ODROID-XU4: Much Better Performance Than The Raspberry Pi Plus USB3 & Gigabit Ethernet @ $60

[Enverex]

I have some comparisons to cheap x86 systems if you are interested

Not sure how you managed it, but something's completely wrong with that N3160 system, it should be a decent amount faster than the ancient AMD E-350, not multiple times slower. You need to make sure something's not wrong with that machine because as it stands right now, it's highly misleading.

[t.s.]

Another SBC bench with numbers on 7z, AES-128 (16 byte), AES 256 (16 KB), memcpy, memset, kH/s, what distro and kernel the bench using: https://github.com/ThomasKaiser/sbc-...ter/Results.md

[t.s.]

The 5422 is damn slow in antutu benchmarks;

https://www.fonearena.com/blog/101117/samsung-galaxy-s5-exynos-5-octa-benchmarks.html

For 60 usd you can buy an used Dell intel core 2 duo 2.8ghz laptop. If you want to tinker, mbed boards are more suitable to small embedded projects. STI Nucleo boards are cheap and good: https://os.mbed.com/platforms/ST-Nucleo-F401RE/

1. Depends on what the user need. My client go for laptop at first, then he ask for mini server (so, ARM SBC)
2. core 2 duo laptop use ~34watt when idle. Compare to ~3watt idle.

[t.s.]

Call me when it's ARMv8. Until then I'll stick to the C2.

Comparable performance? https://www.friendlyarm.com/index.ph...product_id=234

[t.s.]

Thanks for that article! Great work! I always have to remind myself when looking for low-power systems that there are many x86 systems which can do the job as well [and arguably better]. Even for a little more money [and energy usage], hardware support from all of the major OS's is miles ahead of most ARM boards.

If you can get CPU + mobo, mITX x86 systems from your local retailer, for ~$75 and use ~19.5v DC adapter with comparable performance, go for it. But then, the now popular RK3399 ARM board is quite a beast and sell for about $39 - $95.

[t.s.]

Hmmm, this could be the first ~$60 board that could run xscpu64 emulation at 100% speed (Commodore 64 with SuperCPU card emulation from Vice emulator, very processor intensive). If it could do that, it would be a buy from me...

There's serveral ARM sbc that has that kind of performance for ~$65, e.g: Rock Pi 4 ($39 - $65); Nanopi NEO4 ($45); nanopi M4 ($65 - $95);

[willmore]

I'm glad that Hardkernel responded to my request to send you a board to evaluate, Michael. I have always found them to be a very forthcoming and supportive company. It's worth pointing out with regards to the XU4, that the board is already three years old and that HK plans to support it for some time to come. In that sense, they are much like the RPi Foundation--they commit to providing and supporting their products for a long term. I believe that makes these two companies separate from the many other board vendors whos products come and go and receive support when their creators feel like it.

The performance of the XU4 surprised me for several of these tests. I never expected it to beat the TX2 in anything. Two Denver cores and *four* A57 cores getting beat by 4 A17's and 4 A7's? I would have expected the A57's alone to beat the XU4.

Hardkernel has put in a lot of effort (and money) to get their board supported by the mainline Linux kernel. I believe that's related to their commitment to provide long term support for their products. The GPUs in the C1, C2, and XU4 are all ARM Mali, so HK can only do so much to provide open support for them. They do provide full hardware acceleration support for them in their OS releases. Don't forget that the Lima project is working towards making an open Mali drivers (and Panfrost is doing similar for the higher end ARM GPUs).

To address the comments about the USB ports. There are three external USB ports on the XU4. One is a 2.0 port that goes directly to the SoC to a proper USB Host controller. One is a USB 3.0 interface that is dedicated to the GigE controller. The last one is a 3.0 interface that goes to a two port switch which connects to two 3.0 ports on the board. So, it has vastly higher I/O bandwidth than the C1, C2, or any of the Rpi boards. It also has an eMMC port which can provide up to 128GB of high performance storage. The same goes for the C1 and C2. Looks like you used a 16GB eMMC card, so that's good.

[m132]

Going in hand with this Exynos5 SoC is 2GB of LPDDR3 RAM, HDMI 1.4 supporting 1080p output, one USB 2.0 port, two USB 3.0 ports, Gigabit Ethernet, 802.11n WiFi, 30-pin GPIO header, and all of this fits on a PCB just 83 x 58 mm.

You sure about that WiFi? It's also worth noting, that despite the SoC having 2 USB 3.0 interfaces, one of them is used (and imho wasted) for GbE and these 2 regular ports are attached to a hub.

[riklaunim]

Nice! Albeit something seems to be very wrong with the Celeron N3160. This 4-core Celeron (Airmont core, Braswell) isn't exactly a speed demon, but it ought to be a lot faster - no way this chip is 20 times slower than the newest Goldmont Plus based Pentium...

I'll be doing more tests of the Chinese boards and in this case it may be some sort of fixed TDP scenario this CPU is locked to. This should be visible in clocks, turbo frequencies and if it will be to hold those turbo frequencies.

[brrrrttttt]

Call me when it's ARMv8. Until then I'll stick to the C2.