Nick A Posted February 16 Author Posted February 16 @Jain Ziad I'm currently working on hardware acceleration, though I can't say yet if it will outperform the official Radxa builds since it uses the same drivers. I'll keep the community updated on the results. 0 Quote
cest73 Posted February 21 Posted February 21 Hello NickA, fire-hound from discord here ... You do a tremendous service for us all, keep it up buddy! I loaded your recent build from here: https://github.com/NickAlilovic/build/releases and can confirm it works on my A7A. It correctly detected my Pimorony NVME Base lsblk NAME MAJ:MIN RM SIZE RO TYPE MOUNTPOINTS mtdblock0 31:0 0 8M 0 disk mmcblk0 179:0 0 29.2G 0 disk ├─mmcblk0p1 179:1 0 256M 0 part /boot └─mmcblk0p2 179:2 0 28.6G 0 part /var/lib/containers/storage/overlay /var/log.hdd / zram0 254:0 0 3.9G 0 disk [SWAP] zram1 254:1 0 50M 0 disk /var/log zram2 254:2 0 0B 0 disk nvme0n1 259:0 0 238.5G 0 disk └─nvme0n1p1 259:1 0 238.5G 0 part /srv/dev-disk-by-uuid-b28b2f5c-ee2e-4ded-959e-22bc63051b1d and: screenfetch _,met$$$$$gg. root@radxa-cubie-a7a ,g$$$$$$$$$$$$$$$P. OS: Debian 13 trixie ,g$$P"" """Y$$.". Kernel: aarch64 Linux 6.6.98-vendor-sun60iw2 ,$$P' `$$$. Uptime: 11h 14m ',$$P ,ggs. `$$b: Packages: 661 `d$$' ,$P"' . $$$ Shell: bash 5.2.37 $$P d$' , $$P Disk: 4.9G / 271G (2%) $$: $$. - ,d$$' CPU: ARM Cortex-A55 Cortex-A76 @ 8x 1.794GHz $$\; Y$b._ _,d$P' RAM: 534MiB / 7928MiB Y$$. `.`"Y$$$$P"' `$$b "-.__ `Y$$ `Y$$. `$$b. `Y$$b. `"Y$b._ `"""" So far so good The issues: 1. The typical Armbian image has a module in the armbian-config and radxa's rsetup where one can set up the two leds conveiently to other than full on and heartbeat for the green and blue gpio leds respectively. I find it to be a nuisance to dig for that manually 2. The Ovralys module in armbian-config is returning an error - when entering DTO001 module (Device Tree Overlays) saying: Invalid overlay_prefix that might require deeper digging into the issue i guess? What works: Open Media Vault registers the NVME nicely as additional storage. I suspect booting off that NVME would work for an updated SPI bootloader i could test that if you want me to? If you need any other info or output I stand at Your disposal 😁 0 Quote
alexc Posted March 23 Posted March 23 Hi all, Over the past few months working on the A7Z, I’ve successfully ported a number of BSP drivers to Linux 6.18. The code is available here: https://github.com/alexcaoys/allwinner-bsp/tree/linux-6.18.y It’s currently tested on 6.18.19 and should, in theory, work with newer 6.18.y releases. I’ve tried to reuse as much upstream (mainline) drivers as possible, and have also upstreamed support for PCK600 and ET7304 used on the A7Z. Hopefully this helps accelerate ongoing mainline efforts. Feedback, issues, and PRs are very welcome. I may also look into creating branches for Linux 7.x in the future. 3 Quote
Nick A Posted April 2 Author Posted April 2 @alexc Thanks for all your hard work! I’ve put together an Armbian build using your kernel—you can check it out here: https://github.com/NickAlilovic/build/tree/Radxa-mainline-WIP 0 Quote
qq20739111 Posted April 11 Posted April 11 @Nick A Is it possible to release a system for Radxa Cubie A7S? 0 Quote
Nick A Posted April 11 Author Posted April 11 (edited) @qq20739111 I'll add it soon. 6.18 might take awhile. Edited April 11 by Nick A 2 Quote
tiobily Posted Wednesday at 01:01 PM Posted Wednesday at 01:01 PM hello i have ths a7z with 8gb ram. Actualy im using with only 4gb (ok for my application). I can boot only if i set mem=4g. Any workaround for that? i tried all images ive found and no one can boot with 8gb. 0 Quote
chris178 Posted Wednesday at 10:37 PM Posted Wednesday at 10:37 PM (edited) @tiobily Is your a7z factory 8GB or modded out of curiosity (lpddr4 or lpddr5) ? Are you able to connect serial console onto uart0 (pb9, pb10) header pins (3.3V TTL uart adapter) and see right after power is applied to the board the debug of dram detection and training? Debug will show if full 8GB is being detected - once this is confirmed, we can take it from there. Edited Wednesday at 10:40 PM by chris178 0 Quote
tiobily Posted Thursday at 08:11 PM Posted Thursday at 08:11 PM hi @chris178! 1 - factory 8gb, not modded, lpddr4 2 - boot log: ▒...... [151]HELLO! BOOT0 is starting! [154]BOOT0 commit : {9f6fab4f-dirty} [168]dram_para_total:0xf [171]vaild para:6 select dram para10 [175][mmc]: mmc driver ver 2024-03-20 10:36 [184][mmc]: Wrong media type 0x0 [188][mmc]: ***Try SD card 0*** [197][mmc]: HSSDR52/SDR25 4 bit [200][mmc]: 50000000 Hz [202][mmc]: 59638 MB [205][mmc]: ***SD/MMC 0 init OK!!!*** [210]DRAM BOOT DRIVE INFO: V0.601 [215]DRAM_VCC set to 1100 mv [217]DRAM CLK =1800 MHZ [220]DRAM Type =8 (8:LPDDR4,9:LPDDR5) [320]Training result is = 7 [323]DRAM Pstate 1 training, frequency is 1200 Mhz [433]Training result is = 7 [436]DRAM Pstate 2 training, frequency is 800 Mhz [642]Training result is = 7 [645]DRAM Pstate 3 training, frequency is 400 Mhz [3829]Training result is = 7 [3832]DRAM Pstate 0 training, frequency is 1800 Mhz [3840]Actual DRAM SIZE =8192 M [3843]DRAM SIZE =8192 MBytes, para1 = 311a, para2 = 20001001, dram_tpr13 = 6067 [3858]DRAM simple test OK. [3864]error:bad magic. [3928]error: dtb not found for scp [3931]mmc not para [3933]Jump to ATF: monitor_base = 0x48000000, uboot_base = 0x4a000000, optee_base = 0x0 NOTICE: BL31: OP-TEE 32bit detected NOTICE: BL31: U-BOOT 32bit detected NOTICE: BL31: v2.5(debug):5fc237a6a NOTICE: BL31: Built : 09:05:28, Feb 26 2025 NOTICE: BL31: No DTB found. ERROR: Error initializing runtime service opteed_fast U-Boot 2018.07-12-boot-aw2501-gb2d229198b2-dirty (Jan 06 2026 - 03:45:33 +0000) Allwinner Technology [03.985]CPU: Allwinner Family [03.988]Model: Radxa Cubie A7A I2C: ready [04.052]DRAM: 8 GiB [04.061]Relocation Offset is: 75d89000 [04.114]secure enable bit: 0 [04.117]PMU: AXP8191 [04.119]PMU: AXP8191 VER_A FDT ERROR:fdt_get_regulator_name:get property handle twi-supply error:FDT_ERR_INTERNAL [04.131]bmu_axp515_probe pmic_bus_read fail [04.135][AXP8191] onoff status: 0x50 = 0x1, 0x51 = 0x0 [04.140][AXP8191] charge status: 0x4 = 0x0 [04.144]CPU=1014 MHz,PLL6=1200 Mhz,AHB=200 Mhz, APB1=24Mhz MBus=600Mhz [04.150]gic: sec monitor mode SPI ALL: ready [04.194]flash init start [04.196]workmode = 0,storage type = 1 [04.200][mmc]: mmc driver ver uboot2018:2024-10-18 9:01:00 -- uboot2018:2025-03-28 10:03:00 [04.208][mmc]: get sdc_type fail and use default host:tm1. [04.216][mmc]: can't find node "mmc0",will add new node [04.221][mmc]: fdt err returned <no error> [04.224][mmc]: Using default timing para [04.228][mmc]: SUNXI SDMMC Controller Version:0x50530 [04.246][mmc]: card_caps:0x3000000a [04.249][mmc]: host_caps:0x3000003f [04.252]sunxi flash init ok get value error [04.262]Loading Environment from SUNXI_FLASH... OK No ethernet found. [04.299]NOT support bridge yet [04.302]Warn: no find panel or bridge [04.403]drm hdmi get mode: 1920x1080@60Hz hdmi0@5520000: use force output detailed mode clock 139190 kHz, flags[5] H: 1920 2008 2052 2080 V: 1080 1082 1087 1115 bus_format: 0 clk_set_rate: <NULL> has NULL parent [04.505]suxni tcon hdmi set rate: 139190000Hz and get diff: -139190000Hz [04.577]the secure storage item0 copy0 magic is bad [04.589]the secure storage item0 copy1 magic is bad [04.593]usb burn from boot delay time 0 [04.599]usb prepare ok [04.902]overtime [04.906]do_burn_from_boot usb : no usb exist skip update boot_param cann't get the boot_base from the env [04.917]update part info [04.922]battery temp is -1 [04.926]update bootcmd [04.934](not enough data: expected >= 8 cells, got 4 cells) Link up timeout Speed change timeout PCIe speed of Gen1 Override default fdtfile to allwinner/sun60i-a733-cubie-a7z.dtb [07.376]libfdt fdt_check_header(): FDT_ERR_BADMAGIC [07.382]libfdt fdt_path_offset() returned FDT_ERR_BADMAGIC [07.388][mmc]: get sunxi-mmc0 string failed [07.392]The storage not support sample function [07.398]## error: update_fdt_dram_para_from_bootpara : FDT_ERR_NOTFOUND ret 0 [07.408]chip id : 0x00005100 0x00004900 0x00004810 0x00001d62 [07.414]mac addr => 08:51:49:58:7f:71 [07.417]chip id : 0x00005100 0x00004900 0x00004810 0x00001d62 [07.423]mac addr => 08:51:49:58:7f:70 [07.427]/chosen/rng-seed node not exit [07.430]no battery, disabled battery functons [07.434]Could not find nodeoffset for bat_supply [07.439]update dts curr_device:0 Device: SUNXI SD/MMC Manufacturer ID: ff OEM: 0 Name: SDU1 Bus Speed: 50000000 Mode : SD High Speed (50MHz) Rd Block Len: 512 SD version 3.0 High Capacity: Yes Capacity: 58.2 GiB Bus Width: 4-bit Erase Group Size: 512 Bytes [ufs]info:Driver version 0.0.24 2024.12.30 16:00 [ufs]info:Cal words 0x60:val 0xb0880000, 0x64:val 0x4f21a959 [ufs]err:uic c 16,uic a 0,1,0 [ufs]err:e 80000010,0,0,0,0 [ufs]err:int s 0,400 [ufs]err:Device not present [ufs]err:wait device link startup failed [ufs]info:peer link startup timeout [ufs]info:Cal words 0x60:val 0xb0880000, 0x64:val 0x4f21a959 [ufs]err:uic c 16,uic a 0,1,0 [ufs]err:e 80000010,0,0,0,0 [ufs]err:int s 0,400 [ufs]err:Device not present [ufs]err:wait device link startup failed [ufs]info:peer link startup timeout [ufs]info:Cal words 0x60:val 0xb0880000, 0x64:val 0x4f21a959 [ufs]err:uic c 16,uic a 0,1,0 [ufs]err:e 80000010,0,0,0,0 [ufs]err:int s 0,400 [ufs]err:Device not present [ufs]err:wait device link startup failed [ufs]info:peer link startup timeout [ufs]info:Cal words 0x60:val 0xb0880000, 0x64:val 0x4f21a959 [ufs]err:uic c 16,uic a 0,1,0 [ufs]err:e 80000010,0,0,0,0 [ufs]err:int s 0,400 [ufs]err:Device not present [ufs]err:wait device link startup failed [ufs]info:peer link startup timeout [ufs]err:link startup failed -6 [ufs]err:ufshcd_probe() failed -6 fail to init ufs sunxi_flash - sunxi_flash sub-system Usage: sunxi_flash sunxi_flash init storage_type sunxi_flash read mem_addr part_name [size] sunxi_flash read_mtd mem_addr part_name [size] sunxi_flash write <mem_addr> <part_name> [size] sunxi_flash write <mem_addr> <part_name> [offset] [size] sunxi_flash write_mtd <mem_addr> <part_name> sunxi_flash boot0 force_dram_update_size <new_val> sunxi_flash boot0 force_dram_update_flag <new_val> Hit any key to stop autoboot: 0 starting USB... USB0: start sunxi USB-DRD... config usb clk ok sunxi USB-DRD init ok... USB EHCI 1.00 scanning bus 0 for devices... 1 USB Device(s) found USB1: start sunxi USB1-Host... config usb clk ok sunxi USB1-Host init ok... USB EHCI 1.00 scanning bus 1 for devices... 1 USB Device(s) found scanning usb for storage devices... 0 Storage Device(s) found Device 0: device type unknown ... is now current device ** Bad device usb 0 ** ** Bad device usb 0 ** mmc0 is current device Scanning mmc 0:2... Scanning mmc 0:3... Found /boot/extlinux/extlinux.conf Retrieving file: /boot/extlinux/extlinux.conf 1307 bytes read in 19 ms (66.4 KiB/s) U-Boot menu 1: Debian GNU/Linux 11 (bullseye) 5.15.147-14-a733 2: Debian GNU/Linux 11 (bullseye) 5.15.147-14-a733 (rescue target) Enter choice: 1: Debian GNU/Linux 11 (bullseye) 5.15.147-14-a733 Retrieving file: /boot/initrd.img-5.15.147-14-a733 31153067 bytes read in 1310 ms (22.7 MiB/s) Retrieving file: /boot/vmlinuz-5.15.147-14-a733 19030528 bytes read in 808 ms (22.5 MiB/s) append: root=UUID=6f750720-329a-45f0-a4b5-abc5797b040a console=ttyAS0,115200n8 rootwait clk_ignore_unused mac_addr=08:51:49:58:7f:71 mac1_addr=08:51:49:58:7f:70 quiet splash loglevel=4 rw earlycon consoleblank=0 console=tty1 coherent_pool=2M irqchip.gicv3_pseudo_nmi=0 cgroup_enable=cpuset cgroup_memory=1 cgroup_enable=memory swapaccount=1 kasan=off Retrieving file: /usr/lib/linux-image-5.15.147-14-a733/allwinner/sun60i-a733-cubie-a7z.dtb 193692 bytes read in 145 ms (1.3 MiB/s) [16.025]libfdt fdt_path_offset() returned FDT_ERR_BADMAGIC [16.034]The storage not support sample function ret 0 [16.136]mac addr => 08:51:49:58:7f:71 [16.140]mac addr => 08:51:49:58:7f:70 [16.145]no battery, disabled battery functons [16.152]update dts stop sunxi USB-DRD ok... stop sunxi USB1-Host ok... Loading Ramdisk to 4824a000, end 49fffbab ... OK ERROR: reserving fdt memory region failed (addr=48000000 size=1000000) [16.232]Starting kernel ... [16.235]total: 16235 ms [16.237][mmc]: MMC Device 2 not found [16.240][mmc]: mmc 2 not find, so not exit NOTICE: [SCP] :wait arisc ready.... NOTICE: [SCP] :arisc version: [d463b9da43dc50320f21ba51c6c51afe2db20d83] NOTICE: [SCP] :arisc startup ready NOTICE: [SCP] :arisc startup notify message feedback NOTICE: [SCP] :sunxi-arisc driver is starting NOTICE: BL3-1: Next image address = 0x40200000 NOTICE: BL3-1: Next image spsr = 0x3c5 nothing happen after that my workaround ( with claude help ) edit extlinux.conf and add mem=4G append root=UUID=0e1a597a-1884-4b41-a07c-733efecf176c console=ttyAS0,115200n8 earlyprintk=sunxi-uart,0x2500000 rootwait mem=4G clk_ignore_unused any higher value dosent work. 0 Quote
chris178 Posted Thursday at 09:20 PM Posted Thursday at 09:20 PM (edited) @tiobily Ram detect and training looks just fine above.... you could interrupt the boot and test memory. what OS image did you use above? I could try it on my device (also 8GB but modded). Did you try different microSD cards? Just tried @Nick A image and u-boot section is very similar. However boot continues with linux kernel at the point your debug stalled.... I can't see much clue as to why you're having that issue from the log above unfortunately... Just tried https://github.com/radxa-build/radxa-a733/releases/download/rsdk-r5/radxa-a733_bullseye_cli_r5.output_512.img.xz on a 8GB moded A7Z and it boot fine and free mem as expected... Edited Thursday at 10:13 PM by chris178 0 Quote
tiobily Posted Friday at 01:07 AM Posted Friday at 01:07 AM @chris178 im using this image ( r5 512 ) with the mem=4g param i tried the nick A image too, same problem. ( i tried the r4, r1, 512 and 4096 ) i didnt tried a different microsd, im going to buy a new one and make a new test maybe i got a faulty board. (im using npu and a lot of cpu, everything running fine) Srry about my poor english, is not my native language 😅 0 Quote
qq20739111 Posted Friday at 11:27 AM Posted Friday at 11:27 AM @tiobily You can post your issue on the official community as long as you haven't done any hardware modifications yourself. By the way, How did you get the NPU running on this board? 0 Quote
tiobily Posted Friday at 03:36 PM Posted Friday at 03:36 PM @qq20739111 thks, official community = github? https://github.com/radxa-docs/docs/ i asked claude to make a summary of my project with npu: We use the vendor's ACUITY toolkit inside a Docker container (ubuntu-npu:v2.0.10.1) to convert ONNX → quantized .nb files. - Pipeline: onnxsim → pegasus_import → pegasus_quantize → pegasus_export, uint8 quantization, deployed to ~/npu_models/ via push.sh. - Runtime: custom npu_server.c (pre-allocated input buffer, no per-call mmap) that the Python app talks to. - Hard lesson #1 — quantization: only pure Conv+BN+ReLU survives uint8. Attention, SE blocks, hard-swish, LayerNorm all collapse to constant outputs. So no MobileNetV3+, no transformers. - Hard lesson #2 — concurrency hang: NPU IRQs get lost when camera ISP DMA runs in parallel (shared memory bus). Fix: suppress GStreamer buffer copies during inference. Never STREAMOFF/ON the sunxi-vin driver — instant kernel crash. - Result: 12 models running (9 NPU + 3 CPU) at ~40ms/inference. 1 Quote
Nick A Posted Friday at 03:41 PM Author Posted Friday at 03:41 PM (edited) @tiobily https://radxa.com/community Try these forums. Edited Friday at 04:21 PM by Nick A 0 Quote
chris178 Posted Friday at 03:52 PM Posted Friday at 03:52 PM @tiobily no worries. you can still interrupt u-boot. In u-boot shell you can try calculating crc32 of memory regions to test memory. memtester tool does not seem to be working. 0 Quote
qq20739111 Posted Sunday at 08:44 AM Posted Sunday at 08:44 AM (edited) @Nick A hi, I noticed a big difference in my tests on A7Z. The official image runs cool at 60°C, but your image boosts performance by roughly 30% while hitting 90°C. Did you apply an overclock, or is the stock image just limiting the potential? Edited Sunday at 08:45 AM by qq20739111 0 Quote
Nick A Posted Sunday at 10:51 AM Author Posted Sunday at 10:51 AM (edited) @qq20739111 The only change I can think of is the Thermal Governor is set to STEP_WISE by default. Without it the fan would run full speed all the time. Radxa had it set to CONFIG_THERMAL_DEFAULT_GOV_POWER_ALLOCATOR=y . I guess it would be safer to switch back to POWER_ALLOCATOR and figure another way to fix the fan issue. The 0.6.x images don't have these changes. Is this an image you compiled yourself? The CPU needs proper cooling you should install one of these. https://radxa.com/products/accessories/heatsink-6530b Extreme throttling introduced on A733 https://forum.radxa.com/t/extreme-throttling-introduced-on-a733/30688/4 Edited Sunday at 01:19 PM by Nick A 0 Quote
qq20739111 Posted Sunday at 10:39 PM Posted Sunday at 10:39 PM (edited) @Nick A I didn't compile the image myself; I downloaded yours directly. What I meant is that when I running a CPU benchmark program, with all 8 cores fully loaded, the score using your image is about 30% higher than with the official image. However, the CPU temperature is also higher, even though I'm using the exact same cooling setup. By the way, that official heatsink seems to be out of stock now. Edited Sunday at 10:44 PM by qq20739111 0 Quote
Nick A Posted yesterday at 12:08 AM Author Posted yesterday at 12:08 AM (edited) @qq20739111 I have this patch included in my latest build. I wonder if this will change the cooling and performance. I'll release a test image so you can perform a benchmark on it. I only have a heat sink and fan for the A7A right now. https://github.com/radxa/allwinner-device/commit/7716c3aacfe0ca4432ad3973ec7ea864d2b91195 Edited yesterday at 12:10 AM by Nick A 0 Quote
qq20739111 Posted 9 hours ago Posted 9 hours ago (edited) @Nick A Rising CPU temps aren't really a concern. I'm using a massive 6x6cm aluminum heatsink for passive cooling. Even without a fan, it keeps the CPU stable under full load for long periods. The real bottleneck is the frequency throttling, which is really holding back performance. Edited 8 hours ago by qq20739111 0 Quote
qq20739111 Posted 5 hours ago Posted 5 hours ago @Nick A As shown in the figure, I tested your newly released patched image on this fanless DIY heatsink dev board using a CPU benchmark tool, and the CPU performance is indeed heavily throttled! With v0.6.1, I scored 2571, with average core temps around 75°C. However, with v0.6.4, the all-core score dropped to just 985, with temps hovering around 60°C. This is even significantly lower than the official system image (1700+). I would really prefer a version without these patches; otherwise, the board's potential is being wasted. After all, sustained operation below 90°C is perfectly safe for this CPU. The official release limits CPU performance to prevent high-temp instability and reduce support tickets, but your image doesn't need to impose these restrictions. 0 Quote
Nick A Posted 5 hours ago Author Posted 5 hours ago @qq20739111 thanks I'll remove the patch and compile a new image. 0 Quote
qq20739111 Posted 5 hours ago Posted 5 hours ago (edited) @Nick A Thanks for taking my advice.😄 This is the test program of mine. You can play around with it if you have time. cpu_benchmark.py.zip Edited 5 hours ago by qq20739111 0 Quote
humanus Posted 1 hour ago Posted 1 hour ago (edited) I ran Geekbench on the A7S with all the clocks at maximum. Here are the results: https://browser.geekbench.com/v5/cpu/24253556 https://browser.geekbench.com/v6/cpu/17727869 And your script gives me: ╔══════════════════════════════════════════════════╗ ║ CPU Performance Benchmark Tool v1.0 ║ ║ Single · Multi-Core · Big/LITTLE Benchmark ║ ║ Auto-detects heterogeneous cores ║ ╚══════════════════════════════════════════════════╝ ════════════════════════════════════════════════════════════ System Info ════════════════════════════════════════════════════════════ OS: Linux 6.6.98-vendor-sun60iw2 (aarch64) Processor: aarch64 Logical cores: 8 Core topology: 8 cores [homogeneous / unprobed] Python: 3.13.5 (CPython) Test time: 2026-04-21 00:46:32 Single-core 5s/test | Per-core 3rounds x 3s | 8 dimensions ════════════════════════════════════════════════════════════ Single Core Benchmark (Fastest Core) ════════════════════════════════════════════════════════════ [1/8] Integer (Prime Sieve) ... 4637 iters / 5.00s (927.39 ops/s) [2/8] Float (Matrix Mul) ... 7 iters / 5.12s (1.37 ops/s) [3/8] Float (FFT) ... 3 iters / 7.05s (0.43 ops/s) [4/8] Hash (SHA-256) ... 114 iters / 5.02s (22.70 ops/s) [5/8] Memory Sort ... 4 iters / 6.56s (0.61 ops/s) [6/8] Recursion (Fibonacci) ... 1 iters / 11.63s (0.09 ops/s) [7/8] Math (Pi) ... 1852 iters / 5.00s (370.35 ops/s) [8/8] Compress (zlib) ... 25 iters / 5.14s (4.86 ops/s) ════════════════════════════════════════════════════════════ Per-Core Power Probe (8 logical cores) ════════════════════════════════════════════════════════════ System reports no heterogeneous topology, probing empirically 3 rounds x 3s/round, warmup 0.5s, pinned mode, 8 workers Probing Integer (3 rounds): Round 1/3 ... max 923.9 min 277.1 ratio 3.33x pinned Round 2/3 ... max 922.6 min 275.2 ratio 3.35x pinned Round 3/3 ... max 921.8 min 273.9 ratio 3.37x pinned Median results (3 rounds): Core 6: 922.63 ops/s [████████████████████] 100.0% Core 7: 916.12 ops/s [███████████████████░] 99.3% Core 4: 276.21 ops/s [█████░░░░░░░░░░░░░░░] 29.9% Core 1: 276.10 ops/s [█████░░░░░░░░░░░░░░░] 29.9% Core 3: 276.04 ops/s [█████░░░░░░░░░░░░░░░] 29.9% Core 5: 275.76 ops/s [█████░░░░░░░░░░░░░░░] 29.9% Core 2: 275.56 ops/s [█████░░░░░░░░░░░░░░░] 29.9% Core 0: 275.17 ops/s [█████░░░░░░░░░░░░░░░] 29.8% -> Heterogeneity detected: 2 big cores (avg 919.4 ops/s) + 6 little cores (avg 275.8 ops/s) ratio 3.33x Probing Float (3 rounds): Round 1/3 ... max 370.4 min 95.4 ratio 3.88x pinned Round 2/3 ... max 371.1 min 95.4 ratio 3.89x pinned Round 3/3 ... max 369.5 min 95.3 ratio 3.88x pinned Median results (3 rounds): Core 6: 370.39 ops/s [████████████████████] 100.0% Core 7: 369.28 ops/s [███████████████████░] 99.7% Core 5: 95.59 ops/s [█████░░░░░░░░░░░░░░░] 25.8% Core 4: 95.51 ops/s [█████░░░░░░░░░░░░░░░] 25.8% Core 1: 95.46 ops/s [█████░░░░░░░░░░░░░░░] 25.8% Core 3: 95.46 ops/s [█████░░░░░░░░░░░░░░░] 25.8% Core 2: 95.45 ops/s [█████░░░░░░░░░░░░░░░] 25.8% Core 0: 95.36 ops/s [█████░░░░░░░░░░░░░░░] 25.7% -> Heterogeneity detected: 2 big cores (avg 369.8 ops/s) + 6 little cores (avg 95.5 ops/s) ratio 3.87x Multi-benchmark cross-validation agrees: 2 big + 6 little System didn't report heterogeneous, but probe found 2 big + 6 little (ratio 3.58x) ════════════════════════════════════════════════════════════ Big/Little Independent Benchmark (2 big + 6 little) ════════════════════════════════════════════════════════════ Three phases: big(2 threads) / little(6 threads) / mixed(8 threads) [1/8] Integer (Prime Sieve) Big ( 2 threads): iters 3637 | throughput 1818.20 ops/s | per-core min/avg/max: 907.88/909.10/910.33 Little ( 6 threads): iters 5881 | throughput 2939.47 ops/s | per-core min/avg/max: 275.72/489.91/924.17 Mixed ( 8 threads): iters 6989 | throughput 3492.59 ops/s | per-core min/avg/max: 275.81/436.57/919.04 Speedup: big 1.97x little 3.18x mixed 3.78x Efficiency: big 98.4% little 53.0% mixed 47.3% Big per-core min/avg/max: 907.88/909.10/910.33 ops/s Little per-core min/avg/max: 275.72/489.91/924.17 ops/s Mixed per-core min/avg/max: 275.81/436.57/919.04 ops/s [2/8] Float (Matrix Mul) Big ( 2 threads): iters 4 | throughput 1.14 ops/s | per-core min/avg/max: 0.57/0.57/0.57 Little ( 6 threads): iters 8 | throughput 1.95 ops/s | per-core min/avg/max: 0.23/0.32/0.51 Mixed ( 8 threads): iters 8 | throughput 2.59 ops/s | per-core min/avg/max: 0.24/0.32/0.48 Speedup: big 1.48x little 2.54x mixed 3.37x Efficiency: big 74.2% little 42.3% mixed 42.1% Big per-core min/avg/max: 0.57/0.57/0.57 ops/s Little per-core min/avg/max: 0.23/0.32/0.51 ops/s Mixed per-core min/avg/max: 0.24/0.32/0.48 ops/s [3/8] Float (FFT) Big ( 2 threads): iters 2 | throughput 0.74 ops/s | per-core min/avg/max: 0.37/0.37/0.37 Little ( 6 threads): iters 6 | throughput 1.51 ops/s | per-core min/avg/max: 0.18/0.25/0.36 Mixed ( 8 threads): iters 8 | throughput 1.79 ops/s | per-core min/avg/max: 0.16/0.22/0.35 Speedup: big 1.93x little 3.93x mixed 4.65x Efficiency: big 96.3% little 65.4% mixed 58.1% Big per-core min/avg/max: 0.37/0.37/0.37 ops/s Little per-core min/avg/max: 0.18/0.25/0.36 ops/s Mixed per-core min/avg/max: 0.16/0.22/0.35 ops/s [4/8] Hash (SHA-256) Big ( 2 threads): iters 90 | throughput 44.91 ops/s | per-core min/avg/max: 22.44/22.45/22.47 Little ( 6 threads): iters 192 | throughput 93.90 ops/s | per-core min/avg/max: 12.15/15.65/22.60 Mixed ( 8 threads): iters 241 | throughput 117.73 ops/s | per-core min/avg/max: 12.10/14.72/22.54 Speedup: big 1.98x little 4.14x mixed 5.19x Efficiency: big 99.0% little 69.0% mixed 64.9% Big per-core min/avg/max: 22.44/22.45/22.47 ops/s Little per-core min/avg/max: 12.15/15.65/22.60 ops/s Mixed per-core min/avg/max: 12.10/14.72/22.54 ops/s [5/8] Memory Sort Big ( 2 threads): iters 4 | throughput 1.06 ops/s | per-core min/avg/max: 0.53/0.53/0.53 Little ( 6 threads): iters 6 | throughput 2.01 ops/s | per-core min/avg/max: 0.23/0.33/0.49 Mixed ( 8 threads): iters 8 | throughput 2.28 ops/s | per-core min/avg/max: 0.19/0.29/0.47 Speedup: big 1.78x little 3.37x mixed 3.83x Efficiency: big 89.1% little 56.2% mixed 47.9% Big per-core min/avg/max: 0.53/0.53/0.53 ops/s Little per-core min/avg/max: 0.23/0.33/0.49 ops/s Mixed per-core min/avg/max: 0.19/0.29/0.47 ops/s [6/8] Recursion (Fibonacci) Big ( 2 threads): iters 2 | throughput 0.17 ops/s | per-core min/avg/max: 0.08/0.08/0.09 Little ( 6 threads): iters 6 | throughput 0.36 ops/s | per-core min/avg/max: 0.04/0.06/0.08 Mixed ( 8 threads): iters 8 | throughput 0.43 ops/s | per-core min/avg/max: 0.04/0.05/0.08 Speedup: big 2.00x little 4.20x mixed 5.07x Efficiency: big 100.0% little 70.1% mixed 63.4% Big per-core min/avg/max: 0.08/0.08/0.09 ops/s Little per-core min/avg/max: 0.04/0.06/0.08 ops/s Mixed per-core min/avg/max: 0.04/0.05/0.08 ops/s [7/8] Math (Pi) Big ( 2 threads): iters 1492 | throughput 745.39 ops/s | per-core min/avg/max: 370.74/372.70/374.65 Little ( 6 threads): iters 2263 | throughput 1130.50 ops/s | per-core min/avg/max: 94.95/188.42/375.07 Mixed ( 8 threads): iters 2642 | throughput 1319.41 ops/s | per-core min/avg/max: 95.16/164.93/374.73 Speedup: big 2.03x little 3.07x mixed 3.58x Efficiency: big 101.3% little 51.2% mixed 44.8% Big per-core min/avg/max: 370.74/372.70/374.65 ops/s Little per-core min/avg/max: 94.95/188.42/375.07 ops/s Mixed per-core min/avg/max: 95.16/164.93/374.73 ops/s [8/8] Compress (zlib) Big ( 2 threads): iters 20 | throughput 9.51 ops/s | per-core min/avg/max: 4.73/4.75/4.78 Little ( 6 threads): iters 40 | throughput 18.53 ops/s | per-core min/avg/max: 2.21/3.09/4.78 Mixed ( 8 threads): iters 50 | throughput 22.02 ops/s | per-core min/avg/max: 2.09/2.75/4.62 Speedup: big 1.96x little 3.81x mixed 4.53x Efficiency: big 97.8% little 63.5% mixed 56.6% Big per-core min/avg/max: 4.73/4.75/4.78 ops/s Little per-core min/avg/max: 2.21/3.09/4.78 ops/s Mixed per-core min/avg/max: 2.09/2.75/4.62 ops/s ════════════════════════════════════════════════════════════ Score Report (Heterogeneous: 2 big + 6 little) ════════════════════════════════════════════════════════════ Test Single Big Little Mixed Big eff Little eff ──────────────────────────────────────────────────────────────────────────────── Integer (Prime Sieve) 11592 22728 36743 43657 98.4% 53.0% Float (Matrix Mul) 2733 2282 3899 5183 74.2% 42.3% Float (FFT) 142 247 503 596 96.3% 65.4% Hash (SHA-256) 1514 2994 6260 7849 99.0% 69.0% Memory Sort 203 354 670 761 89.1% 56.2% Recursion (Fibonacci) 29 57 119 144 100.0% 70.1% Math (Pi) 1852 3727 5652 6597 101.3% 51.2% Compress (zlib) 972 1901 3705 4405 97.8% 63.5% ──────────────────────────────────────────────────────────────────────────────── Single geomean.................... 719 Big group geomean................. 1236 Little group geomean.............. 2289 Mixed geomean..................... 2760 ════════════════════════════════════════════════════════════ Big/Little Power Comparison ════════════════════════════════════════════════════════════ Big/Little power ratio (from per-core probe): 3.58x Big avg ops/s: 1.99 Little avg ops/s: 0.56 Power composition: Big (2 cores): [████████████████░░░░░░░░░░░░░░] 54.4% 54.4% Little (6 cores): [█████████████░░░░░░░░░░░░░░░░░] 45.6% 45.6% Equivalent big cores: 3.7 (little cores converted to big power) ══════════════════════════════════════════════════ Single score: 719 (fastest core) Big group score: 1236 (big cores only) Little group score: 2289 (little cores only) Mixed score: 2760 (big+little mixed) Mixed / Single: 3.84x ══════════════════════════════════════════════════ Benchmark complete! Edited 1 hour ago by humanus 0 Quote
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.