I’ve already reviewed the Khadas Mind 2 AI Maker Kit with Windows 11 Home, and today, I’ll report my experience with Linux on the Intel Core Ultra 7 258V mini PC using Ubuntu 24.10 operating system.
I would usually review systems with Ubuntu 24.04 LTS OS, but considering the Intel Core Ultra 7 258V SoC is so new, I installed Ubuntu 24.10 when I tested whether disabling VT-d (IOMMU) would improve Intel Arc GPU performance (it does to some extent), and it turns out it was a good decision because Ubuntu 24.04 requires lots of fixes and workarounds to work the Khadas Mind 2 AI Maker Kit, at least until Ubuntu 24.04.2 is released later this month with a more recent kernel.
Khadas Mind 2 AI Maker Kit – Ubuntu 24.10 system information
My Ubuntu 24.10 installed has both Linux 6.11 (default) and Linux 6.13 kernels, but I did most of the tests using Linux 6.11 since it’s what comes with Ubuntu 24.10. The About window confirms we have a KHADAS Mind 2 AI Maker Kit running Ubuntu 24.10 on an octa-core Intel Core Ultra 7 258V processor with 32GB RAM and 1.0 TB of storage.
We can get a few more details in a terminal window:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ cat /etc/lsb-release DISTRIB_ID=Ubuntu DISTRIB_RELEASE=24.10 DISTRIB_CODENAME=oracular DISTRIB_DESCRIPTION="Ubuntu 24.10" jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ uname -a Linux Mind-2-AI-Maker-Kit-CNX 6.11.0-13-generic #14-Ubuntu SMP PREEMPT_DYNAMIC Sat Nov 30 23:51:51 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ df -h Filesystem Size Used Avail Use% Mounted on tmpfs 3.1G 2.2M 3.1G 1% /run /dev/nvme0n1p5 384G 21G 344G 6% / tmpfs 16G 0 16G 0% /dev/shm efivarfs 192K 102K 86K 55% /sys/firmware/efi/efivars tmpfs 5.0M 8.0K 5.0M 1% /run/lock tmpfs 1.0M 0 1.0M 0% /run/credentials/systemd-journald.service tmpfs 1.0M 0 1.0M 0% /run/credentials/systemd-udev-load-credentials.service tmpfs 1.0M 0 1.0M 0% /run/credentials/systemd-tmpfiles-setup-dev-early.service tmpfs 1.0M 0 1.0M 0% /run/credentials/systemd-tmpfiles-setup-dev.service tmpfs 1.0M 0 1.0M 0% /run/credentials/systemd-sysctl.service /dev/nvme0n1p1 96M 86M 11M 90% /boot/efi tmpfs 16G 8.0K 16G 1% /tmp tmpfs 1.0M 0 1.0M 0% /run/credentials/systemd-tmpfiles-setup.service tmpfs 1.0M 0 1.0M 0% /run/credentials/systemd-resolved.service tmpfs 3.1G 92K 3.1G 1% /run/user/120 tmpfs 3.1G 80K 3.1G 1% /run/user/1000 jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ free -h total used free shared buff/cache available Mem: 30Gi 1.3Gi 28Gi 81Mi 863Mi 29Gi Swap: 8.0Gi 0B 8.0Gi |
And even more with inxi utility:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ sudo inxi -Fc0 System: Host: Mind-2-AI-Maker-Kit-CNX Kernel: 6.11.0-13-generic arch: x86_64 bits: 64 Console: pty pts/1 Distro: Ubuntu 24.10 (Oracular Oriole) Machine: Type: Laptop System: KHADAS product: Mind 2 AI Maker Kit v: N/A serial: 1600014A000099 Mobo: KHADAS model: MIND-2-AI-MAKER-KIT-K1019 v: V11 serial: N/A UEFI: American Megatrends LLC. v: 1.02 date: 10/31/2024 Battery: ID-1: BAT1 charge: 5.6 Wh (100.0%) condition: 5.6/5.4 Wh (103.2%) CPU: Info: 8-core model: Intel Core Ultra 7 258V bits: 64 type: MCP cache: L2: 14 MiB Speed (MHz): avg: 852 min/max: 400/4700:4800:3700 cores: 1: 400 2: 4017 3: 400 4: 400 5: 400 6: 400 7: 400 8: 400 Graphics: Device-1: Intel Lunar Lake [Intel Arc Graphics 130V / 140V] driver: xe v: kernel Display: unspecified server: X.org v: 1.21.1.13 with: Xwayland v: 24.1.2 driver: gpu: xe tty: 80x24 resolution: 1920x1080 API: EGL v: 1.5 drivers: iris,swrast platforms: gbm,surfaceless,device API: OpenGL v: 4.6 compat-v: 4.5 vendor: mesa v: 24.2.3-1ubuntu1 note: console (EGL sourced) renderer: Mesa Intel Graphics (LNL), llvmpipe (LLVM 19.1.0 256 bits) Audio: Device-1: Intel Lunar Lake-M HD Audio driver: snd_hda_intel API: ALSA v: k6.11.0-13-generic status: kernel-api Network: Device-1: Intel driver: N/A Device-2: ASIX AX88772B driver: asix type: USB IF: enx00606e435f8a state: up speed: 100 Mbps duplex: full mac: 00:60:6e:43:5f:8a Bluetooth: Device-1: Intel AX211 Bluetooth driver: btusb type: USB Report: hciconfig ID: hci0 state: up address: BC:38:98:04:62:0C bt-v: 5.3 Drives: Local Storage: total: 953.87 GiB used: 20.58 GiB (2.2%) ID-1: /dev/nvme0n1 vendor: Western Digital model: WD PC SN740 SDDPTQD-1T00 size: 953.87 GiB Partition: ID-1: / size: 383.43 GiB used: 20.5 GiB (5.3%) fs: ext4 dev: /dev/nvme0n1p5 ID-2: /boot/efi size: 96 MiB used: 85.7 MiB (89.3%) fs: vfat dev: /dev/nvme0n1p1 Swap: ID-1: swap-1 type: file size: 8 GiB used: 0 KiB (0.0%) file: /swap.img Sensors: System Temperatures: cpu: 36.0 C mobo: N/A Fan Speeds (rpm): N/A Info: Memory: total: 32 GiB note: est. available: 30.42 GiB used: 1.29 GiB (4.2%) Processes: 300 Uptime: 0m Init: systemd target: graphical (5) Shell: Sudo inxi: 3.3.35 |
Everything is there, except WiFi which just shows as “Device-1: Intel driver: N/A”. That’s why I’ll be using a 100Mbps Ethernet USB adapter before fixing WiFI on the Intel Core Ultra 7 258V mini PC.
Ubuntu 24.04 benchmarks on Khadas Mind 2 AI Maker Kit
As usual, I’ll start with Thomas Kaiser’s sbc-bench.sh script:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ sudo ./sbc-bench.sh -r Starting to examine hardware/software for review purposes... sbc-bench v0.9.69 Installing needed tools: distro packages already installed, cpuminer. Done. Checking cpufreq OPP. Done. Executing tinymembench. Done. Executing RAM latency tester. Done. Executing OpenSSL benchmark. Done. Executing 7-zip benchmark. Done. Throttling test: heating up the device, 5 more minutes to wait. Done. Checking cpufreq OPP again. Done (14 minutes elapsed). Results validation: * Advertised vs. measured max CPU clockspeed: -1.2% before, -53.4% after -> https://tinyurl.com/32w9rr94 * No swapping * Background activity (%system) OK * Powercap detected. Details: "sudo powercap-info -p intel-rapl" -> https://tinyurl.com/4jh9nevj Full results uploaded to https://0x0.st/8Xve.bin # KHADAS Mind 2 AI Maker Kit / Ultra 7 258V Tested with sbc-bench v0.9.69 on Sun, 26 Jan 2025 15:12:39 +0700. Full info: [https://0x0.st/8Xve.bin](http://0x0.st/8Xve.bin) ### General information: The CPU features 2 clusters of different core types: Ultra 7 258V, Kernel: x86_64, Userland: amd64 CPU sysfs topology (clusters, cpufreq members, clockspeeds) cpufreq min max CPU cluster policy speed speed core type 0 0 0 400 4700 Lion Cove 1 0 1 400 4800 Lion Cove 2 0 2 400 4800 Lion Cove 3 0 3 400 4700 Lion Cove 4 0 4 400 3700 Skymont 5 0 5 400 3700 Skymont 6 0 6 400 3700 Skymont 7 0 7 400 3700 Skymont 31153 KB available RAM ### Policies (performance vs. idle consumption): Status of performance related policies found below /sys: /sys/module/pcie_aspm/parameters/policy: default [performance] powersave powersupersave ### Clockspeeds (idle vs. heated up): Before at 40.0°C: cpu0-cpu3 (Lion Cove): OPP: 4700, Measured: 4685 cpu4-cpu7 (Skymont): OPP: 3700, Measured: 3655 (-1.2%) After at 55.0°C: cpu0-cpu3 (Lion Cove): OPP: 4700, Measured: 2188 (-53.4%) cpu4-cpu7 (Skymont): OPP: 3700, Measured: 3654 (-1.2%) ### Performance baseline * cpu0 (Lion Cove): memcpy: 25504.6 MB/s, memchr: 30358.1 MB/s, memset: 65157.0 MB/s * cpu4 (Skymont): memcpy: 13578.6 MB/s, memchr: 16053.4 MB/s, memset: 23334.7 MB/s * cpu0 (Lion Cove) 16M latency: 30.19 24.89 29.82 24.91 29.85 27.20 25.95 34.48 * cpu4 (Skymont) 16M latency: 152.5 132.0 152.9 139.5 152.9 125.3 112.1 107.4 * cpu0 (Lion Cove) 128M latency: 126.4 113.3 123.4 111.8 121.9 103.9 96.43 87.33 * cpu4 (Skymont) 128M latency: 195.2 194.6 199.5 190.5 196.6 184.0 184.7 186.0 * 7-zip MIPS (3 consecutive runs): 22093, 18967, 18871 (19980 avg), single-threaded: 5797 * `aes-256-cbc 1440087.73k 1611845.85k 1651938.30k 1657943.72k 1665280.68k 1665362.60k (Lion Cove)` * `aes-256-cbc 1118106.86k 1264731.52k 1306177.37k 1316968.11k 1320564.05k 1321020.07k (Skymont)` ### PCIe and storage devices: * Intel Lunar Lake [Intel Arc Graphics 130V / 140V] (Onboard - Video): driver in use: xe * Intel Lunar Lake-M Thunderbolt 4 USB (Onboard - Other): driver in use: xhci_hcd * Intel Lunar Lake-M Thunderbolt 4 NHI #0 (Onboard - Other): driver in use: thunderbolt * Intel Lunar Lake-M USB 3.2 Gen 2x1 xHCI Host (Onboard - Other): driver in use: xhci_hcd * Intel Device a840 (Onboard - Ethernet): driver in use: * Intel Lunar Lake-M HD Audio (Onboard - Sound): driver in use: snd_hda_intel * 953.9GB "WD PC SN740 SDDPTQD-1T00" SSD as /dev/nvme0: Speed 16GT/s, Width x4, 0% worn out, 2 error log entries, drive temp: 45°C, ASPM Disabled * Winbond W25Q256JW 32MB SPI NOR flash, drivers in use: spi-nor/intel-spi "nvme error-log /dev/nvme0 ; smartctl -x /dev/nvme0" could be used to get further information about the reported issues. ### Challenging filesystems: The following partitions are NTFS: nvme0n1p3,nvme0n1p4 -> https://tinyurl.com/mv7wvzct ### Swap configuration: * /swap.img on /dev/nvme0n1p5: 8.0G (0K used) ### Software versions: * Ubuntu 24.10 (oracular) * Compiler: /usr/bin/gcc (Ubuntu 14.2.0-4ubuntu2) 14.2.0 / x86_64-linux-gnu * OpenSSL 3.3.1, built on 4 Jun 2024 (Library: OpenSSL 3.3.1 4 Jun 2024) ### Kernel info: * `/proc/cmdline: BOOT_IMAGE=/boot/vmlinuz-6.11.0-13-generic root=UUID=c4b6c017-6b2e-4679-8bc1-122fb68a434e ro intel_iommu=on quiet splash crashkernel=2G-4G:320M,4G-32G:512M,32G-64G:1024M,64G-128G:2048M,128G-:4096M vt.handoff=7` * Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl * Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization * Kernel 6.11.0-13-generic / CONFIG_HZ=1000 Waiting for the device to cool down...................................... 42.0°C^C |
The Khadas Mind 2 AI Maker Kit may not be able to extra the full performance of the Intel Core Ultra 7 258V under load, as while the CPU temperature never gets over 69.0°C (in the single-core OpenSSL benchmark), the CPU frequency drops sharply in multi-core benchmarks (4700 MHz to 2200 MHz) and the temperature is only up to 59.0°C in that test pointing so suboptimal settings.
The script tells us to check the thermal limits, so let’s do that:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 |
sudo powercap-info -p intel-rapl enabled: 1 Zone 0 name: package-0 enabled: 1 max_energy_range_uj: 262143328850 energy_uj: 19822811924 Constraint 0 name: long_term power_limit_uw: 200000000 time_window_us: 31981568 max_power_uw: 17000000 Constraint 1 name: short_term power_limit_uw: 37000000 time_window_us: 976 max_power_uw: 0 Zone 0:0 name: core enabled: 0 max_energy_range_uj: 262143328850 energy_uj: 5941268176 Constraint 0 name: long_term power_limit_uw: 0 time_window_us: 976 Zone 0:1 name: uncore enabled: 0 max_energy_range_uj: 262143328850 energy_uj: 1783259 Constraint 0 name: long_term power_limit_uw: 0 time_window_us: 976 Zone 0:2 name: dram enabled: 0 max_energy_range_uj: 262143328850 energy_uj: 146804433 Constraint 0 name: long_term power_limit_uw: 0 time_window_us: 976 Zone 1 name: psys enabled: 1 max_energy_range_uj: 262143328850 energy_uj: 32540973579 Constraint 0 name: long_term power_limit_uw: 17000000 time_window_us: 27983872 Constraint 1 name: short_term power_limit_uw: 21250000 time_window_us: 976 |
The PL1 and PL2 power limits are set to 200W!? (PBP) and 37W (MTP), while PL4 (Peak power) is not listed at all… For reference, we had PL1, PL2, and PL4 set to 30W, 37W, and 95W respectively in Windows 11 Home. So something is odd here. I rebooted to Linux 6.13-rc1 to check again, but the results are the same. Back to Linux 6.11, I decided to change the power limits for Zone 1 (PSYS) to match the limits on Windows 11:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ sudo powercap-set intel-rapl -z 1 -c 0 -l 30000000 jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ sudo powercap-set intel-rapl -z 1 -c 1 -l 37000000 jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ sudo powercap-info -p intel-rapl enabled: 1 Zone 0 name: package-0 enabled: 1 max_energy_range_uj: 262143328850 energy_uj: 1319377847 Constraint 0 name: long_term power_limit_uw: 200000000 time_window_us: 31981568 max_power_uw: 17000000 Constraint 1 name: short_term power_limit_uw: 37000000 time_window_us: 976 max_power_uw: 0 Zone 0:0 name: core enabled: 0 max_energy_range_uj: 262143328850 energy_uj: 649695907 Constraint 0 name: long_term power_limit_uw: 0 time_window_us: 976 Zone 0:1 name: dram enabled: 0 max_energy_range_uj: 262143328850 energy_uj: 476649048 Constraint 0 name: long_term power_limit_uw: 0 time_window_us: 976 Zone 1 name: psys enabled: 1 max_energy_range_uj: 262143328850 energy_uj: 1300823024 Constraint 0 name: long_term power_limit_uw: 30000000 time_window_us: 27983872 Constraint 1 name: short_term power_limit_uw: 37000000 time_window_us: 976 |
Let’s run sbc-bench.sh again:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ sudo ./sbc-bench.sh -r Starting to examine hardware/software for review purposes... sbc-bench v0.9.69 Installing needed tools: distro packages already installed. Done. Checking cpufreq OPP. Done. Executing tinymembench. Done. Executing RAM latency tester. Done. Executing OpenSSL benchmark. Done. Executing 7-zip benchmark. Done. Throttling test: heating up the device, 5 more minutes to wait. Done. Checking cpufreq OPP again. Done (13 minutes elapsed). Results validation: * Advertised vs. measured max CPU clockspeed: -1.2% before, -1.2% after -> https://tinyurl.com/32w9rr94 * No swapping * Background activity (%system) OK * Powercap detected. Details: "sudo powercap-info -p intel-rapl" -> https://tinyurl.com/4jh9nevj Full results uploaded to https://0x0.st/8Ko-.bin # KHADAS Mind 2 AI Maker Kit / Ultra 7 258V Tested with sbc-bench v0.9.69 on Sun, 02 Feb 2025 14:15:36 +0700. Full info: [https://0x0.st/8Ko-.bin](http://0x0.st/8Ko-.bin) ### General information: The CPU features 2 clusters of different core types: Ultra 7 258V, Kernel: x86_64, Userland: amd64 CPU sysfs topology (clusters, cpufreq members, clockspeeds) cpufreq min max CPU cluster policy speed speed core type 0 0 0 400 4700 Lion Cove 1 0 1 400 4800 Lion Cove 2 0 2 400 4800 Lion Cove 3 0 3 400 4700 Lion Cove 4 0 4 400 3700 Skymont 5 0 5 400 3700 Skymont 6 0 6 400 3700 Skymont 7 0 7 400 3700 Skymont 31153 KB available RAM ### Policies (performance vs. idle consumption): Status of performance related policies found below /sys: /sys/module/pcie_aspm/parameters/policy: default [performance] powersave powersupersave ### Clockspeeds (idle vs. heated up): Before at 40.0°C: cpu0-cpu3 (Lion Cove): OPP: 4700, Measured: 4685 cpu4-cpu7 (Skymont): OPP: 3700, Measured: 3656 (-1.2%) After at 66.0°C: cpu0-cpu3 (Lion Cove): OPP: 4700, Measured: 4685 cpu4-cpu7 (Skymont): OPP: 3700, Measured: 3654 (-1.2%) ### Performance baseline * cpu0 (Lion Cove): memcpy: 30172.9 MB/s, memchr: 39503.0 MB/s, memset: 99951.2 MB/s * cpu4 (Skymont): memcpy: 13892.6 MB/s, memchr: 16834.2 MB/s, memset: 23418.1 MB/s * cpu0 (Lion Cove) 16M latency: 31.41 24.92 31.32 25.03 31.37 26.91 25.82 34.49 * cpu4 (Skymont) 16M latency: 159.3 139.1 159.4 141.9 159.7 126.8 111.3 103.7 * cpu0 (Lion Cove) 128M latency: 123.2 109.0 123.0 112.2 124.0 106.7 96.59 86.69 * cpu4 (Skymont) 128M latency: 197.2 199.4 201.7 197.6 201.2 185.7 189.8 187.5 * 7-zip MIPS (3 consecutive runs): 32768, 31335, 30330 (31480 avg), single-threaded: 5807 * `aes-256-cbc 1440358.14k 1612205.21k 1652203.43k 1662480.04k 1665529.17k 1661348.52k (Lion Cove)` * `aes-256-cbc 1122405.06k 1268959.79k 1311087.87k 1322728.79k 1325938.01k 1327579.14k (Skymont)` ### PCIe and storage devices: * Intel Lunar Lake [Intel Arc Graphics 130V / 140V] (Onboard - Video): driver in use: xe * Intel Lunar Lake-M Thunderbolt 4 USB (Onboard - Other): driver in use: xhci_hcd * Intel Lunar Lake-M Thunderbolt 4 NHI #0 (Onboard - Other): driver in use: thunderbolt * Intel Lunar Lake-M USB 3.2 Gen 2x1 xHCI Host (Onboard - Other): driver in use: xhci_hcd * Intel BE201 320MHz (Onboard - Ethernet): driver in use: * Intel Lunar Lake-M HD Audio (Onboard - Sound): driver in use: snd_hda_intel * 953.9GB "WD PC SN740 SDDPTQD-1T00" SSD as /dev/nvme0: Speed 16GT/s, Width x4, 0% worn out, 2 error log entries, drive temp: 47°C, ASPM Disabled * Winbond W25Q256JW 32MB SPI NOR flash, drivers in use: spi-nor/intel-spi "nvme error-log /dev/nvme0 ; smartctl -x /dev/nvme0" could be used to get further information about the reported issues. ### Challenging filesystems: The following partitions are NTFS: nvme0n1p3,nvme0n1p4 -> https://tinyurl.com/mv7wvzct ### Swap configuration: * /swap.img on /dev/nvme0n1p5: 8.0G (0K used) ### Software versions: * Ubuntu 24.10 (oracular) * Compiler: /usr/bin/gcc (Ubuntu 14.2.0-4ubuntu2) 14.2.0 / x86_64-linux-gnu * OpenSSL 3.3.1, built on 4 Jun 2024 (Library: OpenSSL 3.3.1 4 Jun 2024) ### Kernel info: * `/proc/cmdline: BOOT_IMAGE=/boot/vmlinuz-6.11.0-14-generic root=UUID=c4b6c017-6b2e-4679-8bc1-122fb68a434e ro intel_iommu=on quiet splash crashkernel=2G-4G:320M,4G-32G:512M,32G-64G:1024M,64G-128G:2048M,128G-:4096M vt.handoff=7` * Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl * Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization * Kernel 6.11.0-14-generic / CONFIG_HZ=1000 Waiting for the device to cool down............................... 47.0°C |
The 7-zip score is now much better at 31,480 MIPS on average, so there’s some work to be done on the BIOS or Linux kernel side to fix the default values… As a side note, the Orion O6 motherboard with a 12-core Armv9 processor achieved a similar score (31,060 MIPS) in the same 7-zip benchmarks. Arm has come a long way.
Let’s carry on testing with Geekbench 6.3.0 single-core and multi-core benchmarks.
That would be 2829 points for the single-core score and 7014 points for the multi-core score before changing the power limits. But what about after?
Similar single-core score (2882) and a much better multi-core score (9618).
I’ve also run Geekbench AI 1.2.0. to check whether the Intel AI Boost accelerator is supported, but by default, the benchmark relies on the TensorFlow Lite framework and runs on the CPU…
You’ll find the full results on the Geekbench website, even though they are not super interesting since it’s CPU only, and we’d like it to run on the GPU and/or NPU. So let’s check the options available in Linux:
1 2 3 4 5 6 7 8 9 10 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~/GeekbenchAI-1.2.0-Linux$ ./banff --ai-list Geekbench AI 1.2.0 : https://www.geekbench.com/ai/ Geekbench AI requires an active internet connection and automatically uploads benchmark results to the Geekbench Browser. Framework | Backend | Device 1 TensorFlow Lite | 1 CPU | 0 Intel(R) Core(TM) Ultra 7 258V 3 ONNX | 1 CPU | 0 Intel(R) Core(TM) Ultra 7 258V 4 OpenVINO | 1 CPU | 0 Intel(R) Core(TM) Ultra 7 258V |
So while we could run Geekbench AI on the CPU, GPU, and NPU on Windows 11, the program only supports CPU processing on Linux, and it does not matter whether I try with Linux 6.13… So I tried to install the Intel AI Boost drivers following the instructions on GitHub:
1 2 3 4 5 6 7 8 |
sudo apt install libtbb12 wget https://github.com/intel/linux-npu-driver/releases/download/v1.13.0/intel-driver-compiler-npu_1.13.0.20250131-13074932693_ubuntu24.04_amd64.deb wget https://github.com/intel/linux-npu-driver/releases/download/v1.13.0/intel-fw-npu_1.13.0.20250131-13074932693_ubuntu24.04_amd64.deb wget https://github.com/intel/linux-npu-driver/releases/download/v1.13.0/intel-level-zero-npu_1.13.0.20250131-13074932693_ubuntu24.04_amd64.deb sudo dpkg -i intel-*.deb wget https://github.com/oneapi-src/level-zero/releases/download/v1.18.5/level-zero_1.18.5+u24.04_amd64.deb sudo dpkg -i level-*.deb sudo reboot |
I did get a few more kernel log messages related to intel_vpu after that:
1 2 3 4 5 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ dmesg | grep intel_vpu [ 2.085553] intel_vpu 0000:00:0b.0: enabling device (0000 -> 0002) [ 2.099330] intel_vpu 0000:00:0b.0: [drm] Firmware: intel/vpu/vpu_40xx_v0.0.bin, version: Jan 15 2025*NPU40xx*ci_tag_ud202504_vpu_rc_20250115_1905*ae83b65d01ccb4696594af0cafffea50a52520da [ 2.099338] intel_vpu 0000:00:0b.0: [drm] Scheduler mode: OS [ 2.131233] [drm] Initialized intel_vpu 1.0.0 for 0000:00:0b.0 on minor 0 |
The NPU device is also there:
1 2 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ ls -l /dev/accel/accel0 crw-rw---- 1 root render 261, 0 Feb 2 10:53 /dev/accel/accel0 |
But I still don’t get an NPU option in Geekbench AI:
1 2 3 4 5 6 7 8 9 10 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~/GeekbenchAI-1.2.0-Linux$ sudo ./banff --ai-list Geekbench AI 1.2.0 : https://www.geekbench.com/ai/ Geekbench AI requires an active internet connection and automatically uploads benchmark results to the Geekbench Browser. Framework | Backend | Device 1 TensorFlow Lite | 1 CPU | 0 Intel(R) Core(TM) Ultra 7 258V 3 ONNX | 1 CPU | 0 Intel(R) Core(TM) Ultra 7 258V 4 OpenVINO | 1 CPU | 0 Intel(R) Core(TM) Ultra 7 258V |
I’ll stop for now, and I may revisit this once I test AI on the Intel Core Ultra 7 258V in more detail.
We’ve already tested the GPU performance with Unigine Heaven Benchmark 4.0 and the Intel Core Ultra 7 258V mini PC last time around when we disabled VT-d and tried Linux 6.11 and Linux 6.13 kernel. Results varied between 42 FPS and 48.2 FPS with the best result achieved with VT-d/IOMMU disabled and Linux 6.13-RC1, but note that disabling VT-d/IOMMU has some security implications. I’ll use the score for Linux 6.11 and VT-d enabled since those are the default values on Ubuntu 24.10: 1,057 points / 42 FPS on average at the standard 1920×1080 resolution. For reference, the score was 2045 points / 81.2 FPS on Windows 11 with Direct3D11.
The test above was made with the default power limits. So I repeated the benchmark with PSYS PL1 = 30 W, PL2 = 37W, Linux 6.11, and IOMMU/VT-d enabled.
We’re getting closer to the optimal performance with 67.4 FPS and a score of 1698 points.
I also tested several 4K or 8K YouTube videos in Firefox and the results were rather disappointing (with default power limits). Note that I had enabled WiFi 6 at that point (See below how I did that) since the 100 Mbps Ethernet USB dongle might have been a bottleneck otherwise.

I was a little too overconfident since 8K 60 FPS YouTube videos would play perfectly in Windows 11, and I started directly with an 8K 60 FPS video. But I had to quickly stop because the video was unwatchable and “Stats for Nerds” reported 1062 frames dropped out of 1589.

4K 60 FPS plays better, but it’s not perfect with 723 frames dropped out of 27640 when playing the video for a little over 6 minutes.

No problem for 4K 30 FPS.

But when I switched to 8K 30 FPS it was also unwatchable. It looks like the system can’t handle 8K AV1 video decoding with the GPU.
I installed Google Chrome and tried again the 8K 60 FPS video, but the video was still unwatchable and I could see the loading logo show up often despite the “buffer health” being just fine.

What happens when I change the PSYS power limits to 30W and 37W?
Still not watchable with some stuttering from time to time, but it’s a big improvement with much fewer frames dropped, and I can imagine that with further optimizations in the future the video may play as smoothly as in Windows 11…
I finally ran Speedometer 2.0 in Firefox to test web browser performance. That’s mostly a single-core benchmark.
That would be 295 runs per minute, a pretty good/expected score for a high-end mini PC.
Since Speedometer 2.0 is deprecated, I’ve also run Speedometer 3.0 to get data for future reviews.
The Khadas Mind 2 AI Maker Kit achieved 18.4 points in Speedometer 3.0 in Firefox.
Khadas Mind 2 AI Maker Kit Ubuntu benchmarks comparison against other mini PCs
Let’s compare the Ubuntu 24.10 benchmark results of the Khadas Mind 2 AI Maker Kit against the Raptor Lake-powered Khadas Mind Premium mini PC (Intel Core i7-1360p) and other high-end mini PCs namely the GEEKOM GT1 Mega (Intel Core Ultra 9 185H), GEEKOM A8 (AMD Ryzen 9 8945HS), and GEEKOM Mini IT13 (Intel Core i9-13900H) which were tested with Ubuntu 22.04, 24.04, or 24.10.
Let’s list the basic specifications of the five systems under test first.
Khadas Mind Maker Kit | Khadas Mind Premium | GEEKOM GT1 Mega | GEEKOM A8 | GEEKOM IT13 | |
---|---|---|---|---|---|
SoC | Intel Core Ultra 7 285V | Intel Core i7-1360P | Intel Core Ultra 9 185H | AMD Ryzen 9 8945HS | Intel Core i9-13900H |
CPU | 8-core (4P+4E) Lunar Lake processor up to 4.8 GHz (P-Cores) and 3.7 GHz (E-Cores) | 12-core (4P+8E) Raptor Lake processor up to 5.0 GHz (P-Cores), up to 3.7 GHz (E-Cores) | 16-core/22-thread (6P+8E+2LP) Meteor Lake processor up to 5.1 GHz (P-cores), up to 3.8 GHz (E-cores), up to 2.5 GHz (LP-cores) | 8-core/16-thread processor up to 5.2 GHz | 14-core/20-thread (6P+8E) Raptor Lake processor up to 5.40 GHz (P-cores), up to 4.10 GHz (E-Cores) |
GPU | 8 Xe cores Intel Arc 140V GPU | 96 EU Intel Iris Xe Graphics up to 1.5 GHz | 8 Xe cores Intel Arc Graphics | AMD Radeon 780M Graphics | 96 EU Intel Iris Xe Graphics |
NPU | Intel AI Boost (47 TOPS) | N/A | Intel AI Boost (34 TOPS) | Ryzen AI (16 TOPS) | N/A |
Memory | 32GB LPDDR5-8533 | 32GB LPDDR5-5200 | 32GB DDR5-5600 | 32GB DDR5-5600 | 32GB DDR4-3200 |
Storage | 1TB NVMe SSD | 1TB NVMe SSD | 2TB NVMe SSD | 2TB NVMe SSD | 2TB NVMe SSD |
Default OS | Windows 11 Home | Windows 11 Home | Windows 11 Pro | Windows 11 Pro | Windows 11 Pro |
Here are the benchmark results.
Khadas Mind 2 AI Maker Kit | Khadas Mind Premium | GEEKOM GT1 Mega | GEEKOM A8 | GEEKOM Mini IT13 | |
---|---|---|---|---|---|
sbc-bench.sh | |||||
- memcpy | 25,504.6 MB/s (P-Core) | 25,389.5 MB/s (P-core) | 21,364.6 MB/s (P-core) | 20,318.5 MB/s | 24,014.4 MB/s (P-core) |
- memset | 65157.0 MB/s (P-Core) | 24,731.8MB/s (P-core) | 36,928.3 MB/s (P-Core) | 62,156.7 MB/s | 26,647.9 MB/s (P-Core) |
- 7-zip (average) | 19,980 31,480 (adjusted PL) | 44,430 | 67,960 | 68,790 | 56,540 |
- 7-zip (top result) | 22,093 32,768 (adjusted PL) | 50,396 | 71,623 | 69,297 | 60,981 |
- OpenSSL AES-256 16K | 1,665,362.60k (P-Core) | 1,771,334.31k (P-Core) | 1,698,239.83k (P-Core) | 1,422,136.66k | 1,844,401.49k (P-Core) |
Geekbench 6 Single | 2,829 | 2093 | 2,605 | 2,661 | 2,745 |
Geekbench 6 Multi | 7,014 9,618 (adjusted PL) | 8,891 | 13,728 | 13,275 | 11,974 |
Unigine Heaven score | 1,057 1,698 (adjusted PL) | 1,349 | 1,956 | 1,972 | 1,333 |
Speedometer 2.0 (Firefox) | 295 | 242 | 278 | 298 | 273 |
The good news is that the Khadas Mind 2 AI Maker Kit has the highest memory bandwidth of all mini PC we’ve tested, and its single-core performance is excellent as demonstrated with Geekbench 6 single-core and Speedometer 2.0 benchmarks. But GPU performance is much lower than in Windows 11 and compared to the other high-end mini PCs we’ve tested in Linux. Multi-core performance is also quite lower both because it only has eight cores, but also due to thermal throttling with default power limits. It was only when I changed the power limits to the same as in Windows 11 (PL1 = 30W, PL2 = 37W) that the scores (shown as adjusted PL in the table above) were closer to expectations. Khadas engineers will have some work to do once they are back from Chinese New Year’s holidays…
Storage and USB testing
Let’s check the SSD performance with iozone3:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ sudo iozone -e -I -a -s 1000M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2 Iozone: Performance Test of File I/O Version $Revision: 3.506 $ Compiled for 64 bit mode. Build: linux-AMD64 random random bkwd record stride kB reclen write rewrite read reread read write read rewrite read fwrite frewrite fread freread 1024000 4 207104 231727 199622 208043 84230 275961 1024000 16 876328 965343 767053 855757 259547 925521 1024000 512 4467967 4433268 4188088 4039209 2313172 4013531 1024000 1024 4810402 4710557 4501473 4346830 1892598 4484580 1024000 16384 4797723 4832139 4615778 4634793 4405063 4770973 iozone test complete. |
Sequential read and write speeds are about 4615 MB/s and 4797 MB/s respectively, both excellent as expected. For reference, the computer achieved 5267 MB/s sequential reads and 4939.70 MB/s sequential writes with CrystralDiskMark on Windows 11.
I also tested the USB ports to find out if we could get the advertised speed using lsusb (USB 3.2) or boltctl (USB4/Thunderbolt) utility as well as iozone3 to measure the speed using an ORICO M234C3-U4 M.2 NVMe SSD enclosure.
First USB Type-A (#1) port from the left:
1 2 3 4 5 6 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ lsusb -t | grep uas |__ Port 001: Dev 004, If 0, Class=Mass Storage, Driver=uas, 10000M jaufranc@Mind-2-AI-Maker-Kit-CNX:/media/sda3$ sudo iozone -e -I -a -s 1000M -r 16384k -i 0 -i 1 random random bkwd record stride kB reclen write rewrite read reread read write read rewrite read fwrite frewrite fread freread 1024000 16384 984718 991621 993384 996661 |
Left USB Type-C port:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ boltctl ● Intel USB4.0 SSD ├─ type: peripheral ├─ name: USB4.0 SSD ├─ vendor: Intel ├─ uuid: ba010000-0052-541e-03d5-47dc2cd4b008 ├─ generation: Thunderbolt 3 ├─ status: authorized │ ├─ domain: 9c2d8780-70df-3214-ffff-ffffffffffff │ ├─ rx speed: 40 Gb/s = 2 lanes * 20 Gb/s │ ├─ tx speed: 40 Gb/s = 2 lanes * 20 Gb/s │ └─ authflags: none ├─ authorized: Sun 02 Feb 2025 08:04:38 UTC ├─ connected: Sun 02 Feb 2025 08:04:38 UTC └─ stored: Sun 02 Feb 2025 08:04:38 UTC ├─ policy: iommu └─ key: no jaufranc@Mind-2-AI-Maker-Kit-CNX:/media/nvme1n1p3$ sudo iozone -e -I -a -s 1000M -r 16384k -i 0 -i 1 random random bkwd record stride kB reclen write rewrite read reread read write read rewrite read fwrite frewrite fread freread 1024000 16384 2171982 2197125 2942240 2955395 |
Here’s a summary for all four ports from left to right:
- USB-C #1 – USB4 – 40 Gbps – Read speed: 2942 MB/s; write speed: 2171 MB/s
- USB-C #2 – USB4 – 40 Gbps – Read speed: 2968 MB/s; write speed: 2186 MB/s
- USB-A #1 – USB 3.2 – 10 Gbps – Read speed: 993 MB/s; write speed: 984 MB/s
- USB-A #2 – USB 3.2 – 10 Gbps – Read speed: 994 MB/s; write speed: 992 MB/s
All good. No problem here.
WiFi 6 performance
Since the Khadas Mind 2 AI Maker Kit does not come with Ethernet, we’ve only tested WiFi 6 performance using Xiaomi Mi AX6000 router and UP Xtreme i11 Edge mini PC on the other end connected through 2.5GbE. There’s a small problem though the AX211 WiFi is detected as a BE201 module and unsurpringly loading the firmware does not work meaning we don’t have WiFi:
1 2 3 4 5 6 7 8 9 10 11 12 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ dmesg | grep iw [ 2.210331] iwlwifi 0000:00:14.3: enabling device (0000 -> 0002) [ 2.218497] iwlwifi 0000:00:14.3: Detected crf-id 0x400410, cnv-id 0x1080900 wfpm id 0x80005b20 [ 2.218513] iwlwifi 0000:00:14.3: PCI dev a840/0094, rev=0x461, rfid=0x2010d000 [ 2.218516] iwlwifi 0000:00:14.3: Detected Intel(R) Wi-Fi 7 BE201 320MHz [ 2.220714] iwlwifi 0000:00:14.3: Direct firmware load for iwlwifi-bz-b0-gf-a0-92.ucode failed with error -2 [ 2.220945] iwlwifi 0000:00:14.3: Direct firmware load for iwlwifi-bz-b0-gf-a0-91.ucode failed with error -2 [ 2.220981] iwlwifi 0000:00:14.3: Direct firmware load for iwlwifi-bz-b0-gf-a0-90.ucode failed with error -2 [ 2.220984] iwlwifi 0000:00:14.3: no suitable firmware found! [ 2.220988] iwlwifi 0000:00:14.3: minimum version required: iwlwifi-bz-b0-gf-a0-90 [ 2.220990] iwlwifi 0000:00:14.3: maximum version supported: iwlwifi-bz-b0-gf-a0-92 [ 2.220992] iwlwifi 0000:00:14.3: check git://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git |
Khadas provides a solution in the link we included in the introduction. In Linux 6.13-RC1, I only had to install the new firmware:
1 2 3 4 5 |
wget https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/plain/iwlwifi-bz-b0-gf-a0-94.ucode -P /tmp wget https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/plain/iwlwifi-bz-b0-gf-a0.pnvm -P /tmp/ sudo cp /tmp/iwlwifi-bz-b0-gf-a0* /lib/firmware/ sync sudo reboot |
But with Linux 6.11, I had to install the new firmware as above and compile the backported wifi driver:
1 2 3 4 5 6 7 8 9 |
sudo apt update sudo apt install -y git build-essential gcc-14 net-tools git clone git://git.kernel.org/pub/scm/linux/kernel/git/iwlwifi/backport-iwlwifi.git -b release/core91 --depth=1 cd backport-iwlwifi make defconfig-iwlwifi-public make -j8 sudo make install sync sudo reboot |
The firmware is now loaded and iwlwifi driver is listed in inxi:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ dmesg | grep iw [ 2.180861] iwlwifi 0000:00:14.3: enabling device (0000 -> 0002) [ 2.194909] iwlwifi 0000:00:14.3: Detected crf-id 0x400410, cnv-id 0x1080900 wfpm id 0x80005b20 [ 2.194925] iwlwifi 0000:00:14.3: PCI dev a840/0094, rev=0x461, rfid=0x2010d000 [ 2.194928] iwlwifi 0000:00:14.3: Detected Intel(R) Wi-Fi 7 BE201 320MHz [ 2.204305] iwlwifi 0000:00:14.3: TLV_FW_FSEQ_VERSION: FSEQ Version: 0.0.4.191 [ 2.204863] iwlwifi 0000:00:14.3: loaded firmware version 94.62990553.0 bz-b0-gf-a0-94.ucode op_mode iwlmvm [ 2.548069] iwlwifi 0000:00:14.3: loaded PNVM version da3b5cf2 [ 2.548226] iwlwifi 0000:00:14.3: RFIm is deactivated, reason = 4 [ 2.566299] iwlwifi 0000:00:14.3: Detected RF GF, rfid=0x2010d000 [ 2.628510] iwlwifi 0000:00:14.3: base HW address: bc:38:98:04:62:08 [ 2.696617] iwlwifi 0000:00:14.3 wlo1: renamed from wlan0 [ 4.665018] iwlwifi 0000:00:14.3: RFIm is deactivated, reason = 4 [ 4.747681] iwlwifi 0000:00:14.3: Registered PHC clock: iwlwifi-PTP, with index: 0 jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ inxi -n Network: Device-1: Intel driver: iwlwifi IF: wlo1 state: up mac: bc:38:98:04:62:08 Device-2: ASIX AX88772B driver: asix type: USB IF: enx00606e435f8a state: up speed: 100 Mbps duplex: full mac: 00:60:6e:43:5f:8a |
I could connect to 5 GHz WiFi 6…
…before running iperf3:
- Download
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ iperf3 -t 60 -c 192.168.31.12 -i 10 -R Connecting to host 192.168.31.12, port 5201 Reverse mode, remote host 192.168.31.12 is sending [ 5] local 192.168.31.177 port 34424 connected to 192.168.31.12 port 5201 [ ID] Interval Transfer Bitrate [ 5] 0.00-10.01 sec 1.59 GBytes 1.36 Gbits/sec [ 5] 10.01-20.01 sec 1.54 GBytes 1.33 Gbits/sec [ 5] 20.01-30.01 sec 1.59 GBytes 1.37 Gbits/sec [ 5] 30.01-40.01 sec 1.55 GBytes 1.33 Gbits/sec [ 5] 40.01-50.01 sec 1.59 GBytes 1.37 Gbits/sec [ 5] 50.01-60.01 sec 1.54 GBytes 1.32 Gbits/sec - - - - - - - - - - - - - - - - - - - - - - - - - [ ID] Interval Transfer Bitrate Retr [ 5] 0.00-60.04 sec 9.40 GBytes 1.35 Gbits/sec 0 sender [ 5] 0.00-60.01 sec 9.40 GBytes 1.35 Gbits/sec receiver iperf Done. |
- Upload:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 |
jaufranc@Mind-2-AI-Maker-Kit-CNX:~$ iperf3 -t 60 -c 192.168.31.12 -i 10 Connecting to host 192.168.31.12, port 5201 [ 5] local 192.168.31.177 port 55830 connected to 192.168.31.12 port 5201 [ ID] Interval Transfer Bitrate Retr Cwnd [ 5] 0.00-10.01 sec 1.73 GBytes 1.48 Gbits/sec 0 3.07 MBytes [ 5] 10.01-20.01 sec 1.89 GBytes 1.62 Gbits/sec 0 3.07 MBytes [ 5] 20.01-30.01 sec 1.80 GBytes 1.55 Gbits/sec 963 2.92 MBytes [ 5] 30.01-40.01 sec 1.88 GBytes 1.61 Gbits/sec 0 3.00 MBytes [ 5] 40.01-50.01 sec 1.77 GBytes 1.52 Gbits/sec 0 3.10 MBytes [ 5] 50.01-60.01 sec 1.87 GBytes 1.61 Gbits/sec 52 3.03 MBytes - - - - - - - - - - - - - - - - - - - - - - - - - [ ID] Interval Transfer Bitrate Retr [ 5] 0.00-60.01 sec 10.9 GBytes 1.57 Gbits/sec 1015 sender [ 5] 0.00-60.05 sec 10.9 GBytes 1.57 Gbits/sec receiver iperf Done. |
1.35 Gbps download speed and 1.57 Gbps upload speed are pretty good, albeit not quite as fast as in Windows (1.78 Gbps / 1.93 Gbps) in the same test environment.
Khadas Mind 2 AI Maker Kit stress test in Ubuntu 24.10
We’ve seen the CPU temperature never exceeds 60°C with the default power limits in Linux, so it’s pretty clear thermal throttling would not happen under those conditions. So I run a stress test on all eight cores of the Intel Core Ultra 7258V processor after setting PL1 to 30W and PL2 to 37W in PSYS Zone (1). I monitored CPU temperature and frequency with Psensor and the sbc-bench.sh script.
The temperature maxes out at 70°C and the frequency stabilizes at 3300 MHz on the P-cores and 3701 MHz on the E-cores. There’s still some legroom at least for CPU-only workloads even after adjusting the power limits.
Fan noise
Khadas Mind AI Maker Kit is cooled with a fan that’s inaudible at idle and light workloads but becomes noisier under stress, although I don’t find it annoying. I measured the fan noise with a sound level meter placed at around 5 centimeters from the top of the device:
- Idle – 37.9 – 38.3 dBA (low-speed fan)
- YouTube 8K 60FPS video in Firefox (volume off, adjusted power limits) – 46.9 – 47.8 dBA
- Stress test on all 8 cores (adjusted power limited) – 48.8 – 49.4 dBA
For reference, the meter measures around 37.2 dBA in a quiet room.
Mind 2 AI Maker Kit power consumption in Ubuntu 24.10
After making sure the internal battery is fully charged, I measured the power consumption with a wall power meter:
- Power off – 1.6 Watt
- Idle – 3.9 – 4.2 Watts
- Video playback – 32.6 – 35.9 Watts (Youtube 8K 60FPS in Firefox – AV1 codec)
- CPU stress test (stress -c 8)
- First ~30 seconds – 40.8 – 40.9 Watts
- Longer run – 33.5 – 33.6 Watts
During the measurements, the mini PC was connected to WiFi 6, an HDMI display with its own power adapter, and a USB RF dongle for a wireless keyboard and mouse combo.
Conclusion
If the Khadas Mind 2 AI Maker Kit was a mini PC for end users, I would have no hesitation telling people to stay away if they wanted to run Ubuntu 24.10 or other Linux distributions on the mini PC. That’s because it’s a complete mess and workarounds are needed for it to work properly. But considering it’s made for developers, once you install the WiFi driver and firmware, Intel AI Boost NPU drivers, and adjust the power limits, it works relatively well albeit I was unable to reach Windows 11’s performance levels in tasks like 8K 60 FPS YouTube video playback and 3D graphics, meaning more work is needed. Geekbench AI is unable to use the GPU or NPU on Linux, while it can on Windows 11.
On a positive note, the processor offers the fastest single-core performance of the many mini PCs we’ve tested, the WD SSD is super fast, and so is the AX211 WiFI 6 module once you’ve installed the drivers. All features work as expected after tweaks, except the battery which I have trouble understanding. Sometimes I have no problems waking up the mini PC, and other times I have to press the power button to shut it off before starting it again. Its eight cores may be limiting depending on what you want to do, but its up to 115 TOPS of AI performance makes it a good machine for AI development, although I have not tested the OpenVino SDK just yet.
I’d like to thank Khadas for sending the Mind 2 AI Maker Kit for review, also referred to as the Mind Maker Kit. The company sells the AI developer kit on its website for $1,199 plus shipping. I plan to test AI features on Windows and Ubuntu in the final part of the review once I have some answer from Khadas since some Copilot+ features are not enabled for some reason…

Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
[ was interested what are specification details for the battery: “Mind Maker Kit is equipped with a built-in 5.55Wh battery to ensure an uninterrupted workflow. When the power is disconnected, the device instantly enters sleep mode, preserving your progress without any interruptions. With up to 25 hours of standby time and a 2-second wake-up feature, you can seamlessly resume your desktop projects exactly where you left off. This ensures smooth transitions between tasks and locations, enhancing your productivity and development efficiency.” and astonished, that the expansion interface (“Mind Link Expansion Interface”) is a PCIe 5.0 x8 (256GT/s) “Additionally, the open-source… Read more »
> Idle – 3.9 – 4.2 Watts
That’s *outstandingly low* power consumption for an Intel CPU. Lower than any I’ve seen recently!
Has Intel finally got its game together in this regard with Meteor Lake?
Thanks Jean-Franc foi one more in-depth review. 👍
> Has Intel finally got its game together in this regard with Meteor Lake?
Huh? If a system is carefully designed and software/settings fit then you find many Intel powered systems that idle at or below 4W. And that does not only apply to ‘slow’ machines based on Jasper or Alder Lake-N but you’ll find also reports for e.g. i3-13100 idling that low.
BTW: when measuring at wall replacing a crappy power brick with something better can easily half idle consumption. Check for example Jeff testing a Twin Lake box idling north of 10W with vendor supplied power brick.
Found it: 2.5W idle (vendor provided power brick and all losses included) after adjusting settings to fit our needs: https://www.cnx-software.com/2021/10/15/2-5gbe-mini-pc-asus-pn41/#comment-585825
The keyword here is “fit our needs”. If I remove the HDMI cable and USB dongle, the idle power consumption drops to 2.5-2.7 Watts. That would be with WiFi 6 and only the USB-C power cable connected to the mini PC.
I’ve also noticed the power consumption of workloads using the GPU has dropped on newer Intel processors with Intel Arc.