Open ThomasKaiser opened 1 year ago
Now to another machine. NanoPi Fire3, a board equipped with only 1 GB RAM and a Nexell S5P6818 SoC (two quad-core A53 clusters running at 1.4 GHz). Since the board is that low on RAM we need to add -md=2m
to 7-zip parameters to test with smaller data sizes preventing the oom-killer to kick in.
When running 7-zip single-threaded in parallel on all 8 cores each one scores 1005 (in sum 8043). When testing with 7-zip fully multi-threaded on all 8 cores we get a score of 7377 (91.7% of the theoretical combined score).
When testing multi-threaded through 2 to 7 cores something interesting happens (numbers in brackets show the percentage compared to a theoretical ideal sum score based on 1005 7-ZIP-MIPS per core):
Running with odd number of threads produces odd results. If we compare just the even thread counts it looks like this:
And the strange drop with 6 threads is reproducible.
for i in $(seq 0 7) ; do taskset -c $i 7zr b -mmt=1 -md=2m >7-zip-${i}.log & done
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 64000000 64000000 64000000 - 256000000 - 1024000000 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 30 MB, # Benchmark threads: 1
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 835 100 745 744 | 15909 99 1300 1287
19: 799 100 717 717 | 15823 100 1296 1296
20: 768 100 700 700 | 15557 100 1289 1289
21: 733 100 688 688 | 15308 100 1287 1288
---------------------------------- | ------------------------------
Avr: 100 712 712 | 100 1293 1290
Tot: 100 1003 1001
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 32000000 64000000 - 64000000 - - 512000000 - 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 30 MB, # Benchmark threads: 1
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 845 100 753 753 | 16128 100 1305 1304
19: 806 100 723 723 | 15856 100 1299 1298
20: 773 100 705 705 | 15582 100 1291 1291
21: 739 100 693 692 | 15272 100 1284 1285
---------------------------------- | ------------------------------
Avr: 100 718 718 | 100 1295 1295
Tot: 100 1007 1006
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 64000000 64000000 64000000 128000000 256000000 - 1024000000 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 30 MB, # Benchmark threads: 1
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 839 100 748 748 | 16129 100 1305 1304
19: 803 100 720 720 | 15892 100 1301 1301
20: 771 100 703 703 | 15626 100 1295 1295
21: 738 100 692 692 | 15336 100 1290 1290
---------------------------------- | ------------------------------
Avr: 100 716 716 | 100 1298 1298
Tot: 100 1007 1007
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 64000000 64000000 - 128000000 - 512000000 1024000000 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 30 MB, # Benchmark threads: 1
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 842 100 750 750 | 16143 100 1306 1305
19: 803 100 720 720 | 15869 100 1299 1300
20: 769 100 701 701 | 15598 100 1293 1293
21: 734 100 688 688 | 15343 100 1291 1290
---------------------------------- | ------------------------------
Avr: 100 715 715 | 100 1297 1297
Tot: 100 1006 1006
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 64000000 - 64000000 128000000 256000000 512000000 - 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 30 MB, # Benchmark threads: 1
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 843 100 751 751 | 16059 100 1299 1299
19: 801 100 718 718 | 15801 100 1294 1294
20: 771 100 703 703 | 15562 100 1289 1290
21: 738 100 692 692 | 15285 100 1286 1286
---------------------------------- | ------------------------------
Avr: 100 716 716 | 100 1292 1292
Tot: 100 1004 1004
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 64000000 - 64000000 128000000 256000000 512000000 1024000000 -
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 30 MB, # Benchmark threads: 1
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 847 100 755 755 | 16121 100 1304 1304
19: 805 100 722 722 | 15850 100 1298 1298
20: 774 100 706 706 | 15590 100 1292 1292
21: 740 100 694 694 | 15258 100 1283 1283
---------------------------------- | ------------------------------
Avr: 100 719 719 | 100 1294 1294
Tot: 100 1007 1007
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 64000000 64000000 64000000 128000000 - - - 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 30 MB, # Benchmark threads: 1
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 841 100 750 750 | 16119 100 1304 1304
19: 799 100 716 716 | 15865 100 1299 1299
20: 767 100 700 700 | 15595 100 1293 1292
21: 732 100 687 687 | 15301 100 1287 1287
---------------------------------- | ------------------------------
Avr: 100 713 713 | 100 1296 1295
Tot: 100 1004 1004
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 32000000 64000000 64000000 64000000 128000000 256000000 512000000 1024000000 1024000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 30 MB, # Benchmark threads: 1
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 845 100 753 753 | 16134 100 1305 1305
19: 806 100 723 723 | 15873 100 1300 1300
20: 773 100 705 705 | 15606 100 1293 1293
21: 739 100 693 692 | 15365 100 1292 1292
---------------------------------- | ------------------------------
Avr: 100 718 718 | 100 1297 1298
Tot: 100 1008 1008
root@nanopifire3:/tmp# for i in $(seq 1 7) ; do taskset -c 0-${i} 7zr b -mmt=$(( ${i} + 1 )) -md=2m ; done
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 32000000 64000000 64000000 - 128000000 256000000 - 1024000000 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 36 MB, # Benchmark threads: 2
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 1403 144 869 1250 | 32805 200 1328 2653
19: 1364 150 817 1222 | 32336 200 1325 2648
20: 1344 155 791 1225 | 31841 200 1321 2638
21: 1309 159 773 1227 | 31364 200 1319 2638
---------------------------------- | ------------------------------
Avr: 152 813 1231 | 200 1323 2644
Tot: 176 1068 1938
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 - 64000000 64000000 128000000 256000000 512000000 - 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 36 MB, # Benchmark threads: 3
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 1437 147 872 1281 | 32847 200 1328 2656
19: 1392 153 817 1248 | 32407 200 1327 2654
20: 1345 157 783 1227 | 31957 200 1324 2648
21: 1312 160 770 1230 | 31313 200 1318 2634
---------------------------------- | ------------------------------
Avr: 154 810 1246 | 200 1324 2648
Tot: 177 1067 1947
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 64000000 64000000 64000000 128000000 256000000 - 1024000000 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 72 MB, # Benchmark threads: 4
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 2632 291 807 2345 | 64949 400 1313 5248
19: 2532 296 765 2269 | 63899 400 1308 5226
20: 2459 305 735 2242 | 62752 399 1303 5202
21: 2440 317 722 2287 | 61536 399 1295 5171
---------------------------------- | ------------------------------
Avr: 302 757 2286 | 399 1305 5212
Tot: 351 1031 3749
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 - 64000000 - - 256000000 512000000 1024000000 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 72 MB, # Benchmark threads: 5
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 2795 292 854 2491 | 65205 397 1326 5269
19: 2681 299 805 2403 | 63928 396 1321 5228
20: 2586 306 771 2357 | 63097 397 1317 5231
21: 2580 319 759 2418 | 41902 267 1320 3521
---------------------------------- | ------------------------------
Avr: 304 797 2417 | 364 1321 4812
Tot: 334 1059 3615
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 32000000 64000000 64000000 64000000 128000000 256000000 512000000 512000000 1024000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 108 MB, # Benchmark threads: 6
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 3941 428 820 3511 | 97373 597 1317 7866
19: 3791 443 767 3397 | 95196 594 1311 7789
20: 3642 451 736 3320 | 90796 576 1306 7528
21: 3590 464 724 3364 | 62519 402 1308 5256
---------------------------------- | ------------------------------
Avr: 447 762 3398 | 542 1311 7110
Tot: 495 1036 5254
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 32000000 64000000 64000000 64000000 128000000 256000000 512000000 - 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 108 MB, # Benchmark threads: 7
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 4016 437 818 3578 | 97652 599 1318 7888
19: 3880 457 760 3477 | 96070 598 1314 7861
20: 3766 472 728 3433 | 94367 599 1307 7824
21: 3669 481 715 3438 | 92524 598 1301 7779
---------------------------------- | ------------------------------
Avr: 462 755 3481 | 598 1310 7838
Tot: 530 1033 5660
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 64000000 64000000 64000000 128000000 256000000 512000000 1024000000 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 145 MB, # Benchmark threads: 8
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 5089 576 787 4535 | 128772 797 1308 10415
19: 4905 603 729 4395 | 126795 797 1302 10372
20: 4750 622 696 4330 | 124681 798 1295 10332
21: 4689 639 687 4394 | 121936 796 1287 10246
---------------------------------- | ------------------------------
Avr: 610 725 4413 | 797 1298 10341
Tot: 704 1011 7377
root@nanopifire3:/tmp# for i in $(seq 1 7) ; do taskset -c 0-${i} 7zr b -mmt=$(( ${i} + 1 )) -md=2m ; done
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 32000000 64000000 64000000 - 128000000 256000000 - 1024000000 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 36 MB, # Benchmark threads: 2
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 1403 144 869 1250 | 32805 200 1328 2653
19: 1364 150 817 1222 | 32336 200 1325 2648
20: 1344 155 791 1225 | 31841 200 1321 2638
21: 1309 159 773 1227 | 31364 200 1319 2638
---------------------------------- | ------------------------------
Avr: 152 813 1231 | 200 1323 2644
Tot: 176 1068 1938
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 - 64000000 64000000 128000000 256000000 512000000 - 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 36 MB, # Benchmark threads: 3
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 1437 147 872 1281 | 32847 200 1328 2656
19: 1392 153 817 1248 | 32407 200 1327 2654
20: 1345 157 783 1227 | 31957 200 1324 2648
21: 1312 160 770 1230 | 31313 200 1318 2634
---------------------------------- | ------------------------------
Avr: 154 810 1246 | 200 1324 2648
Tot: 177 1067 1947
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 64000000 64000000 64000000 128000000 256000000 - 1024000000 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 72 MB, # Benchmark threads: 4
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 2632 291 807 2345 | 64949 400 1313 5248
19: 2532 296 765 2269 | 63899 400 1308 5226
20: 2459 305 735 2242 | 62752 399 1303 5202
21: 2440 317 722 2287 | 61536 399 1295 5171
---------------------------------- | ------------------------------
Avr: 302 757 2286 | 399 1305 5212
Tot: 351 1031 3749
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 - 64000000 - - 256000000 512000000 1024000000 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 72 MB, # Benchmark threads: 5
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 2795 292 854 2491 | 65205 397 1326 5269
19: 2681 299 805 2403 | 63928 396 1321 5228
20: 2586 306 771 2357 | 63097 397 1317 5231
21: 2580 319 759 2418 | 41902 267 1320 3521
---------------------------------- | ------------------------------
Avr: 304 797 2417 | 364 1321 4812
Tot: 334 1059 3615
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 32000000 64000000 64000000 64000000 128000000 256000000 512000000 512000000 1024000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 108 MB, # Benchmark threads: 6
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 3941 428 820 3511 | 97373 597 1317 7866
19: 3791 443 767 3397 | 95196 594 1311 7789
20: 3642 451 736 3320 | 90796 576 1306 7528
21: 3590 464 724 3364 | 62519 402 1308 5256
---------------------------------- | ------------------------------
Avr: 447 762 3398 | 542 1311 7110
Tot: 495 1036 5254
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 32000000 64000000 64000000 64000000 128000000 256000000 512000000 - 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 108 MB, # Benchmark threads: 7
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 4016 437 818 3578 | 97652 599 1318 7888
19: 3880 457 760 3477 | 96070 598 1314 7861
20: 3766 472 728 3433 | 94367 599 1307 7824
21: 3669 481 715 3438 | 92524 598 1301 7779
---------------------------------- | ------------------------------
Avr: 462 755 3481 | 598 1310 7838
Tot: 530 1033 5660
7-Zip (a) [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)
LE
CPU Freq: 64000000 64000000 64000000 64000000 128000000 256000000 512000000 1024000000 2048000000
RAM size: 994 MB, # CPU hardware threads: 8
RAM usage: 145 MB, # Benchmark threads: 8
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
18: 5089 576 787 4535 | 128772 797 1308 10415
19: 4905 603 729 4395 | 126795 797 1302 10372
20: 4750 622 696 4330 | 124681 798 1295 10332
21: 4689 639 687 4394 | 121936 796 1287 10246
---------------------------------- | ------------------------------
Avr: 610 725 4413 | 797 1298 10341
Tot: 704 1011 7377
Trying to understand 7-ZIP MIPS with many CPU cores
Started by a conversation with @wtarreau about benchmarking issues on big iron (Ampere Altra with +64 cores) let's look closer how 7-zip MIPS behave in situations where we can manipulate memory access and have two sets of different CPU core types:
RK3588 consists of A55 and A76 cores. This SoC running with BSP kernel is nice since we can dynamically adjust the DRAM frequency between 528 and 2112 MHz (more insights wrt DRAM clock here).
We're testing with 528 MHz (
powersave
DMC governor) and 2112 MHz (performance
DMC governor).When running the tests single-threaded we get higher numbers as when all cores are busy at the same time.
For example when testing with DMC (and cpufreq) governor set to
performance
an A55 scores 1629 7-zip MIPS and an A76 3121. All 8 cores combined should end up with a theoretical 7-ZIP MIPS score of 19000 if there are no further bottlenecks affecting cores running in parallel.In reality with these settings we see scores that are ~15% lower. In this measurement it's 16309 vs. the theoretical 19000 (85.5%).
To check the influence memory access make let's compare with
powersave
DMC settings (DRAM clocked at just 528 MHz). All scores are lower, single-threaded the A55 are at 85.8% and the A76 at 81.8% their performance compared to running with 2112 MHz so the faster core is more bottlenecked which makes sense given 7-zip depending on memory latency. The measured multi-threaded score suffers even more: 79.1% compared toperformance
DMC governor.To check whether mixing different types of cores affects the results (7-zip having to wait longer for results from the slow cores and as such slowing down the faster cores unnecessarily) I fired up a round of tests with 8 individual
7zr b -mmt=1
runs in parallel. So each 7-zip task was isolated and ran a single thread.When adding these individual scores together the numbers are slightly better but differ not that much. The combined score is 16807 at 2112 MHz (vs. 16309 when measuring fully multi-threaded, as such a 3% difference) and at 528 MHz it's 13479 (vs. 12911 when measuring fully multi-threaded, a difference slightly exceeding 4% which hints at the lower DRAM freq bottlenecking further in situations with full memory access by all cores)
All numbers with
performance
dmc governor:And the same with
powersave
dmc governor:Let's repeat the test limited to one cluster at a time. Let's focus only on the A76 cluster 1st and remain on 2112 MHz:
When only the A76 run each a single-threaded 7-zip benchmark the per core score is ~2851 compared to ~2735 when the A55 are busy in parallel. The sum of these single scores is 11404 and based on what we saw above we would expect this value being higher compared to letting 7-zip run multi-threaded on all A76 in parallel.
Though it's not when checking with
taskset -c 4-7 7zr b -mmt=4
since the average score built from 3 runs is 12319 or 8% higher compared to adding the single-threaded scores.Raw data:
Performance:
Single-threaded
3 times mutli-threaded (all cores):
single-threaded running fully parallel:
Only testing the A76 cluster
First testing each A76 core individually but in parallel:
Now testing three times all four A76 together:
Powersave:
Single-threaded
3 times mutli-threaded (all cores):
single-threaded: