ARM Cortex-A53

Amlogic S905 (ARM Cortex-A53), 1536 MHz, (28 nm). 4 cores. 2 GB DDR3-1824 (13-13-13) (32-bit). ODROID-C2 board.

L1 Data cache = 32 KB, 64 B/line, 4-WAY, PIPT.
L1 Instruction cache = 32 KB, 64 B/line, 2-WAY, VIPT.
L2 Cache = 512 KB, 64 B/line, 16-WAY, shared by all cores.

Branch Target Address Cache: 256-entry.
Branch predictor: 3072-entry pattern history prediction table.
Return stack : 8-entry.

L1 Data Cache Latency = 3 cycles for simple access via pointer
L1 Data Cache Latency = 3 cycles for access with complex address calculation (size_t n, *p; n = p[n]).
L2 Cache Latency = 15 cycles
RAM Latency = 15 cycles + 128 ns

2 MB pages (64-bit Linux)

Micro Data TLB L1 (4 KB pages): 10 items. ?-way. Miss penalty = 2 cycles. Parallel miss: 3 cycles per L1 data cache read access.
TLB L2 (2 MB pages): 512 items. 4-way. Miss penalty = 20 ? cycles.

  Size        Latency        Increase   Description

  32 K     3                            
  64 K    10                        6   + 12 (L2) + 2 (Micro TLB miss)                 
 128 K    14                        4   
 256 K    16                        2
 512 K    17                        1   
   1 M    17 +  70 ns         + 70 ns   + 128 ns (RAM)
   2 M    17 + 100 ns         + 30 ns
   4 M    17 + 114 ns         + 14 ns   
   8 M    17 + 121 ns         +  7 ns 
  16 M    17 + 124 ns         +  3 ns   
  32 M    17 + 126 ns         +  2 ns   
  64 M    17 + 127 ns         +  1 ns
 128 M    17 + 128 ns         +  1 ns
 256 M    17 + 128 ns                
 512 M    17 + 128 ns       
1024 M    17 + 128 ns       
2048 M                                  + 20 ? (L2 TLB miss)

4 KB pages (64-bit Linux)

Micro Data TLB L1 (4 KB pages): 10 items. ?-way. Miss penalty = 2 cycles. Parallel miss: 3 cycles per L1 data cache read access.
TLB L2: 512 items. 4-way. Miss penalty = 11-22 cycles (translation entry from L1/L2 cache). Parallel miss: 22 cycles per L2 cache read access.

  Size        Latency        Increase   Description

  32 K     3                            
  64 K    10                        6   + 12 (L2) + 2 (Micro TLB miss)                 
 128 K    14                        4   
 256 K    16                        2
 512 K    17 +   4 ns       1 +  4 ns
   1 M    17 +  70 ns         + 66 ns   + 128 ns (RAM)
   2 M    17 + 100 ns         + 30 ns
   4 M    25 + 114 ns       8 + 14 ns   + 22 (L2 TLB miss)        
   8 M    31 + 121 ns       6 +  7 ns 
  16 M    35 + 124 ns       4 +  3 ns   
  32 M    37 + 126 ns       2 +  2 ns   
  64 M    39 + 127 ns       2 +  1 ns
 128 M    39 + 141 ns         + 14 ns
 256 M    39 + 192 ns         + 51 ns   + 128 ns (Page walk to RAM)
 512 M    39 + 230 ns         + 38 ns
1024 M    39 + 250 ns         + 20 ns

MISC

Branch misprediction penalty = 7 cycles

4-bytes range cross penalty = 1 cycles (32-bit)
8-bytes range cross penalty = 2 cycles
Page cross - no additional penalty

L1 B/W (Parallel Random Read) = 1 cycle per one access

L2->L1 B/W (Parallel Random Read) = 9.30 cycles per access
L2->L1 B/W (Read, 64 bytes step) = 9.00 cycles per cache line
L2->L1 B/W (Read, 64 bytes step - pointer chasing) = 8.00 cycles per cache line (HW prefetch L2->L1)
L2 Write B/W (skip, 64 Bytes step) = 8.00 cycles per write

RAM Read B/W (Parallel Random Read) = 54 ns per access
RAM Read B/W (Read, 8-64 Bytes step) = 2000 MB/s
RAM Read B/W (Read, 8-64 Bytes step - pointer chasing) = 2000 MB/s (HW prefetch from RAM)
RAM Write B/W (8 Bytes step, 64-bit) = 3700 MB/s
RAM Write B/W (skip, 16 - 64 Bytes step) = 1800 MB/s

7-Zip Benchmark

7z b -mm=* -mmt1

Note: MIPS and Effectiveness values are normalized with AMD K8 cpu.

Note: Core 0 was not used. Core 0 provides only about 1500 MHz for user application. Another cores work at 1536 MHz.

32-bit, arm32


# gcc-6 -O2 -mcpu=cortex-a53

7-Zip (a) [32] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=ru_RU.UTF-8,Utf16=on,HugeFiles=on,32 bits,3 CPUs LE)

LE
CPU Freq:  1534  1535  1535  1535  1535  1535  1535  1535  1535

RAM size:    1719 MB,  # CPU hardware threads:   3
RAM usage:    225 MB,  # Benchmark threads:      1


Method           Speed Usage    R/U Rating   E/U Effec
                 KiB/s     %   MIPS   MIPS     %     %

CPU                      100   1537   1535
CPU                      100   1537   1535
CPU                      100   1533   1535   100   100

LZMA:x1           3205   100   1172   1172    76    76
                 20421   100   1662   1663   108   108
LZMA:x5:mt1        811   100   1014   1014    66    66
                 18597   100   1566   1569   102   102
LZMA:x5:mt2       1269   172    922   1586    60   103
                 18585   100   1566   1568   102   102
Deflate:x1        8810   100   1118   1119    73    73
                 43203   100   1343   1342    87    87
Deflate:x5        2947   100   1137   1135    74    74
                 43270   100   1344   1343    88    87
Deflate:x7        1103   100   1223   1223    80    80
                 43554   100   1353   1352    88    88
Deflate64:x5      2579   100   1115   1114    73    73
                 42955   100   1345   1344    88    88
BZip2:x1          1642   100    993    992    65    65
                  8378   100    909    908    59    59
BZip2:x5          1045   100    873    873    57    57
                  4858   100    954    953    62    62
BZip2:x5:mt2      1834   195    786   1531    51   100
                  9108   191    935   1788    61   116
BZip2:x7           347   100    900    899    59    59
                  4883   100    959    958    62    62
PPMD:x1            990   100   1025   1024    67    67
                   771   100    909    908    59    59
PPMD:x5            665   100   1129   1128    74    73
                   537   100   1007   1007    66    66
Delta:4         189803   100   1168   1166    76    76
                184005   100   1131   1131    74    74
BCJ             200593   100    822    822    54    54
                200797   100    822    822    54    54
AES256CBC:1      37526   100    923    922    60    60
                 37546   100    922    923    60    60
AES256CBC:2 

CRC32:1         166189   100   1211   1210    79    79
CRC32:4         425545   100    949    950    62    62
CRC32:8         620946   100    842    842    55    55
CRC64           297979   100    610    610    40    40
SHA256           52860   100   1079   1078    70    70
SHA1            113298   100   1061   1060    69    69
BLAKE2sp         45058   100    991    991    65    65

CPU                      100   1537   1535
------------------------------------------------------
Tot:                     111   1179   1276    75    83

32-bit, Thumb2


# gcc-6 -O2 -mthumb -mcpu=cortex-a53

7-Zip (a) [32] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=ru_RU.UTF-8,Utf16=on,HugeFiles=on,32 bits,3 CPUs LE)

LE
CPU Freq:  1535  1534  1535  1535  1535  1535  1535  1535  1534

RAM size:    1719 MB,  # CPU hardware threads:   3
RAM usage:    225 MB,  # Benchmark threads:      1


Method           Speed Usage    R/U Rating   E/U Effec
                 KiB/s     %   MIPS   MIPS     %     %

CPU                      100   1537   1535
CPU                      100   1537   1535
CPU                      100   1533   1535   100   100

LZMA:x1           3128   100   1143   1144    74    75
                 17070   100   1390   1390    91    91
LZMA:x5:mt1        819   100   1024   1024    67    67
                 15883   100   1342   1340    87    87
LZMA:x5:mt2       1285   173    927   1605    60   105
                 15912   100   1342   1342    87    87
Deflate:x1        8850   100   1125   1124    73    73
                 42192   100   1311   1311    85    85
Deflate:x5        3016   100   1162   1162    76    76
                 42194   100   1310   1310    85    85
Deflate:x7        1105   100   1226   1225    80    80
                 42417   100   1315   1316    86    86
Deflate64:x5      2631   100   1136   1137    74    74
                 42168   100   1320   1319    86    86
BZip2:x1          1572   100    950    950    62    62
                  9200   100    998    997    65    65
BZip2:x5          1014   100    847    847    55    55
                  5197   100   1021   1020    66    66
BZip2:x5:mt2      1776   195    762   1483    50    97
                  9641   191    992   1892    65   123
BZip2:x7           325   100    844    844    55    55
                  5218   100   1023   1023    67    67
PPMD:x1           1013   100   1048   1048    68    68
                   878   100   1035   1034    67    67
PPMD:x5            676   100   1146   1146    75    75
                   596   100   1118   1118    73    73
Delta:4         173258   100   1066   1064    69    69
                144342   100    886    887    58    58
BCJ             212562   100    871    871    57    57
                212315   100    870    870    57    57
AES256CBC:1      31623   100    777    777    51    51
                 31507   100    774    774    50    50
AES256CBC:2 

CRC32:1         166290   100   1211   1211    79    79
CRC32:4         385831   100    861    861    56    56
CRC32:8         542376   100    735    735    48    48
CRC64           277470   100    568    568    37    37
SHA256           44457   100    907    907    59    59
SHA1            107530   100   1007   1006    66    66
BLAKE2sp         42744   100    940    940    61    61

CPU                      100   1537   1535
------------------------------------------------------
Tot:                     111   1108   1206    71    79

64-bit


# gcc-6 -O2 -march=armv8-a -mcpu=cortex-a53

7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=ru_RU.UTF-8,Utf16=on,HugeFiles=on,64 bits,3 CPUs LE)

LE
CPU Freq:  1534  1535  1535  1535  1535  1535  1535  1535  1535

RAM size:    1719 MB,  # CPU hardware threads:   3
RAM usage:    225 MB,  # Benchmark threads:      1


Method           Speed Usage    R/U Rating   E/U Effec
                 KiB/s     %   MIPS   MIPS     %     %

CPU                      100   1535   1535
CPU                      100   1535   1535
CPU                      100   1535   1535   100   100

LZMA:x1           3204   100   1171   1171    76    76
                 17797   100   1449   1449    94    94
LZMA:x5:mt1        794   100    993    993    65    65
                 16449   100   1388   1387    90    90
LZMA:x5:mt2       1304   170    958   1630    62   106
                 16507   100   1393   1392    91    91
Deflate:x1        9042   100   1149   1148    75    75
                 41510   100   1291   1290    84    84
Deflate:x5        3356   100   1293   1292    84    84
                 41590   100   1292   1291    84    84
Deflate:x7        1174   100   1302   1301    85    85
                 41826   100   1298   1298    85    85
Deflate64:x5      2895   100   1252   1251    82    82
                 41540   100   1300   1299    85    85
BZip2:x1          1649   100    997    997    65    65
                  9855   100   1069   1068    70    70
BZip2:x5          1047   100    875    874    57    57
                  5565   100   1093   1092    71    71
BZip2:x5:mt2      1833   195    785   1530    51   100
                 10063   186   1060   1975    69   129
BZip2:x7           346   100    898    897    58    58
                  5593   100   1098   1097    71    71
PPMD:x1            945   100    978    978    64    64
                   873   100   1029   1029    67    67
PPMD:x5            637   100   1081   1080    70    70
                   591   100   1109   1108    72    72
Delta:4         211588   100   1301   1300    85    85
                152901   100    940    939    61    61
BCJ             231429   100    948    948    62    62
                231500   100    949    948    62    62
AES256CBC:1      41594   100   1023   1022    67    67
                 43796   100   1077   1076    70    70
AES256CBC:2 

CRC32:1         149673   100   1090   1090    71    71
CRC32:4         396745   100    886    886    58    58
CRC32:8         561884   100    762    762    50    50
CRC64           424277   100    869    869    57    57
SHA256           69726   100   1423   1422    93    93
SHA1            119720   100   1121   1121    73    73
BLAKE2sp         73764   100   1624   1623   106   106

CPU                      100   1535   1535
------------------------------------------------------
Tot:                     110   1159   1256    74    82