Skip to content

Latest commit

 

History

History
249 lines (234 loc) · 41.8 KB

Results.md

File metadata and controls

249 lines (234 loc) · 41.8 KB
Raw

Results

Below some results collected. Please keep in mind that these are NOT hardware performance numbers but depend on software/settings (see the differences kernel version makes for RockPro64 for example). The purpose of sbc-bench is to generate insights and not colorful graphs representing numbers without meaning. It's perfectly fine for the same hardware appearing multiple times with different numbers since those differ for a reason (software/settings).

Especially openssl numbers should be taken with a huge grain of salt since the benchmark numbers depend on kernel features and performance with other use cases (e.g. disk/filesystem encryption) might look differently.

So do not rely on collected numbers unless you carefully read through all the explanations and insights below and be prepared to conduct your own benchmarks if you really want to choose appropriate hardware for your use case.

Some numbers

ODROID-M1, Quartz64, ROCK 3A, ROCK 5B, RK3568-ROC-PC and Khadas VIM4 numbers are preliminary since software support situation for RK3566/RK3568/RK3588 and A311D2 is still in a very early stage. Same applies to Apple M1 Pro. Please also note that with RK35xx SBC so far measured clockspeeds differ from what's defined in device-tree due to PVTM.

Device / details Clockspeed Kernel Distro 7-zip AES-128 (16 byte) AES-256 (16 KB) memcpy memset kH/s
Akaso M8S 1200 MHz 5.10 Buster armhf 3050 32050 32120 1160 3330 -
Amazon a1.xlarge 2300 MHz 4.15 Bionic arm64 8610 458500 1297960 4280 14220 -
AMedia X96 Max+ 2100 MHz 5.15 Focal arm64 5270 197690 981830 2630 5150 -
Apple M1 Pro 3030/2060 MHz 5.18 Gentoo 2.8 arm64 43800 620960 1064450 27110 71910 48.28
BPi M2U 1010 Mhz 5.16 Buster armhf 2230 15550 19540 790 2540 -
BPi M4 1400 MHz 4.9 Bionic arm64 3500 125430 651460 1010 4360 5.48
BPi R2 1300 MHz 4.4 Xenial armhf 2600 27550 25350 1500 3800 -
Clearfog Pro 1600 MHz 4.14 Stretch armhf 2185 44500 43900 935 4940 -
ClockworkPi R-01 1000 MHz 5.4 Focal riscv64 450 8900 9040 1220 2640 -
Cubietruck 960 MHz 5.10 Bullseye armhf 1030 15260 18640 440 2010 -
Cubox-i4 980 MHz 5.15 Jammy armhf 2360 25750 27000 340 340 -
EspressoBin 800 MHz 4.17 Stretch arm64 1138 54290 368330 1040 2490 1.23
EspressoBin 1200 MHz 4.18 Stretch arm64 1630 81900 544240 1000 2400 1.82
Gigabyte H270-T70 2000 Mhz 5.16 Sid arm64 107180 110340 340750 4180 17130 -
Helios4 1600 MHz 4.14 Stretch armhf 2210 44785 *1280 42500 *98560 910 4840 -
Honeycomb LX2 2200 MHz 5.16 Fedora 35 aarch64 30690 418100 1251710 5050 16220 46.09
Hugsun X99 2088/1800 MHz 5.9 Focal arm64 7710 412105 1184306 2270 5970 -
HummingBoard Pulse i.MX8M Quad 1500 MHz 4.19 Buster arm64 4330 97380 695540 2230 9900 -
Jetson Nano 1430 MHz 4.9 Bionic arm64 5060 276890 513700 3680 8560 6.64
Jetson Nano 2000 MHz 4.9 Bionic arm64 6260 376900 717500 4100 11760 8.72
Jetson Xavier NX 1890 MHz 4.9 Bionic arm64 13230 992690 706280 9190 18480 -
Kendryte K510 790 MHz 4.17 Sid riscv64 690 6750 7410 280 440 -
Khadas Edge/Captain 2000/1500 MHz 4.4 Bionic arm64 6550 402150 1130400 2810 4860 10.50
Khadas Edge/Captain 2000/1500 MHz 4.4 Stretch arm64 6600 269485 1127780 2860 4880 8.85
Khadas VIM1 1415 MHz 5.1 Buster arm64 3860 90160 659460 1930 5900 -
Khadas VIM2 1415/1000 MHz 4.9 Xenial arm64 4800 177600 659000 1690 5610 -
Khadas VIM2 1415/1000 MHz 4.17 Bionic arm64 5450 126770 659600 1920 5920 8.59
Khadas VIM3 2200/1800 MHz 4.9 Bionic arm64 8600 398370 1256910 4980 9300 13.12
Khadas VIM3 2400/2015 MHz 5.10 Bullseye arm64 9650 365710 1366350 4850 7380 -
Khadas VIM3 2400/2015 MHz 5.10 Focal arm64 9760 405960 1365900 4840 8260 -
Khadas VIM3L 1900 MHz 4.9 Bionic arm64 5160 194360 892110 3670 6360 7.29
Khadas VIM3L 1900 MHz 5.16 Bullseye arm64 5110 174110 890730 3700 5140 -
Khadas VIM4 2200/1970 MHz 5.4 Focal arm64 12090 372300 1253200 7810 11600 -
Khadas VIM4 2200/2010 MHz 5.4 Jammy arm64 12100 339708 1252070 7800 11590 -
Le Potato 1410 MHz 4.18 Stretch arm64 3780 96680 657200 1810 5730 3.92
LeMaker Banana Pi 960 MHz 5.10 Bullseye armhf 1040 15080 18640 440 2020 -
Libre Computer AML-S912-PC 1415/1000 MHz 5.15 Bullseye arm64 5980 91890 659290 1650 5170 -
Lime A10 910 MHz 4.14 Stretch armhf 550 25200 28250 440 1300 -
Marvell PXA1908 1245 MHz 3.14 Bullseye arm64 3180 81260 581840 740 2220 -
MT6580 K9M1 1300 MHz 5.19 Sid armhf 2930 21580 25300 1250 3300 -
NanoPC T3+ 1400 MHz 4.4 Xenial armhf 6400 143800 651000 1650 3700 -
NanoPC T3+ 1400 MHz 4.14 Bionic arm64 7480 126000 652600 1440 4540 10.99
NanoPC T4 1800/1400 MHz 4.17 Stretch arm64 6250 307200 1022500 4100 9000 8.24
NanoPC T4 1800/1400 MHz 4.17 Stretch arm64 6380 230280 1022600 4160 9000 9.36
NanoPC T4 1800/1400 MHz 4.17 Stretch arm64 6230 299600 1023600 4100 9060 10.30
NanoPC T4 2000/1500 MHz 4.4 Stretch arm64 5870 308370 1124040 2810 4890 8.70
NanoPi Fire3 1380 MHz 4.14 Stretch arm64 7420 95700 645400 1520 4570 8.53
NanoPi Fire3 1400 MHz 4.14 Bionic arm64 7440 126050 653000 1560 4600 10.96
NanoPi Fire3 1400 MHz 4.14 Focal arm64 7350 118250 652640 1530 4590 11.18
NanoPi K1 Plus 1370 MHz 5.10 Focal arm64 3520 115020 638880 1070 3680 5.50
NanoPi K2 1480 MHz 4.14 Stretch arm64 3850 43020 50370 1660 3870 4.61
NanoPi K2 1480 MHz 5.10 Bullseye arm64 3880 42840 51490 1850 3790 -
NanoPi M1 Plus 1370 MHz 4.19 Bionic armhf 3030 25720 26660 830 3450 -
NanoPi M4 2000/1500 MHz 4.19 Stretch arm64 6400 334650 1128330 4080 8270 8.86
NanoPi M4v2 2015/1510 MHz 5.10 Bullseye arm64 6680 230410 921980 3110 7640 -
NanoPi NEO4 2000/1500 MHz 4.4 Stretch arm64 6510 320600 1128860 2260 4770 8.71
NanoPi NEO4 2000/1500 MHz 4.4 Stretch arm64 6030 342620 1121380 2230 4770 8.57
NanoPi NEO4 2000/1500 MHz 4.4 Stretch arm64 6520 268720 1123190 2280 4770 8.83
NanoPi NEO4 2000/1500 MHz 4.19 Stretch arm64 6750 278200 1139850 2370 6110 8.84
NanoPi NEO4 2016/1512 MHz 5.10 Focal arm64 6970 383690 1145030 2450 6190 11.36
Nintendo Switch 1780 MHz 4.9 Fedora 30 arm-64 6170 346340 642670 2500 3570 -
Nintendo Switch 2090 MHz 4.9 Bionic arm64 6720 389030 746680 2370 3670 9.25
ODROID-C2 1750 MHz 3.14 Xenial arm64 4070 50500 48500 1750 3100 -
ODROID-C2 1530 MHz 5.10 Bullseye arm64 4010 44090 51490 1600 2730 -
ODROID-C4 2010 MHz 4.9 Focal arm64 5450 195130 941590 3310 6270 7.71
ODROID-C4 2100 MHz 5.10 Buster arm64 5770 189990 981940 3540 5150 -
ODROID-HC4 2100 MHz 5.10 Buster arm64 5730 191560 980970 3540 5150 -
ODROID-M1 1930 MHz 4.19 Focal arm64 5010 185660 898610 3070 6220 7.14
ODROID-N2 1800/1900 MHz 4.9 Bionic arm64 8140 324900 1024680 4120 8610 11.39
ODROID-N2 2000/1900 MHz 5.10 Buster arm64 9090 298890 1085350 4260 9080 -
ODROID-N2+ 2400/2015 MHz 5.10 Focal arm64 9680 381930 1366730 4030 7120 -
ODROID-N2+ 2400/2015 MHz 5.14 Impish arm64 9790 421730 1366930 4300 7480 -
ODROID-N2+ 2400/2015 MHz 5.14 Hirsute arm64 9780 421750 1366090 4030 7120 -
ODROID-XU4 1900/1400 MHz 3.10 Jessie armhf 6750 74100 68200 2200 4800 -
ODROID-XU4 2000/1400 MHz 4.9 Stretch armhf 6400 73350 72075 2230 4850 -
ODROID-XU4 2000/1400 MHz 5.4 Focal armhf 8980 72450 72020 2280 4910 -
Olimex A20-Lime2 960 MHz 5.10 Bullseye armhf 1040 15280 18670 460 2020 -
Orange Pi PC Plus 1370 MHz 5.10 Focal armhf 3060 25740 26590 890 3450 -
Orange Pi PC 2 1370 MHz 5.10 Focal arm64 3500 114830 637410 1070 3680 -
Orange Pi Plus 2 1300 MHz 4.14 Stretch armhf 2890 21480 25250 830 3240 -
Orange Pi Prime 1370 MHz 5.4 Buster 3590 89210 637980 1180 3540 -
Phytium D2000 1xSO-DIMM 2300 MHz 5.19 Jammy arm64 16390 427440 827090 2820 6490 -
Phytium D2000 2xSO-DIMM 2300 MHz 5.19 Jammy arm64 16670 427490 828130 3480 16110 -
PineH64 1800 MHz 4.18 Stretch arm64 4650 123400 836900 1380 5530 5.62
PineH64 1800 MHz 5.4 Buster arm64 4710 116900 839870 1420 5560 7.10
Quartz64 1800 MHz 5.13 Buster arm64 4840 165250 845490 2980 7650 -
Radxa ROCK 3A 2000 MHz 5.18 Bullseye arm64 5110 181400 935920 3150 6250 7.58
Radxa ROCK 5B 2350/1830 MHz 5.10 Focal arm64 16450 683350 1337540 10830 29220 25.31
Radxa Zero 1800 MHz 5.10 Focal arm64 4610 151370 840080 1600 5370 -
Radxa Zero 1800 MHz 5.10 Buster arm64 4570 114530 839080 1610 5250 6.82
Radxa Zero 1800 MHz 5.10 Bullseye arm64 4580 105170 838360 1600 5360 7.13
Renegade 1400 MHz 4.4 Stretch arm64 3710 95030 644200 1565 7435 3.92
Raspberry Pi B 700 MHz 5.10 Raspberry Pi OS Buster 310 5900 11310 340 1400 -
Raspberry Pi B 700 MHz 5.10 Raspberry Pi OS Bullseye 320 7900 11630 360 1420 -
Raspberry Pi Zero 1000 MHz 5.10 Bullseye armv6l/armhf 470 10450 17060 430 1670 -
Raspberry Pi 2 B+ 900 MHz 4.14 Debian Stretch 2070 14350 17450 615 1175 -
Raspberry Pi 2 B+ 900 MHz 5.10 Raspberry Pi OS Bullseye 2150 13830 16500 1000 1180 -
Raspberry Pi 3 B+ original 4.9 Raspbian Stretch 3600 35500 42700 1230 1640 -
Raspberry Pi 3 B+ normal 4.14 Raspbian Stretch 3240 30500 36600 1130 1530 -
Raspberry Pi 3 B+ normal 4.14 Raspbian Stretch 3040 29500 36600 1050 1500 -
Raspberry Pi 3 B+ UV/normal 4.14 Raspbian Stretch 2100 29500 36400 1040 1460 -
Raspberry Pi 3 B+ OC/normal 4.14 Raspbian Stretch 3130 30500 36620 1230 1780 -
Raspberry Pi 3 B+ with fan 4.14 Raspbian Stretch 3670 35800 42600 1120 1600 -
Raspberry Pi Zero 1000 MHz 4.14 Raspbian Stretch 450 13400 16820 400 1590 -
Raspberry Pi Zero 2 600 MHz 5.10 Raspberry Pi OS Buster 1900 14470 18150 1040 1130 -
Raspberry Pi Zero 2 1000 MHz 5.10 Raspberry Pi OS Buster 3030 23860 29860 1300 1570 -
Raspberry Pi Zero 2 1200 MHz 5.10 Raspberry Pi OS Buster 3640 29120 36300 1320 1790 -
Raspberry Pi 4 B 1500 MHz 4.19 Raspbian Buster 5500 62350 64860 2460 3170 -
Raspberry Pi 4 B 1500 MHz 5.10 Raspberry Pi OS Buster 5750 62310 64930 2550 3430 -
Raspberry Pi 4 B 1800 MHz 5.10 Raspberry Pi OS Buster 6550 75040 77830 2780 3080 -
Raspberry Pi 4 B 1800 MHz 5.10 Raspberry Pi OS Bullseye 5940 74620 77670 2310 2690 -
Raspberry Pi 4 B 1800 MHz 5.10 Raspberry Pi OS Buster arm64 5760 45570 36240 2240 3120 9.46
Raspberry Pi 4 B 1800 MHz 5.10 Raspberry Pi OS Bullseye arm64 arm64 5790 45120 36260 2330 3120 8.74
Raspberry Pi 4 B 1800 MHz 5.15 Armbian Jammy arm64 5640 45210 36260 2580 3110 -
Raspberry Pi 4 B 1800 MHz 5.15 Jammy armhf 6300 77160 82750 1190 3110 -
Raspberry Pi 400 1800 MHz 5.4 Raspberry Pi OS Buster 6550 75050 77890 2680 3110 -
RK3228A TV Box 1200 MHz 4.4 Buster armhf 2310 18520 23070 410 1230 -
RK3568-ROC-PC 1960 MHz 4.19 Bullseye arm64 5040 179090 912800 3130 6240 -
RK3318 BOX 1300 MHz 5.15 Bullseye arm64 3120 84160 603700 700 2510 -
Rock64 1300 MHz 4.4 Bionic arm64 3410 89060 601200 1310 5680 4.46
Rock64 1300 MHz 4.18 Bionic arm64 3530 116100 605250 1340 5770 4.65
Rock64 1300 MHz 4.4 Stretch arm64 3430 88600 601000 1350 5680 3.64
Rock64 1300 MHz 4.18 Stretch arm64 3560 89070 603800 1340 5770 3.80
Rock64 1400 MHz 4.4 Stretch arm64 3610 95000 644250 1330 5700 3.85
Rock64 1400 MHz 4.4 Stretch arm64 3590 95000 643700 1320 5640 4.40
Rock64 1400 MHz 4.4 Stretch arm64 3580 94800 644380 1330 5680 4.63
Rock64 1400 MHz 4.4 Stretch armhf 3620 99400 624000 1430 3620 -
Rock Pi 4 2000/1500 MHz 4.4 Stretch armhf ~6450 301470 1113900 1870 4860 -
Rock Pi 4 2000/1500 MHz 5.3 Bionic arm64 6910 402750 1147370 3660 8310 10.71
Rock Pi 4 2000/1500 MHz 5.10 Focal arm64 6900 383680 1146500 3430 8260 -
Rock Pi S 1300 MHz 4.4 Buster 2590 68740 282290 830 1880 -
RockPro64 1800/1400 MHz 4.4 Stretch arm64 6140 298800 1015600 2770 4850 8.14
RockPro64 1800/1400 MHz 4.4 Stretch armhf 6250 275000 1000150 2000 4835 -
RockPro64 1800/1400 MHz 4.4 Stretch arm64 6420 312690 1018480 3720 8400 8.24
RockPro64 1800/1400 MHz 4.18 Stretch arm64 6300 237700 1021500 3650 8450 8.20
RockPro64 2010/1510 MHz 5.4 Focal arm64 6920 397740 1145300 3700 8430 11.55
RockPro64 2010/1510 MHz 5.8 Bullseye arm64 7000 369820 1144950 3690 8360 11.08
SBC2D70 (SSD202D) 1700 MHz 5.16 Sid armhf 1960 26240 33120 770 3190 -
Qualcomm QRB5165 2830/2400/1750 MHz 4.19 Focal arm64 17930 1121460 1589900 14610 25590 25.38
T-HEAD C910 RVB-ICE 1200 MHz 5.10 Sid riscv64 1760 24420 26930 3340 6470 -
Teres-I 1050 MHz 4.19 Stretch arm64 2785 72261 491590 1080 2820 -
Tinkerboard 1730 MHz 4.14 Stretch armhf 5350 63150 66600 1480 3900 -
Tinkerboard 1800 MHz 4.4 Buster armhf 5440 62410 66300 1340 3510 -
Tinkerboard 1800 MHz 5.10 Buster armhf 5770 64100 67060 1540 4110 -
TRONFY MXQ S805 1536 MHz 5.10 Focal armhf 3100 26250 29080 980 2990 -
Tronsmart MXIII Plus 1560 MHz 5.10 Buster armhf 3880 42850 42570 1470 3430 -
Tronsmart S82 1600 MHz 5.14 Focal armhf 3640 43850 43150 500 3200 -
Ugoos UT2 1560 MHz 5.10 Jammy armhf 3320 41960 43250 320 2020 -
Atom N270 1600 MHz 4.19 Buster i386 1220 19810 18760 1420 2840 -
Atom E3826 1460 MHz 5.18 Jammy amd64 2140 98300 182190 2840 2760 -
x5-Z8300 1420 MHz 4.9 Stretch amd64 3900 101580 178010 2380 2380 7.81
x5-Z8350 1920/1680 MHz 4.15 Bionic amd64 4710 120300 207640 2740 3140 -
x5-Z8350 1920/1680 MHz 5.4 Focal amd64 4790 129810 237230 3170 2960 9.38
Celeron J1900 2000/1333 MHz 5.4 Focal amd64 5530 34060 28860 3550 3400 -
Celeron J3455 2300/1500 MHz 4.17 Stretch amd64 7000 316480 429660 4090 4050 17.26
Pentium N4200 2560/1100 MHz 4.14 Bionic amd64 7469 354328 468008 4682 4997 18.75
Pentium J4205 2560/1500 MHz 4.17 Stretch amd64 7570 355540 480640 5070 5170 18.82
Ryzen R1606G 2600/1400 MHz 5.4 Focal amd64 7970 560110 700780 8230 5970 16.45
Celeron N4100 2300/1100 MHz 4.15 Bionic amd64 8510 435030 669350 4750 5240 18.33
Celeron J4105 2400/1500 MHz 4.15 Bionic amd64 9020 458670 697100 5500 7410 19.07
Celeron J4105 2400/1500 MHz 4.15 Bionic amd64 8960 453860 697080 5620 7650 19.13
Pentium J5005 2700/1500 MHz 5.0 Bionic amd64 9230 379740 778360 5530 7130 20.74
Celeron N4500 2800/1100 MHz 5.13 Impish amd64 6300 663550 783840 8100 8350 -
Celeron N5100 2800/1100 MHz 5.13 Focal amd64 10550 661820 783800 7750 8090 19.22
Celeron N5105 2900/2000 MHz 5.13 Focal amd64 11450 632410 811760 7710 9290 21.79
Pentium G4600 3600 MHz 4.19 Buster amd64 11810 878790 984820 15120 33380 21.88

* Number obtained with cryptodev (Marvell's CESA).

Explanations

  • 7-zip number is an averaged multi threaded score from 3 consecutive 7z b runs. Only relevant for server workloads where stuff happens in parallel. Check the links for single threaded results (on big.LITTLE SoCs individually) to get an idea how most typical (single threaded) workloads perform. Attention: all single-threaded 7-zip results made prior to v0.8.5 are somewhat flawed since execution happened pinned to a single core but with as much threads as cores available. With up to 4 cores the effect is negligible but with 6 or 8 cores the difference in scores is up to 7%. That's why all sbc-bench results made with CPUs featuring more than 8 cores have been removed in Nov 2021 and need to be resubmitted.
  • AES-128 (16 byte) is a single threaded encryption score with very small chunks of data (useful to get an idea how initialization overhead influences crypto performance with small packets). On big.LITTLE SoCs numbers show big core performance
  • AES-256 (16 KB) is a single threaded encryption score with rather huge chunks of data. On big.LITTLE SoCs numbers show big core performance. In case an ARM SoC supports ARMv8 Crypto Extensions scores are pretty much predictable based on CPU clockspeeds.
  • memcpy and memset are tinymembench measurements for memory bandwidth. On big.LITTLE SoCs numbers show big core performance
  • kH/s is a multi threaded cpuminer score showing the board's performance when executing NEON optimized code. To get the performance difference between big and little cores click the links in the left column
  • The Akaso M8S and Tronsmart MXIII Plus numbers may be representative for other Amlogic S812 devices (quad Cortex-A9 @ 1.2/1.55 GHz), Tronsmart S82 for other S802 devices (quad Cortex-A9 @ 1.6 GHz)
  • The Amazon a1.xlarge numbers represent a 1st gen Graviton CPU (64-bit 'ARM Neoverse') limited to four A72 cores and 8GB memory.
  • The 1st Apple M1 number represents a single core Ubuntu 20.10 VM running on an MacBook Air with the VM executed almost all the time on one of Apple's ARMv8.4 power cores. The 2nd number is the same VM setup this time sent to an efficiency core clocking in at just 600 MHz instead of the usual 2064 MHz. Running natively and not inside a VM performance numbers are 5% to 10% higher.
  • Cubietruck and 'LeMaker Banana Pi' numbers are more or less representative for all other Allwinner A20 devices, same with Lime for Allwinner A10, Olimex Teres-I for Allwinner A64, Orange Pi "PC Plus" and "Plus 2" for Allwinner H2+/H3 and NanoPi K1 Plus, Orange Pi "PC 2" and Prime for Allwinner H5. AMedia X96 Max+ numbers represent Amlogic S905X3 devices.
  • Honeycomb LX2 numbers (based on SolidRun's CEx7 LX2160A COM) might vary somewhat with memory configuration but are more or less representative for LX2160A in general.
  • Clearfog Pro and Helios4 use exactly same SoC (Armada 385), kernel and clockspeeds and the only reason why OpenSSL numbers differ is since Helios4 numbers were made using Marvell's CESA crypto accelerator via cryptodev which provides nice speed improvements with larger block sizes but also some initialization overhead with tiny block sizes. Also CPU utilization is way lower so the SoC is free for other stuff while performing better at the same time.
  • EspressoBin's boot BLOB claims to run at up to 1GHz while real clockspeeds are lower maxing out with this setting at 790MHz (obviously a kernel bug -- see details)
  • Gigabyte H270-T70 numbers are for one blade module equipped with two Cavium ThunderX CN8890 (48 cores each). With different DRAM config/settings results vary (see here and there).
  • Hugsun X99 is an overclocked RK3399 TV box, just to show the effect of overclocking the A53 cores to 1.8 GHz and the A72 to 2.1 GHz on an RK3399.
  • Jetson Nano was properly powered with 5V/5A via barrel plug (when powering through Micro USB the board enters a lower consumption/performance profile)
  • Phytium D2000 consists of 8 custom 64-bit ARMv8-compatible FTC663 cores. The 2 numbers above only differ by RAM config: one or two 16GB DDR4 SO-DIMMs. While memory bandwidth differs significantly latency does not and as such the 7-zip benchmark is almost unaffected while the openssl test does not depend on memory performance by design. Though other tasks that are sensitive to memory bandwidth might benefit a lot from a dual channel memory config.
  • NanoPi NEO4 numbers: 1st result is from my NEO4 N°1 running with a NanoPi M4 image. This NEO uses the vendor supplied thermal pad between SoC and heatsink. 2nd number from my 2nd NEO4 this time using NEO4 settings (rk3399-nanopi4-rev04.dtb loaded) with a copper shim between heatsink and SoC which as usual improves 'thermal performance' a lot. Since memory bandwidth and especially latency is too low another test needed with my NEO4 N°2, this time again with M4 settings (rk3399-nanopi4-rev01.dtb loaded) and an additional fan. Memory performance restored, slightly better performance due to colder SoC. 4th result made with 4.19.0-rc4. Please be aware that RK3399 memory performance numbers differ alot between 4.4 and mainline kernel for yet unknown reasons!
  • ODROID-N2 number should be taken with a grain of salt since made with a pretty early software stack. Most probably scores will slightly improve over time. 'Overclocked' executions with both CPU clusters set to 2.0 GHz showed reliability issues most probably due to DVFS undervoltage (cpuminer quit almost immediately here while it ran only 50 seconds there -- this tool since being a load generator checking for data corruption can also be used for reliability testing but I would prefer our StabilityTester instead)
  • Rock Pi S is based on RK3308 Quad Cortex-A35 but the above numbers are not typical for A35 since the SoC design is severly limited: only a 16-bit RAM bus and 589MHz(*2) DDR clock in Rockchip’s DDR loader
  • RPi 3 B+ performance shown as original was measured with an older ThreadX release (6e08617e7767b09ef97b3d6cee8b75eba6d7ee0b from Mar 13 2018). Back then the 3B+ was faster than the 3B. This changed with a newer ThreadX release (4800f08a139d6ca1c5ecbee345ea6682e2160881 from Jun 7 2018) since RPi Trading people decided to trash performance on every RPi 3 B+ to masquerade instability issues on a fraction of boards (details)
  • RPi 3 B+ performance numbers shown as normal were made with no or just a heatsink (in contrast to with fan)
  • RPi 3 B+ marked as 'UV/normal' means: normal settings and average Micro USB cable resulting in UV (undervoltage). Once the demanding 7-zip benchmark started voltage dropped below 4.63V and 'frequency capping' (downclocking to 600 MHz) happened destroying performance. See the detailed log: 1400 MHz are reported by the kernel while it's 600 MHz in reality. Is this just highly misleading or already cheating?
  • RPi 3 B+ marked as 'OC/normal' means: OC (overclocked) settings, stable voltage but no fan used. Since SoC temperature exceeds 60°C the 'firmware' starts to cheat and downclocks to 1200 MHz while the kernel reports running at 1570 MHz. At least memory overclocking is somewhat effective.
  • RPi 4 B numbers improved over time partly due to 'firmware' optimisations resulting in faster memory access (lower latency). Using a 64-bit kernel (arm_64bit=1) just takes away 50MB of RAM, the worst choice is to combine 64-bit kernel and 64-bit userland since everything relevant get slower, same tasks require much more memory and the device will start to swap if low on memory or even kill processes due to out of memory. As a rule of thumb you need almost twice as much RAM with a 64-bit userland compared to 32-bit with the same programs/services running.
  • The highest clockspeeds listed for RPi 4/400 are 1800 MHz since this is what can be achieved with little overvoltage on early BCM2711B0 SoC revisions and is default without overvolting on revision C0 or later. While it's possible to 'overclock' the RPi 4 to 2.15 GHz, 2.3 GHz and even 2.4 GHz all of this requires overvolting the SoC up to over_voltage=15. Unfortunately the ThreadX DFVS (Dynamic voltage and frequency scaling) implementation is rather primitive and as such this overvolting happens also when the CPU cores idle at low frequencies which pretty much fries the CPU cores all the time resulting in high(er) temperatures and consumption figures than necessary. That's why 1.8 GHz is considered the reasonable maximum here.
  • Rock Pi 4B numbers are preliminary. Board has been tested without heatsink first so throttling occured as expected. Second time with higher cpufreq OPPs just a fan was added (fan without heatsink == pretty inefficient). Memory performance seems rather low but that's due to testing with vendor's armhf Linaro images -- see other RK3399 devices running same software stack, e.g. RockPro64 numbers above with kernel 4.4, armhf and also being limited to 1.8/1.4GHz.
  • Last RockPro64 entry has been made after ayufan solved memory performance problem with Rockchip's 4.4 kernel on his images (see discussion)
  • SBC2D70 results are somewhat representative for SigmaStar SSD201/SSD202D dual Cortex-A7 in general even though clockspeeds exceeding 1.3 GHz are considered boost frequencies and require appropriate cooling.
  • Ugoos UT2 might be representative for other RK3188 devices though memory performance with UT2 seems severely limited
  • Vim2 is somewhat special: not a real big.LITTLE design but two A53 clusters controlled by a firmware BLOB that allows cluster 0 to clock up to 1414 MHz (reported falsely as 1512 MHz) and cluster 1 able to reach 1 GHz (details)
  • The 'TRONFY MXQ S805' numbers should be similar to ODROID-C1/C1+ since same Amlogic S805 SoC at same clockspeed.
  • All the RISC-V scores (ClockworkPi R-01, Kendryte K510, T-HEAD C910 RVB-ICE) suffer from missing software optimizations. For example the openssl benchmark is currently generic C on RISC-V vs. optimized assembler on ARM or even ARMv8 Crypto Extensions or AES-NI on x64.
  • x86 numbers are meant as comparison. Atom E3826 numbers were made with a Minnowboard Turbot, x5-Z8300 numbers with an UP Board, 1st x5-Z8350 is an Atomic Pi and the 2nd a RockPi X, Celeron J3455 with an ASRock J3455-ITX mainboard, Pentium N4200 on UP2 Board, Pentium J4205 on an ASRock J4205-ITX, Ryzen Embedded R1606G on DFI GHF51 SBC, Celeron J4105 on two ODROID-H2 with different DDR4-PC19200 (2400MT/s) SO-DIMMs (remotely accessed via maze.odroid.com) and Celeron N4100 tested on an ODROID-H2 engineering sample with single channel DRAM config, Pentium J5005 is in an MINIX NEO J50C-4, Pentium G4600 is inside a TK Microserver MI106+.
  • Both Jasper Lake numbers (N4500/N5100) were obtained using passively cooled Mini PC with only one DIMM. With dual channel memory (and better cooling in N5100's case) some scores might be significantly higher.

Insights

  • Benchmarking the Raspberry Pi is useless when not taking into account that there always is a primary operating system running on the primary CPU (VideoCore) that fully controls the hardware. ARM cores are just guests here. That's why sbc-bench starting with v0.2 also logs ThreadX version and configuration (/boot/config.txt)
  • Looking at RPi 2 B+ numbers this is 2 times the same hardware, one time running latest Raspbian Stretch Lite and one time OMV/Armbian. Userland is both times Debian Stretch but Raspbian packages are built for ARMv6 while upstream Debian builds for ARMv7 (though with less effective compiler switches). Overall performance looks more or less the same except a very low memcopy bandwidth value with OMV. What's the reason since same ditro and kernel is used and same GCC to compile tinymembench? Is it firmware 'af8084725947aa2c7314172068f79dad9be1c8b4 from Apr 16 2018' vs. '47b05c853342eb6e4ea5b017d981e0ef247fb8be from Jul 3 2018'?
  • Looking at RPi 3 B+ numbers it's obvious that 'firmware' version is the most important factor. With original firmware (6e08617e7767b09ef97b3d6cee8b75eba6d7ee0b from Mar 13 2018) performance is ok just to get trashed after applying firmware 4800f08a139d6ca1c5ecbee345ea6682e2160881 from Jun 7 2018 which totally changes throttling behaviour. From then on you either need a fan for good performance or add a temp_soft_limit= entry to the firmware config file (we can't have a look what all those partially undocumented settings really do since RPi's main operating system is closed source)
  • tinymembench when executed on an A53 in an armhf userland compared to arm64 seems to generate lower memset numbers (78% on RK3399 -- see RockPro64 arm64 vs. RockPro64 armhf -- and 64% on RK3328 -- see Rock64 arm64 vs. Rock64 armhf). Status: needs further investigation and confirmation
  • Distro version doesn't seem to make a difference with 7-zip scores. Applies to both armhf and arm64 too -- see Rock64 numbers above
  • 7-zip scores benefit slightly from memory performance. See RK3328 equipped Renegade at 1.4 GHz with 4.4 kernel and Rock64 with same setup
  • openssl numbers are not affected by memory performance and are the same with same CPU cores and same clockspeeds. At least with Cortex-A53 running at 1.4 GHz with a Debian Stretch arm64 binary: Le Potato, NanoPi Fire3, Renegade, Rock64 and RockPro64 with openssl pinned to an A53 core: ~96000k with AES-128/16bit and ~650000k with AES-256/16KB
  • It seems the combination arm64 Bionic with very recent kernel improves AES encryption results with small data chunks (less than 1KB -- see Rock64 with 4.18 at 1.3GHz and Vim2 with 4.17 at 1.4GHz vs. Rock64 with 4.4 at 1.3GHz). Status: Needs further investigations (most probably related to GCC version)
  • It seems running an armhf userland on 64-bit SoCs also improves AES encryption results with small data chunks (see armhf entries for NanoPC T3+, Rock64, RockPro64 and Vim2). Status: very interesting, needs further investigations
  • It seems running Xenial binaries even further improves AES/SSL performance when ARMv8 Crypto Extensions are available. Status: while interesting irrelevant, we should get rid of Xenial and Jessie numbers.
  • It makes a huge difference whether ARMv8 Crypto Extensions can be used or not. See the many 64-bit SBC results above and compare with 32-bit SoCs or RPi 3B+, ODROID-C2 and NanoPi K2 (the latter 3 basing on 64-bit ARMv8 SoCs without crypto engine licensed/available)
  • The used distribution makes a big difference with cpuminer. Libs and GCC versions obviously matter (GCC 9.3 on Focal vs. 8.3 on Buster vs. 7.3 on Bionic vs. 6.3 on Stretch -- some benchmarks heavily depend on compiler versions). Stretch with GCC 7.3 provides a 15% performance increase with cpuminer on RK3328 and RK3399 (see Rock64 and NanoPC T4 numbers above and there the logs to compare performance of big and little cores). With GCC 8.2 and Stretch it's 20% with RK3328 and even 25% with RK3399 (the A72 performance increasing more compared to the A53 cores -- check individual kH/s numbers in the logs). With Focal (GCC 9.3) it's even more, compare Rock Pi 4A with last RockPro64 entry or NanoPi Fire3 results.
  • (more to come soon)

The bigger picture

  • To compare different hardware exactly the same software environment (apps, libs, compiler, kernel) is needed. Ignoring this will produce numbers without meaning.
  • ARM's big cores (A15, A17, A72) perform a lot better than the little cores (A7, A53). Everything that needs high single threaded performance will hugely benefit from running on such a core. This puts SoCs like RK3288 (Tinkerboard), Exynos 5244 (ODROID XU4) or RK3399 in a better position. For the big.LITTLE designs a working HMP scheduler is mandatory since otherwise performance hungry tasks end up on a slow core. This is even true for pseudo big.LITTLE like on the VIM2/S912
  • 7-zip's benchmark still looks like a nice indicator for a 'server workloads' performance index (multi threaded tasks that do not rely on floating point arithmetics but partially on memory performance). Though these scores are totally irrelevant when it's about SBC use cases that focus on something different (e.g. a 'Desktop Linux' needing high single threaded CPU performance, HW accelerated GPU and VPU and also fast random IO on the rootfs)
  • We see a huge variation in tinymembench numbers with some boards outperforming others by magnitudes while the effect in reality for CPU bound workloads is rather minimal though high memory bandwidth is a requirement for certain other tasks (e.g. playing 4K video). At least numbers are there to generate further insights.
  • Identical SoCs perform more or less identical if 'environmental conditions' (clockspeeds) are the same -- see Renegade vs. Rock64 numbers or NanoPC T4 vs. RockPro64 or ODROID-C2 vs. NanoPi K2.
  • Same could be said for different Cortex-A cores. One A53 performs like the other as long as both run at the same clockspeed (with some exceptions most probably due to internal cache sizes -- see cpuminer numbers for Amlogic S905 vs. S905X/RK3328). With same count of cores you get similar performance (if the task(s) in question benefits from parallel execution)
  • Cortex-A53 running at the same clockspeed as A7 shows almost ~30% better performance (~3500 7-zip MIPS vs. ~2700). This is even true when running ARMv7 code (see RPi 3 B+ numbers). In general it seems irrelevant whether the A53 cores run an armhf or arm64 userland, some numbers are even higher when running armhf code. This is very interesting since there are scenarios where running an armhf userland results in needing way less physical memory for the same task while performing identical. Please note: it's about the userland (32-bit vs. 64-bit) and not kernel (64-bit of course)

TODO