.74 78.04 +3%
CP2077 - Medium 77.40 76.28 +1%
CP2077 - High 70.19 71.03 -1%
CP2077 - Ultra 70.30 70.68 -<1%
DiRT 2 - Low 127.15 127.16 -<1%
DiRT 2 - High 127.05 127.16 -<1%
DiRT 2 - Ultra 127.14 127.39 -<1%
Hitman 3 - Low 184.29 186.53 -1%
Hitman 3 - Medium 181.98 184.21 -1%
Hitman 3 - Ultra 171.82 171.01 -<1%
Unigine Super - Low 22.1 23.7 -6%
Unigine Super - Medium 11.9 12.3 -3%
Unigine Super - High 8.6 8.8 -2%
Unigine Super - Ultra 3.9 3.9 0%
Unigine Valley 21.08 22.47 -6%

Next steps

First, I am going to immediately verify the unigine results on Nvidia with more runs. Next I'll move into the next round of benchmarks, which should include Arc graphics and 4k benchmarks.

XV-02 commented 4 months ago

After re-running the Unigine benchmarks several times, it looks like the regressions were the result of spurious runs. I'm uncertain of the cause, but subsequent runs were extremely consistent. The same hardware was used, as were the same package versions. Unigine Superposition tests conducted on Ultra were omitted as they crashed before completion 2 out of 3 times on both gnome and cosmic.

Benchmark Cosmic DE Gnome DE % +/-
Unigine Super - Low 268.8 270.9 -<1%
Unigine Super - Low 269.7 271.6 -<1%
Unigine Super - Low 269.1 271.5 -<1%
Unigine Super - Medium 139.2 140.0 -<1%
Unigine Super - Medium 139.3 140.1 -<1%
Unigine Super - Medium 138.9 140.2 -<1%
Unigine Super - High 103.4 104.1 -<1%
Unigine Super - High 103.5 104.2 -<1%
Unigine Super - High 103.7 104.3 -<1%
Unigine Valley 218.89 220.11 -<1%
Unigine Valley 219.34 220.59 -<1%
Unigine Valley 219.33 220.88 -<1%

I'll keep an eye out on the next test run, which will have packages updated to their latest as of 11th April.