昇腾论坛

参加香橙派的尝鲜活动，自费买了一块8T 16G的板子，没想到活动又中了一块相同规格的板子套装。而且新收到的板子改掉了预售版本的关机风扇狂转的问题（电路优化，预售版无法通过软件修复问题）。再且，风扇运转噪音明显小了很多很多很多。总结，预售版用户哭了。不过好消息是，放出了1.6GHz固件给AIPro升级，说是能提升性能50%（单指控制CPU吧）。参考：https://www.hiascend.com/forum/thread-0261154700143118107-1-1.html ，发现确实AICORE从500MHz升到了750MHz，真50%提升。就是操作过程有些迷惑，和npu-smi指令的相关提示说明也对不上。不知是昇腾的问题还是香橙派的问题了。

本帖就跑个分看看吧。

固件

7.3.t10.0.b528

chmod +x Ascend310B-firmware-7.3.t10.0.b528-rc-signed-opiaipro-12t-1.6ghz-20240605.run
./Ascend310B-firmware-7.3.t10.0.b528-rc-signed-opiaipro-12t-1.6ghz-20240605.run --full

y 确认安装重启即可完成升级

核心分配

(base) root@orangepiaipro:~# npu-smi info -t cpu-num-cfg -i 0
        NPU ID                         : 0
        Chip Count                     : 1

        Chip ID                        : 0
        Current AI CPU number          : 1
        Current control CPU number     : 3
        Current data CPU number        : 0

用大佬提供的程序查询cpu频率

(base) root@orangepiaipro:~# ./cpu_info
Memory frequency is 1596 Mhz
Control cpu frequency is 1600 Mhz
AI CORE now frequency is 500 Mhz
AI CORE default frequency is 500 Mhz

另外，我测试了20TOPS 24GB 版本AIPro开发板的cpu频率没有变化，内存频率提升到了2131 MHz，AICore 提升到了1120 MHz 的水平。所以UnixBench看不太出来和8TOPS版本的差距。

UnixBench

项目地址：https://github.com/aliyun/byte-unixbench/releases

========================================================================
   BYTE UNIX Benchmarks (Version 5.1.6)

   System: orangepiaipro: GNU/Linux
   OS: GNU/Linux -- 5.10.0+ -- #1 SMP Mon Feb 26 15:29:07 CST 2024
   Machine: aarch64 (aarch64)
   Language: en_US.utf8 (charmap="ANSI_X3.4-1968", collate="ANSI_X3.4-1968")
   22:51:37 up 3 min,  2 users,  load average: 16.80, 9.05, 3.67; runlevel Jul

------------------------------------------------------------------------
Benchmark Run: Sat Jul 13 2024 22:51:37 - 23:19:44
4 CPUs in system; running 1 parallel copy of tests

Dhrystone 2 using register variables       19053006.1 lps   (10.0 s, 7 samples)
Double-Precision Whetstone                     4350.8 MWIPS (10.0 s, 7 samples)
Execl Throughput                                997.0 lps   (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks        513044.2 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks          184873.9 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks       1174981.8 KBps  (30.0 s, 2 samples)
Pipe Throughput                              874758.3 lps   (10.0 s, 7 samples)
Pipe-based Context Switching                 120343.8 lps   (10.0 s, 7 samples)
Process Creation                               5435.4 lps   (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                   3791.6 lpm   (60.0 s, 2 samples)
Shell Scripts (8 concurrent)                    922.4 lpm   (60.0 s, 2 samples)
System Call Overhead                         857484.4 lps   (10.0 s, 7 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0   19053006.1   1632.6
Double-Precision Whetstone                       55.0       4350.8    791.0
Execl Throughput                                 43.0        997.0    231.9
File Copy 1024 bufsize 2000 maxblocks          3960.0     513044.2   1295.6
File Copy 256 bufsize 500 maxblocks            1655.0     184873.9   1117.1
File Copy 4096 bufsize 8000 maxblocks          5800.0    1174981.8   2025.8
Pipe Throughput                               12440.0     874758.3    703.2
Pipe-based Context Switching                   4000.0     120343.8    300.9
Process Creation                                126.0       5435.4    431.4
Shell Scripts (1 concurrent)                     42.4       3791.6    894.3
Shell Scripts (8 concurrent)                      6.0        922.4   1537.3
System Call Overhead                          15000.0     857484.4    571.7
                                                                   ========
System Benchmarks Index Score                                         794.2

------------------------------------------------------------------------
Benchmark Run: Sat Jul 13 2024 23:19:44 - 23:47:57
4 CPUs in system; running 4 parallel copies of tests

Dhrystone 2 using register variables       56265407.7 lps   (10.0 s, 7 samples)
Double-Precision Whetstone                    14581.5 MWIPS (8.3 s, 7 samples)
Execl Throughput                               2847.8 lps   (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks        876262.5 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks          352453.2 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks       2407717.3 KBps  (30.0 s, 2 samples)
Pipe Throughput                             2601061.6 lps   (10.0 s, 7 samples)
Pipe-based Context Switching                 157180.6 lps   (10.0 s, 7 samples)
Process Creation                              12144.3 lps   (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                   7070.4 lpm   (60.0 s, 2 samples)
Shell Scripts (8 concurrent)                    960.1 lpm   (60.1 s, 2 samples)
System Call Overhead                        2273271.1 lps   (10.0 s, 7 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0   56265407.7   4821.4
Double-Precision Whetstone                       55.0      14581.5   2651.2
Execl Throughput                                 43.0       2847.8    662.3
File Copy 1024 bufsize 2000 maxblocks          3960.0     876262.5   2212.8
File Copy 256 bufsize 500 maxblocks            1655.0     352453.2   2129.6
File Copy 4096 bufsize 8000 maxblocks          5800.0    2407717.3   4151.2
Pipe Throughput                               12440.0    2601061.6   2090.9
Pipe-based Context Switching                   4000.0     157180.6    393.0
Process Creation                                126.0      12144.3    963.8
Shell Scripts (1 concurrent)                     42.4       7070.4   1667.5
Shell Scripts (8 concurrent)                      6.0        960.1   1600.2
System Call Overhead                          15000.0    2273271.1   1515.5
                                                                   ========
System Benchmarks Index Score                                        1686.7

更换为 4 CPU 后跑分

(base) root@orangepiaipro:~# npu-smi info -t cpu-num-cfg -i 0
        NPU ID                         : 0
        Chip Count                     : 1

        Chip ID                        : 0
        Current AI CPU number          : 0
        Current control CPU number     : 4
        Current data CPU number        : 0

========================================================================
   BYTE UNIX Benchmarks (Version 5.1.6)

   System: orangepiaipro: GNU/Linux
   OS: GNU/Linux -- 5.10.0+ -- #1 SMP Mon Feb 26 15:29:07 CST 2024
   Machine: aarch64 (aarch64)
   Language: en_US.utf8 (charmap="ANSI_X3.4-1968", collate="ANSI_X3.4-1968")
   23:58:57 up 2 min,  2 users,  load average: 16.91, 7.86, 3.03; runlevel Jul

------------------------------------------------------------------------
Benchmark Run: Sat Jul 13 2024 23:58:57 - 00:27:04
4 CPUs in system; running 1 parallel copy of tests

Dhrystone 2 using register variables       19099670.1 lps   (10.0 s, 7 samples)
Double-Precision Whetstone                     4360.5 MWIPS (10.0 s, 7 samples)
Execl Throughput                                987.2 lps   (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks        492803.3 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks          183286.4 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks       1185908.6 KBps  (30.0 s, 2 samples)
Pipe Throughput                              877132.1 lps   (10.0 s, 7 samples)
Pipe-based Context Switching                 121954.2 lps   (10.0 s, 7 samples)
Process Creation                               5399.1 lps   (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                   3843.3 lpm   (60.0 s, 2 samples)
Shell Scripts (8 concurrent)                   1158.8 lpm   (60.0 s, 2 samples)
System Call Overhead                         857158.9 lps   (10.0 s, 7 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0   19099670.1   1636.6
Double-Precision Whetstone                       55.0       4360.5    792.8
Execl Throughput                                 43.0        987.2    229.6
File Copy 1024 bufsize 2000 maxblocks          3960.0     492803.3   1244.5
File Copy 256 bufsize 500 maxblocks            1655.0     183286.4   1107.5
File Copy 4096 bufsize 8000 maxblocks          5800.0    1185908.6   2044.7
Pipe Throughput                               12440.0     877132.1    705.1
Pipe-based Context Switching                   4000.0     121954.2    304.9
Process Creation                                126.0       5399.1    428.5
Shell Scripts (1 concurrent)                     42.4       3843.3    906.4
Shell Scripts (8 concurrent)                      6.0       1158.8   1931.3
System Call Overhead                          15000.0     857158.9    571.4
                                                                   ========
System Benchmarks Index Score                                         807.9

------------------------------------------------------------------------
Benchmark Run: Sun Jul 14 2024 00:27:04 - 00:55:12
4 CPUs in system; running 4 parallel copies of tests

Dhrystone 2 using register variables       75890250.8 lps   (10.0 s, 7 samples)
Double-Precision Whetstone                    17348.1 MWIPS (10.0 s, 7 samples)
Execl Throughput                               3764.0 lps   (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks       1010705.0 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks          406011.5 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks       2678169.9 KBps  (30.0 s, 2 samples)
Pipe Throughput                             3474692.0 lps   (10.0 s, 7 samples)
Pipe-based Context Switching                 381460.3 lps   (10.0 s, 7 samples)
Process Creation                              14968.9 lps   (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                   9178.8 lpm   (60.0 s, 2 samples)
Shell Scripts (8 concurrent)                   1269.2 lpm   (60.1 s, 2 samples)
System Call Overhead                        3035355.1 lps   (10.0 s, 7 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0   75890250.8   6503.0
Double-Precision Whetstone                       55.0      17348.1   3154.2
Execl Throughput                                 43.0       3764.0    875.4
File Copy 1024 bufsize 2000 maxblocks          3960.0    1010705.0   2552.3
File Copy 256 bufsize 500 maxblocks            1655.0     406011.5   2453.2
File Copy 4096 bufsize 8000 maxblocks          5800.0    2678169.9   4617.5
Pipe Throughput                               12440.0    3474692.0   2793.2
Pipe-based Context Switching                   4000.0     381460.3    953.7
Process Creation                                126.0      14968.9   1188.0
Shell Scripts (1 concurrent)                     42.4       9178.8   2164.8
Shell Scripts (8 concurrent)                      6.0       1269.2   2115.4
System Call Overhead                          15000.0    3035355.1   2023.6
                                                                   ========
System Benchmarks Index Score                                        2231.2

END

跑一次还挺费时间的，约30分钟的样子？

如果不跑矩阵运算程序，切成4 CPU配置挺好的。（嗯？那我买它来作甚？）

另外，我还测试了一下双路至强4310的UnixBench跑分，单核在1000+，多核在10000+。

BUG？

当核心配置为 3:1:0 or 2:2:0 时。

UnixBench 执行到 "Pipe-based Context Switching" 会发生错误而中止。

求大神解释！

1 x Dhrystone 2 using register variables  1 2 3 4 5 6 7 8 9 10

1 x Double-Precision Whetstone  1 2 3 4 5 6 7 8 9 10

1 x Execl Throughput  1 2 3

1 x File Copy 1024 bufsize 2000 maxblocks  1 2 3

1 x File Copy 256 bufsize 500 maxblocks  1 2 3

1 x File Copy 4096 bufsize 8000 maxblocks  1 2 3

1 x Pipe Throughput  1 2 3 4 5 6 7 8 9 10

1 x Pipe-based Context Switching  1
**********************************************
Run: "Pipe-based Context Switching": child sched_setaffinity failed: Invalid argument; aborting

固件

核心分配

UnixBench

更换为 4 CPU 后跑分

END

BUG？

关于昇腾

新闻与活动

交流与资讯

支持与服务

开源社区