Rockchip · AI & Linux Hub

Linux & boot

Rockchip Linux Benchmark KPI

Rockchip Linux Benchmark KPI — RK3588/RK3576/RK3562/RK3568/RK3566/RK3399/RK3326/RK3288, V1.3.0

· ArmSoM Editorial · Rockchip source documentation · 5 min read

ArmSoM platform reference based on Chinese and English RK-CS-YF-375 V1.3.0. All technical content and benchmark data for eight chips are retained for comparison during further development. These are Rockchip EVB results, not ArmSoM measurements or guarantees for other boards. For an ArmSoM board, verify SoC, memory, storage, SDK/BSP, kernel, cooling and display configuration. No hardware validation was performed.

1. Test environment

The table specifies hardware and software, including ARM chip, CPU core count, memory capacity and storage configuration. G/M notation for frequency and capacity is preserved from the source rather than silently converted.

Chip Rockchip EVB board Operating system SDK tag Hardware configuration
RK3588 rk3588_evb1_lp4_v10 Buildroot linux-5.10-gen-rkr4 CPU: eight cores, big 2.3G / little 1.8G; DDR: 2112M, 8G; GPU: 1G; FLASH: 32G eMMC
RK3576 rk3576_evb1_v10 Buildroot linux-6.1-stan-rkr4 CPU: eight cores, big 2.2G / little 2.0G; DDR: 2112M, 4G; GPU: 950MHz; NPU: 900MHz; FLASH: 256G UFS
RK3562 rk3562_evb2_ddr4_v10 Buildroot linux-5.10-gen-rkr4 CPU: 4-core A53, 2016M; DDR: 1332M, 2G; GPU: 900M; NPU: 1000MHz; FLASH: 256G eMMC
RK3568 rk3568_evb1_ddr4_v10 Buildroot linux-5.10-gen-rkr4 CPU: 4-core A55, 1992M; DDR: 1560M, 2G; GPU: 800M; NPU: 900MHz; FLASH: 32G eMMC
RK3566 rk3566_evb2_lp4x_v10 Buildroot linux-5.10-gen-rkr4 CPU: 4-core A55, 1.8G; DDR: 1056M, 2G; GPU: 800M; NPU: 900MHz; FLASH: 16G eMMC
RK3399 rk3399_evb_ind_lpddr4 Buildroot linux-5.10-gen-rkr4 CPU: six cores, big 2xA72 at 1.8G, little 4XA53 at 1.4G; DDR: 856M, 4G; GPU: 800M; FLASH: 16G eMMC
RK3326 rk3326_evb_lp3_v12 Buildroot linux-5.10-gen-rkr4 CPU: 4-core A35, 1.3G; DDR: 666M, 2G; GPU: 520M; FLASH: 8G eMMC
RK3288 rk3288_evb_rk808 Buildroot linux-5.10-gen-rkr4 CPU: 4-core A17, 1.6G; DDR: 528M, 2G; GPU: 600M; FLASH: 8G eMMC

2. Testing methods

Document the test methods, tools and settings for each item in enough detail to reproduce the tests. The source sets system performance mode with the command below. It changes governors, is not read-only, and was not executed for this article.

echo performance | tee $(find /sys/ -name *governor) /dev/null || true

2.1 Glmark2 testing

Execute the following command:

/rockchip-test/gpu/gpu_test.sh

2.2 UnixBench testing

Execute the following command:

/opt/unixbench/Run

3. Test results

3.1 Buildroot system

3.1.1 Glmark2

glmark2 benchmarks OpenGL 2.0 and ES 2.0. Normal, off-screen and full-screen scores follow.

Item RK3588 RK3576 RK3562 RK3568 RK3566 RK3399 RK3326 RK3288
Normal Screen (800x600) 4851 1713 676 560 485 812 369 57
Off Screen 3550 1864 644 596 531 687 419 604
Full Screen 2780 788 287 240 224 183 253 52
GPU Type Valhall-G610 Mali-G52 Mali-G52 Mali-G52 Mali-G52 Mali-T864 Mali-G31 Mali-T764
Screen resolution 1080x1920 1080x1920 1080x1920 1080x1920 1080x1920 1536x2048 720x1280 1536x2048

3.1.2 UnixBench

UnixBench provides basic performance indicators for Unix-like systems. Reference values follow. The source does not give units for each row; the values are not silently relabeled with one common throughput unit.

System performance with a single task:

Item RK3399 RK3288 RK3326 RK3566 RK3568 RK3562 RK3576 RK3588
Dhrystone 2 using register variables 19191210.4 10626086.6 5704897.1 12332588.0 13176039.3 10821835.3 22932325.5 34493007.3
Double-Precision Whetstone 3303.5 1718.9 1565.2 2965.9 3164.5 2464.3 4103.7 6665.4
Execl Throughput 2730.8 1538.1 787.9 1483.6 1703.0 1815.3 2678.3 5695.8
File Copy 1024 bufsize 2000 maxblocks 263262.3 163001.5 125333.2 166135.4 175490.6 247437.6 432878.6 832875.7
File Copy 256 bufsize 500 maxblocks 98335.8 50635.1 37871.9 48956.5 51574.7 74740.7 145359.2 253871.4
File Copy 4096 bufsize 8000 maxblocks 677993.2 384632.9 321189.7 440209.9 461129.3 610768.8 1098724.6 1831983.9
Pipe Throughput 775302.3 357578.5 300305.5 521804.1 557997.0 619105.4 779828.1 1644863.6
Pipe-based Context Switching 87345.3 54247.5 37434.5 51766.2 53873.8 61437.0 105369.5 217180.2
Process Creation 4274.2 3512.1 2086.0 3782.4 4041.7 5113.1 4232.7 5272.1
Shell Scripts (1 concurrent) 2944.0 2973.3 1474.2 2352.0 2817.2 2301.5 2598.5 3254.2
Shell Scripts (8 concurrent) 832.4 703.2 431.7 567.3 675.6 647.6 976.8 2092.6
System Call Overhead 721899.8 624614.1 568868.6 783414.7 836985.7 838824.4 621437.6 1322359.7
System Benchmarks Index Score 654.7 421.7 290.6 456.9 497.3 533.3 761.6 1342.5

System performance with multiple tasks:

Item RK3399 RK3288 RK3326 RK3566 RK3568 RK3562 RK3576 RK3588
Dhrystone 2 using register variables 61892645.4 41527276.3 22821903.2 47931915.4 51737187.8 42342893.2 133846256.4 182453805.7
Double-Precision Whetstone 13192.2 6870.1 6265.7 11545.6 12431.5 9699.1 27875.6 38815.5
Execl Throughput 6638.9 4127.0 2449.4 3272.1 3951.1 4359.4 10082.5 23542.0
File Copy 1024 bufsize 2000 maxblocks 253903.6 265838.2 194293.5 236042.5 246968.0 436766.8 756330.5 681901.0
File Copy 256 bufsize 500 maxblocks 74647.0 74156.4 54107.4 65182.9 67873.4 117806.2 225876.4 189411.0
File Copy 4096 bufsize 8000 maxblocks 715699.5 709343.4 565091.8 639035.4 700693.2 1285002.9 1999953.4 1873825.2
Pipe Throughput 3159789.0 1323176.0 1191104.7 2031168.7 2191846.5 2427332.4 5664856.1 8951711.8
Pipe-based Context Switching 298324.6 134686.9 154652.2 202912.5 212649.7 306251.1 707269.5 1148297.7
Process Creation 11834.7 7412.8 5183.1 7036.5 7932.7 10979.8 17803.4 44316.5
Shell Scripts (1 concurrent) 7420.0 5710.9 3587.9 4613.8 5473.0 5027.4 9575.2 24379.9
Shell Scripts (8 concurrent) 952.5 744.6 477.4 593.1 707.7 688.4 1279.3 3455.2
System Call Overhead 2514699.7 2392234.7 2206337.3 2815382.8 3028273.6 3152888.0 3589591.2 3030469.3
System Benchmarks Index Score 1402.8 989.5 746.4 1039.1 1146.5 1365.4 2715.3 4036.2

Source reconciliation

Both sources are V1.3.0. The eight hardware configurations and every results cell agree; pagination differs and the environment table is joined across pages. GPU source spellings such as Valhall-G610 are retained, not corrected from assumptions. Neither source contains technical figures; all tables are editable. Test variability, repeat counts and confidence intervals are not specified and are not invented.

Official references: Rockchip; ArmSoM documentation.

Explore this topic → · Learning paths →