MetaComputing AI PC with Framework Laptop 13 硬件测评
MetaComputing AI PC + Framework Laptop 13 的组合，是一台极具特色和前瞻性的设备。它在多核CPU性能、系统响应速度、内存带宽和集成GPU性能上带来了远超传统ARM笔记本甚至部分X86轻薄本的惊喜表现。模块化设计赋予了它长久的生命力。
·
本次测试基于 MetaComputing AI PC with Framework Laptop 13，系统为 Debian GNU/Linux 12.11（bookworm）aarch64，内核版本 Linux 6.6.89。所有测试均在系统刚启动、无额外后台负载的情况下进行，结果更偏向理论性能上限，实际日常使用中会受功耗策略、温控与负载情况影响。
基础信息
系统信息
roma@debian ~> fastfetch
_,met$$$$$gg. roma@debian
,g$$$$$$$$$$$$$$$P. -----------
,g$$P"" """Y$$.". OS: Debian GNU/Linux 12.11 (bookworm) aarch64
,$$P' `$$$. Host: MC FML13V04 Board (1.0)
',$$P ,ggs. `$$b: Kernel: Linux 6.6.89-cix-build-generic
`d$$' ,$P"' . $$$ Uptime: 11 mins
$$P d$' , $$P Packages: 2632 (dpkg)
$$: $$. - ,d$$' Shell: fish 3.6.0
$$; Y$b._ _,d$P' Display (BOE095F): 2256x1504 @ 2x in 13", 60 Hz [Built-in]
Y$$. `.`"Y$$$$P"' DE: GNOME 43.9
`$$b "-.__ WM: Mutter (Wayland)
`Y$$b WM Theme: Adwaita
`Y$$. Theme: Adwaita [GTK2/3/4]
`$$b. Icons: Adwaita [GTK2/3/4]
`Y$$b. Font: Cantarell (11pt) [GTK2/3/4]
`"Y$b._ Cursor: Adwaita (24px)
`"""" Terminal: /dev/pts/0
CPU: Cortex-A720*2 + Cortex-A520*4 + Cortex-A720*6 (12) @ 2.60 GHz
GPU: Mali-G720-Immortalis [Integrated]
Memory: 7.74 GiB / 30.94 GiB (25%)
Swap: 0 B / 976.00 MiB (0%)
Disk (/): 21.29 GiB / 466.95 GiB (5%) - ext4
Local IP (wlan0): 192.168.50.38/24
Battery (ABC): 95% [AC Connected]
Locale: en_US.UTF-8
基准测试结果
系统响应、开关机测试
-
系统启动时间测试：系统从内核加载到进入图形界面总耗时 约 9.7 秒，其中 userspace 仅 6.1 秒 即达到 graphical.target。
roma@debian ~ > systemd-analyze
Startup finished in 3.587s (kernel) + 6.134s (userspace) = 9.722s
graphical.target reached after 6.075s in userspace.-
系统关闭时间测试：关机时间仅需约 46.45毫秒（实际断电时间）。这几乎是“瞬间”完成，体现了系统进程管理的效率和底层固件响应的迅速。用户完全无需等待。
roma@debian ~> time sudo shutdown -h now
Broadcast message from root@debian on pts/2 (Wed 2025-12-17 14:00:28 CST):
The system will power off now!
________________________________________________________
Executed in 46.45 millis fish external
usr time 3.12 millis 993.00 micros 2.13 millis
sys time 5.48 millis 0.00 micros 5.48 millis
-
系统响应时间测试
roma@debian ~> time ls
Desktop/ Documents/ Downloads/ Geekbench-6.2.2-LinuxARMPreview/ Music/ Pictures/ Public/ Templates/ tinymembench/ top500-benchmark/ Videos/
________________________________________________________
Executed in 6.07 millis fish external
usr time 0.84 millis 844.00 micros 0.00 millis
sys time 5.29 millis 696.00 micros 4.60 millisCPU
-
Geekbench 6
-
Geekbench 6跑分对比图（部分数据来源于网络（Bilibili、Youtube））
-
Coremark：单核性能已经明显超过传统嵌入式 ARM多；核扩展性良好，线程数拉高后仍能线性增长
-
单核测试：24832.38（单位：Iterations/Sec）
roma@debian ~/coremark (main)> make compile XCFLAGS="-DMULTITHREAD=1 -DUSE_PTHREAD" roma@debian ~/coremark (main)> ./coremark.exe 2K performance run parameters for coremark. CoreMark Size : 666 Total ticks : 12081 Total time (secs): 12.081000 Iterations/Sec : 24832.381425 Iterations : 300000 Compiler version : GCC12.2.0 Compiler flags : -O2 -DMULTITHREAD=1 -DUSE_PTHREAD -lrt -lpthread 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 : 24832.381425 / GCC12.2.0 -O2 -DMULTITHREAD=1 -DUSE_PTHREAD -lrt -lpthread / Heap-
多核测试：186003.26（单位：Iterations/Sec）
-
roma@debian ~/coremark (main)> make compile XCFLAGS="-DMULTITHREAD=64 -DUSE_PTHREAD"
roma@debian ~/coremark (main)> ./coremark.exe
2K performance run parameters for coremark.
CoreMark Size : 666
Total ticks : 103224
Total time (secs): 103.224000
Iterations/Sec : 186003.255057
Iterations : 19200000
Compiler version : GCC12.2.0
Compiler flags : -O2 -DMULTITHREAD=64 -DUSE_PTHREAD -lrt -lpthread
Parallel PThreads : 64
Memory location : Please put data memory location here
(e.g. code in flash, data on heap etc)
seedcrc : 0xe9f5
Correct operation validated. See README.md for run and reporting rules.
CoreMark 1.0 : 186003.255057 / GCC12.2.0 -O2 -DMULTITHREAD=64 -DUSE_PTHREAD -lrt -lpthread / Heap / 64:PThreads
GPU
-
Glmark2测试结果:得分 3845,这个分数需要结合场景和分辨率来看。在 800x600 窗口化下跑出近 5000 FPS 的单项测试，以及最终 3845 的总分，表明 Mali-G720 Immortalis 的性能远超以往的 ARM 集成显卡（如 Mali-G78/G710）。
roma@debian ~/coremark (main)> make compile XCFLAGS="-DMULTITHREAD=64 -DUSE_PTHREAD"
roma@debian ~/coremark (main)> ./coremark.exe
2K performance run parameters for coremark.
CoreMark Size : 666
Total ticks : 103224
Total time (secs): 103.224000
Iterations/Sec : 186003.255057
Iterations : 19200000
Compiler version : GCC12.2.0
Compiler flags : -O2 -DMULTITHREAD=64 -DUSE_PTHREAD -lrt -lpthread
Parallel PThreads : 64
Memory location : Please put data memory location here
(e.g. code in flash, data on heap etc)
seedcrc : 0xe9f5
Correct operation validated. See README.md for run and reporting rules.
CoreMark 1.0 : 186003.255057 / GCC12.2.0 -O2 -DMULTITHREAD=64 -DUSE_PTHREAD -lrt -lpthread / Heap / 64:PThreads
Memory
-
LPDDR5(32G)
-
高带宽： 多种复制 (
copy) 操作速度达到 14-16 GB/s，而填充 (fill) 操作更是飙升至 35-41 GB/s (如standard memset达 41.25 GB/s)。这明显是 LPDDR5 或 LPDDR5X 内存的特征，提供了充沛的数据吞吐能力，对 CPU 多核性能、GPU 纹理填充、大数据处理至关重要。 -
低延迟： 内存延迟测试 (
Memory latency test) 结果优秀。在常见应用使用的大小范围内（几MB到几十MB），额外延迟普遍在 几十纳秒 (ns) 级别。即使在最大的 64MB 块测试中，单/双随机读取延迟也仅为 194.1 ns / 228.1 ns (普通页) 和 175.6 ns / 210.6 ns (大页)。这确保了系统的快速响应。
tinymembench v0.4.9 (simple benchmark for memory throughput and latency)
==========================================================================
== Memory bandwidth tests ==
== ==
== Note 1: 1MB = 1000000 bytes ==
== Note 2: Results for 'copy' tests show how many bytes can be ==
== copied per second (adding together read and writen ==
== bytes would have provided twice higher numbers) ==
== Note 3: 2-pass copy means that we are using a small temporary buffer ==
== to first fetch data into it, and only then write it to the ==
== destination (source -> L1 cache, L1 cache -> destination) ==
== Note 4: If sample standard deviation exceeds 0.1%, it is shown in ==
== brackets ==
==========================================================================
C copy backwards : 14892.4 MB/s (1.0%)
C copy backwards (32 byte blocks) : 15205.4 MB/s (0.4%)
C copy backwards (64 byte blocks) : 15140.4 MB/s (0.3%)
C copy : 14994.8 MB/s (0.3%)
C copy prefetched (32 bytes step) : 14132.7 MB/s (0.2%)
C copy prefetched (64 bytes step) : 14301.2 MB/s (0.8%)
C 2-pass copy : 14294.2 MB/s
C 2-pass copy prefetched (32 bytes step) : 13718.6 MB/s
C 2-pass copy prefetched (64 bytes step) : 14677.6 MB/s (0.3%)
C fill : 36013.7 MB/s (2.1%)
C fill (shuffle within 16 byte blocks) : 35850.4 MB/s (2.0%)
C fill (shuffle within 32 byte blocks) : 35094.1 MB/s (0.5%)
C fill (shuffle within 64 byte blocks) : 35493.6 MB/s (0.4%)
---
standard memcpy : 16168.9 MB/s (1.9%)
standard memset : 41249.7 MB/s (5.9%)
---
NEON LDP/STP copy : 16664.6 MB/s (1.0%)
NEON LDP/STP copy pldl2strm (32 bytes step) : 15130.8 MB/s (0.4%)
NEON LDP/STP copy pldl2strm (64 bytes step) : 15379.7 MB/s (0.2%)
NEON LDP/STP copy pldl1keep (32 bytes step) : 15705.0 MB/s
NEON LDP/STP copy pldl1keep (64 bytes step) : 16216.1 MB/s (0.4%)
NEON LD1/ST1 copy : 16102.1 MB/s (0.2%)
NEON STP fill : 37033.2 MB/s (1.9%)
NEON STNP fill : 34190.7 MB/s (2.6%)
ARM LDP/STP copy : 16468.4 MB/s (1.9%)
ARM STP fill : 39243.4 MB/s (2.7%)
ARM STNP fill : 33605.0 MB/s (1.6%)
==========================================================================
== Framebuffer read tests. ==
== ==
== Many ARM devices use a part of the system memory as the framebuffer, ==
== typically mapped as uncached but with write-combining enabled. ==
== Writes to such framebuffers are quite fast, but reads are much ==
== slower and very sensitive to the alignment and the selection of ==
== CPU instructions which are used for accessing memory. ==
== ==
== Many x86 systems allocate the framebuffer in the GPU memory, ==
== accessible for the CPU via a relatively slow PCI-E bus. Moreover, ==
== PCI-E is asymmetric and handles reads a lot worse than writes. ==
== ==
== If uncached framebuffer reads are reasonably fast (at least 100 MB/s ==
== or preferably >300 MB/s), then using the shadow framebuffer layer ==
== is not necessary in Xorg DDX drivers, resulting in a nice overall ==
== performance improvement. For example, the xf86-video-fbturbo DDX ==
== uses this trick. ==
==========================================================================
NEON LDP/STP copy (from framebuffer) : 2235.3 MB/s
NEON LDP/STP 2-pass copy (from framebuffer) : 1515.1 MB/s
NEON LD1/ST1 copy (from framebuffer) : 2231.8 MB/s
NEON LD1/ST1 2-pass copy (from framebuffer) : 1525.6 MB/s
ARM LDP/STP copy (from framebuffer) : 1911.1 MB/s
ARM LDP/STP 2-pass copy (from framebuffer) : 1175.7 MB/s
==========================================================================
== Memory latency test ==
== ==
== Average time is measured for random memory accesses in the buffers ==
== of different sizes. The larger is the buffer, the more significant ==
== are relative contributions of TLB, L1/L2 cache misses and SDRAM ==
== accesses. For extremely large buffer sizes we are expecting to see ==
== page table walk with several requests to SDRAM for almost every ==
== memory access (though 64MiB is not nearly large enough to experience ==
== this effect to its fullest). ==
== ==
== Note 1: All the numbers are representing extra time, which needs to ==
== be added to L1 cache latency. The cycle timings for L1 cache ==
== latency can be usually found in the processor documentation. ==
== Note 2: Dual random read means that we are simultaneously performing ==
== two independent memory accesses at a time. In the case if ==
== the memory subsystem can't handle multiple outstanding ==
== requests, dual random read has the same timings as two ==
== single reads performed one after another. ==
==========================================================================
block size : single random read / dual random read, [MADV_NOHUGEPAGE]
1024 : 0.0 ns / 0.0 ns
2048 : 0.0 ns / 0.0 ns
4096 : 0.0 ns / 0.0 ns
8192 : 0.0 ns / 0.0 ns
16384 : 0.0 ns / 0.0 ns
32768 : 0.0 ns / 0.0 ns
65536 : 0.0 ns / 0.0 ns
131072 : 1.0 ns / 1.5 ns
262144 : 2.0 ns / 2.8 ns
524288 : 3.0 ns / 3.9 ns
1048576 : 19.3 ns / 27.6 ns
2097152 : 27.1 ns / 34.1 ns
4194304 : 31.9 ns / 36.2 ns
8388608 : 37.2 ns / 37.3 ns
16777216 : 52.8 ns / 50.5 ns
33554432 : 128.8 ns / 168.6 ns
67108864 : 194.1 ns / 228.1 ns
block size : single random read / dual random read, [MADV_HUGEPAGE]
1024 : 0.0 ns / 0.0 ns
2048 : 0.0 ns / 0.0 ns
4096 : 0.0 ns / 0.0 ns
8192 : 0.0 ns / 0.0 ns
16384 : 0.0 ns / 0.0 ns
32768 : 0.0 ns / 0.0 ns
65536 : 0.0 ns / 0.0 ns
131072 : 1.0 ns / 1.5 ns
262144 : 1.5 ns / 2.1 ns
524288 : 1.8 ns / 2.4 ns
1048576 : 17.5 ns / 26.1 ns
2097152 : 25.2 ns / 32.3 ns
4194304 : 28.9 ns / 34.3 ns
8388608 : 31.1 ns / 35.0 ns
16777216 : 41.8 ns / 47.3 ns
33554432 : 123.6 ns / 163.9 ns
67108864 : 175.6 ns / 210.6 ns
Disk
-
ZHITAI TiPlus7100 512GB ,iozone 基准测试
-
顺序读写 (1M)： 表现强劲，读取 ~4.9 GB/s，写入 ~4.6 GB/s。这达到了 PCIe 4.0 x4 NVMe SSD 的主流高性能水平，确保了大文件传输、视频加载、系统启动/休眠的速度。
-
4K 随机读写： 读取 ~74.5 MB/s，写入 ~203.7 MB/s。写入性能不错，读取性能相对普通。这反映了 SSD 控制器和闪存颗粒的特性。对于日常使用（系统操作、程序加载）影响不大，但对于高并发随机读密集的数据库类应用会是瓶颈。
|
基准 |
结果(MB/s) |
|
iozone 4K 随机写入 |
203.7 |
|
iozone 4K 随机读取 |
74.5 |
|
iozone 4K 顺序写入 |
141.8 |
|
iozone 4K 顺序读取 |
195.5 |
|
iozone 1M 随机写入 |
4,677.2 |
|
iozone 1M 随机读取 |
4,240.2 |
|
iozone 1M 顺序写入 |
4,575.3 |
|
iozone 1M 顺序读取 |
4,911.3 |
WLAN（Wi-Fi 6 AX210 160MHz）
-
测试结果受路由器和网络环境影响较大。单向上传 (TX) 达到 355 Mbps 和下载 (RX) 215 Mbps，表明 Intel AX210 网卡在良好的 Wi-Fi 6/6E (160MHz) 环境下能提供接近有线千兆网的无线速度。同时收发时性能有所下降是正常现象，但吞吐量总和 (278 Mbps) 依然可观，满足高清视频流、大文件同步、在线协作等需求绰绰有余。
|
测试项 |
结果 |
|
iperf3 发 |
355 Mbits/sec |
|
iperf3 收 |
215 Mbits/sec |
|
iperf3 同时收发 |
上传 (TX)：161 Mbits/sec |
|
下载 (RX)：117 Mbits/sec |
总结
MetaComputing AI PC + Framework Laptop 13 的组合，是一台极具特色和前瞻性的设备。它在多核CPU性能、系统响应速度、内存带宽和集成GPU性能上带来了远超传统ARM笔记本甚至部分X86轻薄本的惊喜表现。模块化设计赋予了它长久的生命力。
有“AI”的1024 = 2048，欢迎大家加入2048 AI社区
更多推荐
21
0- 0
已为社区贡献2条内容
所有评论(0)