Last month, we wrote about Allwinner D1 SBC & processor that promised to offer a relatively low-cost RISC-V Linux solution. We were not given a name at the time, but there was a logo of Nezha, a fictional character from Chinese literature.
The board is now known as the Nezha SBC and has been launched on Indiegogo for $99 and up as a board designed for IoT projects running Linux, but can also be purchased directly on Aliexpress for the same price. [Update: It can also be purchased on Taobao for 599 RMB]
- SoC – Allwinner D1 single-core XuanTie C906 64-bit RISC-V processor @ 1.0 GHz with HiFi4 DSP, G2D 2D graphics accelerators
- Memory – 1GB DDR3 memory
- Storage – 256MB SPI NAND flash, MicroSD card slot
- Output – HDMI 1.4 port up to 4Kp30, MIPI DSI & touch panel interface up to 1080p60
- Decoding – H.265 up to 1080p60 or 4Kp30, H.264 up to 1080p60 or 4Kp24, MPEG-1/2/4, JPEG, VC1 up to 1080p60
- Encoding – JPEG/MJPEG up to 1080p60
- Audio – 3.5mm audio jack, digital audio output via HDMI, a connector for microphone board
- Connectivity – Gigabit Ethernet via RTL8211F PHY, 2.4GHz WiFi 4 & Bluetooth module via XR829 module
- USB – 1x USB 2.0 host port, 1x USB Type-C OTG portport
- Expansion – 40-pin GPIO connector
- Debugging – 4-pin UART header, USB ADB debugging supported
- Misc – Power LED, tri-color user LED, OK & FEL buttons
- Power Supply – 5V/2A via one of the two USB-C ports
- Dimensions – 85 x 56 mm (6-layer PCB)
The standard version of the Nezha SBC ships with a 5V/2A power adapter, some cables, and a 16GB MicroSD card pre-loaded with Debian for $99. But there are also bundles with accessories based on the standard version of the kit:
- $109 Nezha Vision Suit with a 720p USB camera board.
- $115 Nezha Vision Suit with a 6-mic array also including 5 buttons
- $149 Nezha Panel Suit with an 8-inch 1280×800 IPS LCD with a capacitive touch panel and acrylic bracket.
- $170 Nezha Full Suit with the USB camera board, the 6-mic array, and the 8-inch display
The Indiegogo campaign has a flexible funding goal of around $10,000 with perks expected to ship in June 2021 right after the campaign ends. Shipping adds from around $2 to China and to about $14 to most of the rest of the world. It’s not faster to purchase the board on Aliexpress with shipping scheduled for June 23.
Note the board may not be for everybody, as there’s no 3D GPU, and performance will be on the low-side with a single-core 64-bit RISC-V core clocked at 1.0 GHz. The board is mostly suitable for people wanting to play around with Linux on RISC-V or develop IoT projects.
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.
49 Replies to “Nezha RISC-V Linux SBC launched for $99 and up”
> Tina, a fork of OpenWrt, officially supported by Allwinner,
It looks strange to read the words “supported” and “Allwinner” that close to each other without the name of someone doing their job for free in the middle. Time will tell but I’d rather parse that as “forked by allwinner” (i.e. without implying any support nor updates, as they did for their (in)famous 3.4.39 kernel long ago).
There are more basic RISC-V architecture issues to be addressed first, like http://lists.infradead.org/pipermail/linux-riscv/2021-May/006481.html before seeing first class mainline support.
I am not aware that anyone from Allwinner or Alibaba contributed to OpenWrt or tried to This SoC is not supported by upstream OpenWrt. If Tina is still based on OpenWrt 14.07, they need some really good people to maintain these pretty old software stack as upstream OpenWrt 14.07 is EOL since many years. 😉
This is AW we are talking about, of course the “good people to maintain” will be the linux-sunxi community.
Like willy said only time will tell
$100 Dollars for a single-core? What is this? BeagleBoards before the BeagleBone?
The indiegogo page mentions the global chip shortage as a factor, citing an article from November here on CNX Software with an estimated price of $12.50
Whether it was ever deliverable at that price is an open question.That’s an 800% markup; sucks to be an early adopter during a supply crisis.
In fairness, the EVB (this board) and the $12.50 (or “less” from Pine64) were always different boards. The really budget boards were likely to be closer the common K210 boards; likely breadboardable, less memory and storage, fewer I/O options, etc.
Allwinner made the decision to allot all their D1 chips to this specific board and NOT resell them to companies – including their launch partner for this product, Sipeed – for use in other boards like the budget board.
So, yeah, the chip shortage drove the price up, but it drove the price up AND allowed/forced the entire supply of D1’s to go to a single board that wasn’t really the board that got the most column space and mindshare.
At $100, the leap to BeagleV is pretty easy to justify. I have one of the prototypes and it’s pretty awesome AND they know how to get code upstreamed before launch. Allwinner doesn’t have a great history of doing that. IMO, this board is too expensive (and too much of a gamble) for a microcontroller and too wimpy for a “real” computer – but they didn’t ask me.
If they dedicated their whole supply to this board, then they must have *very* limited supply as I don’t see this board selling well.
I see the BeagleV selling much better at only $50 more- it’s got 5-10 times more to offer for that $50. I was (and still am) up for the BeagleV. It’s got things like an NVDLA core in it. This? It’s just overpriced. I could see $50-60 ish with the CPU core being hot enough to rate that. $100? XuanTie’s not THAT special.
I think is 100 because they are including an LCD display.
The one with the LCD is $150(USD).
David, I agree. Look at the counts on IGG and they’re not exactly flying off the shelves.
Frank, BeagleV starts at $120, with $150 being the 8GB version. Drew (from Beagle) is confident they can hit their manufacturing costs, but the chip shortage is on everyone’s mind. The Beagle bring-up team (which includes me) has been actively submitting patches upstream to get it landed in various OSes and tools for launch and they’re listening to us on design issues. (“Hey, instead of putting JTAG $HERE, why not have it over $THERE and 3.3V safe”, etc.) I think it’ll be a better launch experience for most people.
Everyone has their eyes on the Vector processing. D1, to the frustration of everyone, is shipping with an unratified version from about two years ago. (See Bruce Holt’s excellent posts here and elsewhere on the binary compatibility issues this makes.) Current Beagles have no vector processing at all, but we know we’re not on the final silicon, either. Implementing 0.7.1 and shipping it at scale could mean a lifetime of runtime detection, proprietary GCC/LLVM/GDB/Libc branches, and general frustration. I don’t get it.
EVB a really unfortunate combination of price and performance, IMO. Still, the people that want vector processing in silicon may see this as the best (only) option for now. Many of us are confused by Allwinner’s decision to sell the chip ONLY on their own board until supply issues are solved. Maybe this is exactly to manage expectations of the new SoC and to have only one board to roll if they do make another with RVV 1.0, but they’ve claimed in other places that this is the silicon they plan to go volume with, so that’s not their public plan.
I’ll be spending the weekend coding on BeagleV…
This is robbery
Compared to ARM’s Juno dev boards for $10,000? I never see anyone complain about their value for money vs a Raspberry Pi.
If you have another board with RISC-V processor (not microcontroller) at lower price I would like one. This is a first and low quantity batch. RISC-V emulation on an ARM board will be slow and will not be a real test. If you are not developer you probably have no interest in this kind of product.
It’s not so much that it’s that- it doesn’t offer half as much for 50 dollars less than the BeagleV, which isn’t “compelling” like a Pi4 is, but it’s awfully close. It’s worth the $150. This? Try $40-50 less.
How much is only the D1 SoC? I wonder if is a viable alternative to those SoCs with a single Cortex A7
Allwinner only sell D1 board and not open to sell D1 SoC
For now, yes. Not once they have good supply of chips. This is a mass-production chip which they will be making in the millions.
I agree with Tuff. Well, this one core is compared to HiFive U54 or ARM Cortex -A7 A35, what does it look like ????
Dhrystone 2.4 DIMPS/MHZ – ARM cortex-A53 -2.24 DIMPS/MHZ
Not too bad for a first RISC-V implementation CPU :). It take years to ARM to reach this level. So Allwinner get experience with it’s ARM based SoC too, but it is very promising for the future.
So, this is the same silicon that has dual A7’s on it and they either disable the Risc-V or the ARM cores? Sounds a lot like they tried to replace the OpenRISC power management core with a Risc-V core and somehow managed to not get them to run at the same time, so they though they’d make the best of it by fusing off parts to make two new exciting chips.
For the Nezha board price to be justifiable, the price for the D1 SoC should be at least 60 US$ which is hard to believe.
All other componnents on board are under 20 US$ even in small quantities (>=100).
I’m very enthusiastic about RISC-V, but if I can get the board with the similar characteristics/features for 3-4 times lower price or the similary priced board with 3-4 times better characteristics/features, why would I choose this board. Only because it is RISC-V?
Yes. This was made to start adoption of RISC-V, to get support and people start porting stuff to this ISA. The future is promising for RISC-V but it has to start somewhere
Sipeed announced to charge less than 13 bucks for their board featuring also the Allwinner D1 though with internal DRAM: https://www.cnx-software.com/2020/11/09/xuantie-c906-based-allwinner-risc-v-processor-to-power-12-linux-sbcs/
This is old news and may no longer true. Allwinner change their position and only selling board. However, Sipeed may be the exception 🙂
Allwinner will be selling the chips to all comers once they have good supply of them. Sipeed confirmed as recently as yesterday that they expect to have the promised $12.50 SBC using the same SoC later in the year.
Indeed this is absurd. And SoC vendors like AW are in part responsible for this. If at least they reused the exact same pinout as one of their other chips like A20, H3, or H5, it would suddenly be much easier and cheaper for their partners to put it on an existing board. Their A20 was a success precisely because it was a drop-in replacement for the A10, thus eliminating any design phase to adopt it.
There is nothing absurd. This is a high end dev board using early supplies of a chip that will be available in high volume later in the year. It allows software developers to get a head start before the cheap mass-production SBCs come out, so that Average Joe will have OSes and applications to run and drivers already available when they buy their $12.50 board.
I pretty much understand the concept of a dev board. The only thing is that until these chips start to reach decent performance levels, few companies will be interested in investing on them for their products, so I guess most developers will be enthousiasts doing it for the fun on their spare time. In this case, better simply give the boards away to developers as is usually done to help porting. This is the best way to get a fast ROI for the time spent designing the board.
> If at least they reused the exact same pinout as one of their other chips
Feature-wise the closest would be A64 but compared to D1/T1033 missing CVBS in/out and support for microphone arrays.
> A20 … was a drop-in replacement for the A10
Not exactly. While being pin-compatible you needed updated A20 information to design boards that can interchange both SoCs.
Just because of this H5 (the only 64 bit Soc you cited), has a 32 bits external bus. This could be enough for its target, but we expect to have a 64bit bus with 64 bits core.
This is a manufacturer’s evaluation board, simply being resold by Sipeed, RVboards, and others. As such it is an expensive full-featured board. Sipeed and Pine64 and others will build much cheaper boards of their own later in the year when production has ramped up and hopefully the general chip shortage has eased.
The most interesting thing about this SoC and board is it is the first RISC-V Vector processing hardware to ship, by probably at least 12 months if not 18. It implements the 0.7.1 draft of the spec, which is incompatible in a number of ways with the 1.0 spec that will be ratified this summer, but some code *is* binary compatible (e.g. a memcpy()) and it’s good for gaining experience with something at least similar to the final spec. I have test results showing considerable speedup:
ARM is also some way from having their similar SVE in consumer SoCs/boards.
Nice one! (Not sure about the price)
3D GPU is no issue here, but RISC-V needs to address this soon.
256MB SPI NAND – Hm I’ve seen 1-8GB NAND and 1-16MB SPI NOR.
Type-C – Does Allwinner D1 have full 3.1 USB Type-C or it is via extra chip?
Kudos to Linux-Sunxi. Mainline/Upstream will be
– quick and easy if same IP blocks.
– slow and hard if new IP blocks.
D1 has no USB 3 internally or externally capability. Just up to single USB 2 IO speed max.
it looks like they they reused their geometric 2D processor and video processor, so those part are already working on Linux. As they work well and already have driver, I wonder why their would restart from scratch ? That’s probably why for so much Linux distribution are supported. After Debian wiki, more than 95% of their packages are already compiled for RISC-V. I tried it (Bulleye versison) on qemu, I was able to compile everything I need without problems. There are nvidia/ati (xorg)video + (mesa)dri + (kernle)drm drivers too, so their GPU on cards could probably be used using USB external PCI connector ? Has someone tried this kind of things on ARM boards ?
The Ali link says “Sorry, this item is no longer available!”
… that went quickly …
I have more trust in Armbian providing an image for this board, instead of a Chinese fire-and-forget fork. So … will Armbian play a role (despite ARM in their name)?
Probably. The Yocto community might as well. I won’t speak for others, though, I’m a bit hard-pressed to sign off on $100. Claims of a, “high end board,” not withstanding (Guys, I work with real eval boards in my day job for Motorola Solutions…this is the same class of HW as the stuff we’re used to getting for $50…c’mon now…) it’s not one and it’s your basic eval board and I’ve limited budgets for supporting configurations with my embedded Linux distros. I’ll likely support the BeagleV. I’ll probably support the PolarFire config. I’m not so likely to drop $100 for this thing.
Let me coin the name “Fivebian”.
No it won’t. Forget about Armbian if you’re seeking ‘trust’.
I’m glad to see the way you speak your mind, while always aggressive there is truth to your arguments.
Ouch, the performance is lower than my very slow, 10-year old NAS with a very old ARM.
$ python3 pystone.py
Pystone(1.1) time for 50000 passes = 27.7582
This machine benchmarks at 1801.27 pystones/second
So: 1800. That is very low.
My old, slow ARM NAS: 3230
My Celeron does 98.000
My i3 laptop does 180.000
Use the SSH account provided on https://www.indiegogo.com/projects/nezha-your-first-64bit-risc-v-linux-sbc-for-iot#/
Then run pystone.py, as provided by SABnzbd for example
Why running pystone without benchmarking the benchmark first? The usual kitchen-sink benchmarks test compilers or interpreters, settings and whatever else’s performance without giving you the ability to compare different hardware, see e.g. https://stackoverflow.com/a/22354756
When I ran sbc-bench a few weeks ago on the Sipeed provided D1 board two things were obvious:
1) memory performance is abysmal but this might change every moment since we don’t know whether DRAM initialisation with D1 is final or not. As such benchmarks that depend on memory performance will suffer right now.
2) benchmarks test different stuff on different architectures. The openssl calls sbc-bench uses run optimised assembly on ARMv7 (and offload to ARMv8 Crypto Extensions if possible) vs. generic C on RISC-V. While this is what you get right now as OpenSSL user until someone does the same for RISC-V (assembly and/or using the RISC-V Cryptography Extensions) the produced benchmark scores are totally useless to compare these two platforms.
“The usual kitchen-sink benchmarks test compilers or interpreters, settings” … exactly. So pystone tells me what I as user would get.
The 1800 tells me I cannot use it a SABnzbd download station
Since I was curious why CPU horsepower should matter on something like a ‘download station’ I came across this (I guess that’s you?): https://forums.sabnzbd.org/viewtopic.php?t=18190
If this ‘benchmark’ gives a 1st generation RPi (ARM11 at 700 MHz) a 5555 score and the D1 not even a third I guess it’s safe to trash all these pystone scores since completely irrelevant. As to why the underlying ‘benchmark’ back from 1984 already sucks for comparing different systems you might want to do a web search for ‘brendan gregg dhrystone’.
When looking at the title i knew there would be ton of comments about the prices, and wow they are alot of them.
We should take a look at the targer no AW is producing.
Indigogo target is only for 10K USD, or around ~100pcs.
Aliexpress sold out at only 16pcs.
I asume about the same qty on taobao, at <20pcs
At less than 200~300pcs target for production boards, no wonder its expensive.
this is too little IMO, even RPi original target is 1K board on release date.
Not giving AW anymore of my money, when they dont even have the confidence in their product.
Not that desperate yet.
Now plenty on Aliexpress just search Nezha Risc-V or for a different Risc-V board. Search Sipeed Maixdunino K210.