Некоторое время назад появились тесты процессора Байкал-S, поэтому я решил сравнить
производительность данного процессора с китайским процессором Kunpeng 920 (920-4826), к которому некоторое время назад получил доступ.
Характеристики сравниваемых процессоров:
Байкал-S | Kunpeng 920 | |
Архитектура | aarch64 | aarch64 |
ISA | ARMv8.2-A | ARMv8.2-A |
Микроархитектура | Cortex-A75 | TaiShan v110 |
Частота (МГц) | 2000 | 2600 |
Ядра; Потоки | 48 | 48 |
Техпроцесс (нм) | 16 | 7 |
TDP (Вт) | 100 | 150 |
Тип ОЗУ | DDR4-3200 | DDR4-2933 |
Сокет | FCLGA-3467 | BGA |
Каналов ОЗУ | 6 | 8 |
Макс ОЗУ (ГБ) | 768 | 1024 |
ГФлопс (DP) | 384 | 500 |
ГФлопс (SP) | 768 | 2000 |
Год | 2021 | 2019 |
Были проведены следующие тесты:
7zip встроенный бенчмарк
Dhrystone, Whetsone
Coremark
Scimark 2
HPL
Stream
Stockfish
Blender
Geekbench 5
Немедленно переходим к результатам, детали тестов смотрите далее.
Результаты
|
| Байкал-S | Kunpeng 920 | ||
Тест | Единица измерения | Результат | Результат | ||
1 поток | 48 потоков | 1 поток | 96 (48) потоков | ||
CoreMark |
| 13 540 | 647 292 | 18 398 | 1 291 450 (874 104) |
Dhrystone | DMIPS | 13 410 |
| 21 830 |
|
Whetstone | MWIPS | 4 702 | 225 078 | 3 857 | 362 558 |
HPL 2.3 |
| 7 | 294 |
| 327 (194) |
7z | Total; Compress; Decompress | 2 378; 2 400; 2 356 | 90 039; 71 542; 108 536 |
| 194 573; 150 105 ; 239 042 (119 705; 101 272; 138 139) |
STREAM | МБ/с Copy; Scale; Add; Triad |
| 81 787; 78 037; 76 156; 75 113 |
| 93 271; 75 846; 78 009; 79 063 |
Blender | Секунд | 16,96 |
|
| 8,2 (13,56) |
Stockfish | Nodes/second |
| 43 097 684 |
| 77 469 854 |
Geekbench 5 [2] |
| 405 | 13 549 | 599 | 28 112 (18 817) |
Немного об архитектурах процессоров Байкал-S и Kunpeng 920
Байкал-S
Байкал-S — процессор на основе 64 разрядной RISC архитектуре ARM (armv8, aarch64), работает на частоте 2 ГГц (возможна 2,5 ГГц) имеет 48 ядер, которые реализуют микроархитектуру ядер Cortex-A75.
Особенности ядер процессора Байкал-S на основе микроархитектуры Cortex-A75:
64 битная архитектура armv8.2
FP/SIMD расширения VFPv4 и NEON
Внеочередное исполнение
Предсказание ветвлений
Поддержка виртуализации
Конвейер 11 – 13 стадий
8 портов на исполнение микроопераций:
2 Загрузки/Сохранения
2 целых АЛУ (сложение, сдвиг)
1 блок ветвлений
1 АЛУ для умножения, деления
2 блока SIMD/FPU
3 уровневый декодер команд (6 микроинструкций за цикл)
Кеши
64 КБ L1 кэш команд (4 канальный, ассоциативный, размер линии 64 байта)
64 КБ L1 кэш данных (4 канальный, ассоциативный, размер линии 64 байта)
В Байкал-S 512 КБ L2 на 1 ядро (4 ядра в кластере), в сумме 24 МБ
Кэш L3: 24 МБ (2 МБ в кластере)
Кэш L4: 32 МБ
Kunpeng 920
Kunpeng 920 — процессор разработанный команией HiSilicon (производит для Huawei) процессор на основе глубоко модифицированной микроархитектуре Arm Cortex-A72 (TaiShan V110). Существуют 32, 48 и 64 ядерные модели (3226, 4826, 6426).
Особенности ядер TaiShan V110 (для 48 ядер):
64 битная архитектура armv8.2
FP/SIMD расширения VFPv4 и NEON
Внеочередное исполнение
Улучшенный предсказатель ветвлений
Поддержка виртуализации
Конвейер 11 – 13 стадий
8 портов на исполнение микроопераций:
2 Загрузки/Сохранения
2 целых АЛУ (сложение, сдвиг)
1 блок ветвлений
1 АЛУ для умножения, деления
2 блока SIMD/FPU
4 уровневый декодер команд (3 уровня в Cortex-A72)
Кеши
64 КБ L1 кэш команд (4 канальный, ассоциативный, размер линии 64 байта)
64 КБ L1 кэш данных (4 канальный, ассоциативный, размер линии 64 байта)
В Kunpeng 920 512 КБ L2 на 1 ядро (4 ядра в кластере), в сумме 24 МБ
Кэш L3: 48 МБ (4 МБ в кластере)
Разбираем результаты
Coremark
Современный тест, который должен заменить Dhrystone и Whetstone. Написан на C. Считает различные массивы, матрицы, сортировка массивов и т. д. Предназначался для запуска на всём: от микроконтроллеров до мощных процессоров.
Вывод однопоточного теста Coremark процессора Байкал-S
./coremark.exe
2K performance run parameters for coremark.
CoreMark Size : 666
Total ticks : 14771
Total time (secs): 14.771000
Iterations/Sec : 13540.044682
Iterations : 200000
Compiler version : GCC10.2.1 20210110
Compiler flags : -static -Ofast -funroll-all-loops -EL -falign-functions=16
-DPERFORMANCE_RUN=1 -lrt
Memory location : Please put data memory location here
(e.g. code in flash. data on heap etc)
seedcrc : 0xe9f5
[0]crclist : 0xe714
[0]crcmatrix : 0x1fd7
[0]crcstate : 0x8e3a
[0]crcfinal : 0x4983
Correct operation validated. See readme.txt for run and reporting rules.
CoreMark 1.0 : 13540.044682 / GCC10.2.1 20210110 -static -Ofast -funroll-all-
loops -EL -falign-functions=16 -DPERFORMANCE_RUN=1 -lrt / HeapВывод многопоточного теста Coremark процессора Байкал-S
./coremarkMT.exe
2K performance run parameters for coremark.
CoreMark Size : 666
Total ticks : 14815
Total time (secs): 14.815000
Iterations/Sec : 647991.900101
Iterations : 9600000
Compiler version : GCC10.2.1 20210110
Compiler flags : -static -Ofast -funroll-all-loops -EL -falign-functions=16
-DUSE_PHTREAD -DMULTITHREAD=48 -DPERFORMANCE_RUN=1 -lrt
Parallel PThreads : 48
Memory location : Please put data memory location here
(e.g. code in flash. data on heap etc)
seedcrc : 0xe9f5
[0]crclist : 0xe714
[1]crclist : 0xe714
.....
[46]crclist : 0xe714
[47]crclist : 0xe714
[0]crcmatrix : 0x1fd7
[1]crcmatrix : 0x1fd7
.....
[46]crcmatrix : 0x1fd7
[47]crcmatrix : 0x1fd7
[0]crcstate : 0x8e3a
[1]crcstate : 0x8e3a
[2]crcstate : 0x8e3a
.....
[46]crcstate : 0x8e3a
[47]crcstate : 0x8e3a
[0]crcfinal : 0x4983
[1]crcfinal : 0x4983
[2]crcfinal : 0x4983
.....
[46]crcfinal : 0x4983
[47]crcfinal : 0x4983
Correct operation validated. See readme.txt for run and reporting rules.
CoreMark 1.0 : 647991.900101 / GCC10.2.1 20210110 -static -Ofast -funroll-
all-loops -EL -falign-functions=16 -DUSE_PHTREAD -DMULTITHREAD=48 -
DPERFORMANCE_RUN=1 -lrt / Heap / 48:PThreads
2K performance run parameters for coremark.Вывод однопоточного теста Coremark процессора Kunpeng 920
2K performance run parameters for coremark.
CoreMark Size : 666
Total ticks : 16306
Total time (secs): 16.306000
Iterations/Sec : 18398.135656
Iterations : 300000
Compiler version : GCC9.3.0
Compiler flags : -Ofast -march=native -DPERFORMANCE_RUN=1 -DUSE_FORK=1 -lrt
Memory location : Please put data memory location here
(e.g. code in flash, data on heap etc)
seedcrc : 0xe9f5
[0]crclist : 0xe714
[0]crcmatrix : 0x1fd7
[0]crcstate : 0x8e3a
[0]crcfinal : 0xcc42
Correct operation validated. See README.md for run and reporting rules.
CoreMark 1.0 : 18398.135656 / GCC9.3.0 -Ofast -march=native -DPERFORMANCE_RUN=1 -DUSE_FORK=1 -lrt / Heap
Вывод многопоточного теста Coremark процессора Kunpeng 920
2K performance run parameters for coremark.
CoreMark Size : 666
Total ticks : 14867
Total time (secs): 14.867000
Iterations/Sec : 1291450.864330
Iterations : 19200000
Compiler version : GCCUbuntu Clang 12.0.0
Compiler flags : -Ofast -march=armv8.2-a -DPERFORMANCE_RUN=1 -DUSE_FORK=1 -lrt
Parallel Fork : 96
Memory location : Please put data memory location here
(e.g. code in flash, data on heap etc)
seedcrc : 0xe9f5
[0]crclist : 0xe714
[95]crcfinal : 0x4983
Correct operation validated. See README.md for run and reporting rules.
CoreMark 1.0 : 1291450.864330 / GCCUbuntu Clang 12.0.0 -Ofast -march=armv8.2-a -DPERFORMANCE_RUN=1 -DUSE_FORK=1 -lrt / Heap / 96:Fork
Dhrystone
Dhrystone достаточно древний тест 80х годов, написан на C. Тестирует целочисленную арифметику и работу со строками. Результаты измеряются в Dhrystone/s и DMIPS. (DMIPS = Dhrystone/s делить на 1757).
Вывод теста Dhrystone процессора Байкал-S
#####################################################
Dhrystone Benchmark 2.1 arm armv8.1-a optimized, Thu Feb 4 19:41:16 2010
Nanoseconds one Dhrystone run: 42.44
Dhrystones per Second: 23561181
VAX MIPS rating = 13409.89
Numeric results were correctВывод теста Dhrystone процессора Kunpeng 920
Dhrystone Benchmark, Version 2.1 (Language: C or C++)
Optimisation aarch64 armv8.2-a optimized
Register option not selected
10000 runs 0.00 seconds
100000 runs 0.00 seconds
1000000 runs 0.03 seconds
10000000 runs 0.28 seconds
20000000 runs 0.55 seconds
40000000 runs 1.10 seconds
80000000 runs 2.19 seconds
Final values (* implementation-dependent):
Int_Glob: O.K. 5 Bool_Glob: O.K. 1
Ch_1_Glob: O.K. A Ch_2_Glob: O.K. B
Arr_1_Glob[8]: O.K. 7 Arr_2_Glob8/7: O.K. 80000010
Ptr_Glob-> Ptr_Comp: * 0xc0cf2a0
Discr: O.K. 0 Enum_Comp: O.K. 2
Int_Comp: O.K. 17 Str_Comp: O.K. DHRYSTONE PROGRAM, SOME STRING
Next_Ptr_Glob-> Ptr_Comp: * 0xc0cf2a0 same as above
Discr: O.K. 0 Enum_Comp: O.K. 1
Int_Comp: O.K. 18 Str_Comp: O.K. DHRYSTONE PROGRAM, SOME STRING
Int_1_Loc: O.K. 5 Int_2_Loc: O.K. 13
Int_3_Loc: O.K. 7 Enum_Loc: O.K. 1
Str_1_Loc: O.K. DHRYSTONE PROGRAM, 1'ST STRING
Str_2_Loc: O.K. DHRYSTONE PROGRAM, 2'ND STRING
Nanoseconds one Dhrystone run: 27.32
Dhrystones per Second: 36598414
VAX MIPS rating = 20830.06
Whetstone
Тестирует арифметику с плавающей/фиксированной запятой, математические функции, ветвления, вызовов функций, присваиваний, работы с числами с фиксированной запятой, ветвлений. Результаты измеряются в MMIPS.
Вывод однопоточного теста Whetstone процессора Байкал-S
=======================================================================
BYTE UNIX Benchmarks (Version 5.1.3)
System: dbs: GNU/Linux
OS: GNU/Linux -- 5.4.197-baikal-arm64 -- #72 SMP PREEMPT Mon Sep 06
20:59:55 MSK 2022
Machine: aarch64 (unknown)
Language: en_US.utf8 (charmap="UTF-8". collate="UTF-8")
17:40:34 up 38 min. 2 users. load average: 8.44. 17.34. 9.11; runlevel 5
------------------------------------------------------------------------
Benchmark Run: Sep 09 2022 17:40:34 - 17:45:20
0 CPUs in system; running 1 parallel copy of tests
Double-Precision Whetstone 4702.4 MWIPS (9.9 s. 7 samples)Вывод многопоточного теста Whetstone процессора Байкал-S
BYTE UNIX Benchmarks (Version 5.1.3)
System: dbs: GNU/Linux
OS: GNU/Linux -- 5.4.197-baikal-arm64 -- #72 SMP PREEMPT Mon Jun 21
20:59:55 MSK 2022
Machine: aarch64 (unknown)
Language: en_US.utf8 (charmap="UTF-8". collate="UTF-8")
17:34:20 up 32 min. 2 users. load average: 0.01. 0.01. 0.00; runlevel 5
------------------------------------------------------------------------
Benchmark Run: Sep 09 2022 17:34:20 - 17:39:07
0 CPUs in system; running 48 parallel copies of tests
Double-Precision Whetstone 225078.7 MWIPS (9.9 s. 7 samples)Вывод однопоточного теста Whetstone процессора Kunpeng 920
##########################################
Single Precision C Whetstone Benchmark aarch64 armv8-a optimized, Tue Jan 18 14:59:09 2022
Calibrate
0.01 Seconds 1 Passes (x 100)
0.02 Seconds 5 Passes (x 100)
0.06 Seconds 25 Passes (x 100)
0.32 Seconds 125 Passes (x 100)
1.62 Seconds 625 Passes (x 100)
8.11 Seconds 3125 Passes (x 100)
Use 3854 passes (x 100)
Single Precision C/C++ Whetstone Benchmark
Loop content Result MFLOPS MOPS Seconds
N1 floating point -1.12475013732910156 831.272 0.089
N2 floating point -1.12274742126464844 804.396 0.644
N3 if then else 1.00000000000000000 3895.706 0.102
N4 fixed point 12.00000000000000000 62234.039 0.020
N5 sin,cos etc. 0.49911010265350342 115.110 2.786
N6 floating point 0.99999982118606567 519.520 4.001
N7 assignments 3.00000000000000000 2597.914 0.274
N8 exp,sqrt etc. 0.75110864639282227 69.101 2.075
MWIPS 3857.517 9.991
Вывод многопоточного теста Whetstone процессора Kunpeng 920
MWIPS 362558 Based on time for last thread to finishHPL
HPL – переносимый высокопроизводительный тест, который используется для суперкомпьютеров. Решает системы линейных уравнений, использует библиотеки BLAS и MPI. Результаты выдаёт в GFLOPS.
Вывод теста HPL процессора Байкал-S
mpirun.openmpi -np 48 --
allow-run-as-root ./xhpl
=============================================================================
HPLinpack 2.3 -- High-Performance Linpack benchmark -- December 2. 2018
Written by A. Petitet and R. Clint Whaley. Innovative Computing Laboratory.
UTK
Modified by Piotr Luszczek. Innovative Computing Laboratory. UTK
Modified by Julien Langou. University of Colorado Denver
=============================================================================
An explanation of the input/output parameters follows:
T/V : Wall time / encoded variant.
N : The order of the coefficient matrix A.
NB : The partitioning blocking factor.
P : The number of process rows.
Q : The number of process columns.
Time : Time in seconds to solve the linear system.
Gflops : Rate of execution for solving the linear system.
The following parameter values will be used:
N : 98304
NB : 96
PMAP : Row-major process mapping
P : 6
Q : 8
PFACT : Left
NBMIN : 4
NDIV : 2
RFACT : Crout
BCAST : 1ring
DEPTH : 0
SWAP : Mix (threshold = 64)
L1 : transposed form
U : transposed form
EQUIL : yes
ALIGN : 8 double precision words
-----------------------------------------------------------------------------
- The matrix A is randomly generated for each test.
- The following scaled residual check will be computed:
||Ax-b||_oo / ( eps * ( || x ||_oo * || A ||_oo + || b ||_oo ) * N )
- The relative machine precision (eps) is taken to be
1.110223e-16
- Computational tests pass if scaled residuals are less than
16.0
=============================================================================
T/V N NB P Q Time Gflops
-----------------------------------------------------------------------------
WR00C2L4 98304 96 6 8 2149.68 2.9462e+02
HPL_pdgesv() start time Wed Aug 17 18:56:05 2022
HPL_pdgesv() end time Wed Aug 17 19:31:55 2022
-----------------------------------------------------------------------------
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= 1.19811356e-01 ......
PASSED
=============================================================================Вывод теста HPL процессора Kunpeng 920
$ mpiexec -n 48 ./xhpl
HPL ERROR from process # 0, on line 419 of function HPL_pdinfo:
>>> Need at least 96 processes for these tests <<<
HPL ERROR from process # 0, on line 621 of function HPL_pdinfo:
>>> Illegal input in file HPL.dat. Exiting ... <<<
user@iot:~/hpl/bin/linux$ nano HPL.dat
user@iot:~/hpl/bin/linux$ mpiexec -n 48 ./xhpl
================================================================================
HPLinpack 2.3 -- High-Performance Linpack benchmark -- December 2, 2018
Written by A. Petitet and R. Clint Whaley, Innovative Computing Laboratory, UTK
Modified by Piotr Luszczek, Innovative Computing Laboratory, UTK
Modified by Julien Langou, University of Colorado Denver
================================================================================
An explanation of the input/output parameters follows:
T/V : Wall time / encoded variant.
N : The order of the coefficient matrix A.
NB : The partitioning blocking factor.
P : The number of process rows.
Q : The number of process columns.
Time : Time in seconds to solve the linear system.
Gflops : Rate of execution for solving the linear system.
The following parameter values will be used:
N : 18000 24000 36000 40800
NB : 240
PMAP : Row-major process mapping
P : 1
Q : 48
PFACT : Left
NBMIN : 4
NDIV : 4
RFACT : Left
BCAST : 1ring
DEPTH : 1
SWAP : Mix (threshold = 64)
L1 : transposed form
U : transposed form
EQUIL : yes
ALIGN : 16 double precision words
--------------------------------------------------------------------------------
- The matrix A is randomly generated for each test.
- The following scaled residual check will be computed:
||Ax-b||_oo / ( eps * ( || x ||_oo * || A ||_oo + || b ||_oo ) * N )
- The relative machine precision (eps) is taken to be 1.110223e-16
- Computational tests pass if scaled residuals are less than 16.0
================================================================================
T/V N NB P Q Time Gflops
--------------------------------------------------------------------------------
WR10L4L4 18000 240 1 48 29.67 1.3106e+02
HPL_pdgesv() start time Wed Jan 26 22:02:33 2022
HPL_pdgesv() end time Wed Jan 26 22:03:02 2022
--------------------------------------------------------------------------------
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= 1.42218782e-03 ...... PASSED
================================================================================
T/V N NB P Q Time Gflops
--------------------------------------------------------------------------------
WR10L4L4 24000 240 1 48 55.00 1.6757e+02
HPL_pdgesv() start time Wed Jan 26 22:03:06 2022
HPL_pdgesv() end time Wed Jan 26 22:04:01 2022
--------------------------------------------------------------------------------
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= 1.13864820e-03 ...... PASSED
================================================================================
T/V N NB P Q Time Gflops
--------------------------------------------------------------------------------
WR10L4L4 36000 240 1 48 164.06 1.8960e+02
HPL_pdgesv() start time Wed Jan 26 22:04:09 2022
HPL_pdgesv() end time Wed Jan 26 22:06:54 2022
--------------------------------------------------------------------------------
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= 9.12220668e-04 ...... PASSED
================================================================================
T/V N NB P Q Time Gflops
--------------------------------------------------------------------------------
WR10L4L4 40800 240 1 48 233.34 1.9406e+02
HPL_pdgesv() start time Wed Jan 26 22:07:05 2022
HPL_pdgesv() end time Wed Jan 26 22:10:58 2022
--------------------------------------------------------------------------------
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= 9.95497593e-04 ...... PASSED
================================================================================
Finished 4 tests with the following results:
4 tests completed and passed residual checks,
0 tests completed and failed residual checks,
0 tests skipped because of illegal input values.
--------------------------------------------------------------------------------
End of Tests.
================================================================================
7zip
Встроенный тест архиватора. Тест особо не параллелится, результаты примерно равные на этих процессорах (частота одинаковая).
Вывод теста 7z процессора Байкал-S
root@dbs:~# 7za b
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,48 CPUs LE)
LE
CPU Freq: 64000000 - - - - - - - -
RAM size: 96133 MB, # CPU hardware threads: 48
RAM usage: 10590 MB, # Benchmark threads: 48
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
22: 75097 4088 1787 73055 | 1295732 4780 2312 110496
23: 68987 3987 1763 70290 | 1263902 4784 2286 109363
24: 70152 4293 1757 75428 | 1244122 4783 2283 109200
25: 59027 4190 1610 67395 | 1180858 4659 2256 105084
---------------------------------- | ------------------------------
Avr: 4139 1729 71542 | 4752 2284 108536
Tot: 4446 2007 90039
root@dbs:~# 7za b -mmt1
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,48 CPUs LE)
LE
CPU Freq: - - - - - - - - -
RAM size: 96133 MB, # CPU hardware threads: 48
RAM usage: 435 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
22: 2664 100 2593 2592 | 27461 100 2345 2345
23: 2377 100 2423 2422 | 27216 100 2356 2356
24: 2153 100 2317 2316 | 26935 100 2365 2365
25: 1988 100 2271 2270 | 26524 100 2361 2361
---------------------------------- | ------------------------------
Avr: 100 2401 2400 | 100 2357 2356
Tot: 100 2379 2378Вывод теста 7z процессора Kunpeng 920
7zr b
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,96 CPUs LE)
LE
CPU Freq: - - - - - - - - -
RAM size: 321420 MB, # CPU hardware threads: 96
RAM usage: 21180 MB, # Benchmark threads: 96
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
22: 150159 7512 1945 146075 | 2459507 6909 3037 209731
23: 135991 7105 1950 138559 | 3118221 9048 2984 269812
24: 138858 6878 2171 149301 | 3013256 8945 2958 264461
25: 145816 7561 2203 166488 | 2384245 7438 2853 212162
---------------------------------- | ------------------------------
Avr: 7264 2067 150105 | 8085 2958 239042
Tot: 7675 2513 194573
$ 7zr b -mmt48
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,96 CPUs LE)
LE
CPU Freq: - 64000000 64000000 - 128000000 - 512000000 - -
RAM size: 321420 MB, # CPU hardware threads: 96
RAM usage: 10590 MB, # Benchmark threads: 48
Compressing | Decompressing
Dict Speed Usage R/U Rating | Speed Usage R/U Rating
KiB/s % MIPS MIPS | KiB/s % MIPS MIPS
22: 104046 4168 2429 101216 | 1691792 4670 3089 144270
23: 93868 4477 2136 95641 | 1613868 4618 3024 139645
24: 97195 4096 2552 104505 | 1588574 4629 3012 139434
25: 90848 4125 2515 103728 | 1451915 4396 2939 129206
---------------------------------- | ------------------------------
Avr: 4217 2408 101272 | 4578 3016 138139
Tot: 4397 2712 119705
$ 7zr b -mmt1
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,96 CPUs LE)
LE
CPU Freq: - - - - - - - - -
RAM size: 321420 MB, # CPU hardware threads: 96
RAM usage: 435 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
22: 3620 100 3527 3522 | 37243 100 3183 3180
23: 3226 100 3293 3288 | 36950 100 3201 3198
24: 2872 100 3092 3088 | 36444 100 3202 3199
25: 2591 100 2963 2959 | 35783 100 3188 3185
---------------------------------- | ------------------------------
Avr: 100 3219 3214 | 100 3194 3191
Tot: 100 3206 3202STREAM
Тест производительности ОЗУ.
Вывод теста STREAM процессора Байкал-S
Array size = 4947848500 (elements), Offset = 0 (elements)
Memory per array = 37749.1 MiB (= 36.9 GiB).
Total memory required = 113247.3 MiB (= 110.6 GiB).
Each kernel will be executed 3 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Number of Threads requested = 12
Number of Threads counted = 12
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 1196001 microseconds.
(= 1196001 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 81787.6 0.968455 0.967941 0.968970
Scale: 78037.2 1.015015 1.014459 1.015572
Add: 76156.5 1.559800 1.559267 1.560332
Triad: 75113.8 1.582214 1.580914 1.583514
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------Вывод теста STREAM процессора Kunpeng 920
Array size = 20000000 (elements), Offset = 0 (elements)
Memory per array = 152.6 MiB (= 0.1 GiB).
Total memory required = 457.8 MiB (= 0.4 GiB).
Each kernel will be executed 20 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Number of Threads requested = 96
Number of Threads counted = 96
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 3901 microseconds.
(= 3901 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 93271.5 0.004566 0.003431 0.006441
Scale: 75846.4 0.005011 0.004219 0.006355
Add: 78009.4 0.007845 0.006153 0.010400
Triad: 76593.7 0.007606 0.006267 0.009237
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arraysBlender
Вывод теста Blender процессора Байкал-S
blender -b RyzenGraphic_27.blend -f 1 -- --cycles-
device CPU
Blender 2.83.5
Fra:1 Mem:310.57M (0.00M. Peak 831.94M) | Time:00:16.02 | Mem:260.67M.
Peak:262.29M | Scene. RenderLayer | Rendered 625/625 Tiles
Fra:1 Mem:310.56M (0.00M. Peak 831.94M) | Time:00:16.02 | Mem:260.67M.
Peak:262.29M | Scene. RenderLayer | Finished
Fra:1 Mem:48.06M (0.00M. Peak 831.94M) | Time:00:16.05 | Sce: Scene Ve:0 Fa:0
La:0
Saved: '/tmp/0001.png'
Time: 00:16.96 (Saving: 00:00.91)Вывод теста Blender процессора Kunpeng 920
blender -b RyzenGraphic_27.blend -f 1
Fra:1 Mem:49.27M (0.00M, Peak 822.50M) | Time:00:07.54 | Sce: Scene Ve:0 Fa:0 La:0
Saved: '/tmp/0001.png'
Time: 00:08.20 (Saving: 00:00.65)
8 секунд на 96 ядрах
на 48 ядрах (2 проц):
Time: 00:13.56 (Saving: 00:00.66)
на 24 потоках:
Saved: '/tmp/0001.png'
Time: 00:24.92 (Saving: 00:00.65)
на 8 потоках:
Time: 01:10.24 (Saving: 00:00.65)