Ubuntu will not support RISC-V CPUs without RVA23 support going further, so this is stuck on 24.04 forever. There is no official Debian image either, and the Ubuntu version uses fixed kernel 6.6.47 with no further updates (it is not even installed from a repo).
I own one since a couple months ago but I already regret the purchase.
I owned every generation of Raspberry Pi and I still use the 5, 4, 3, and even a 2 Model B. Their software/ecosystem support is unmatched by any other SBC line.
It's actually worse than that, even getting updates kills the image. I did manage to work around it. but overall I don't consider it a platform for anything I would depend on. It's fun for experimentation, that is about it
Ubuntu is the worst of the Debian based Linux distributions. Give MX Linux a try. It's rarely evangelized, unlike other Linux distributions, but it has a surprisingly high user base, and for good reason. Everything it has added to Debian enhances it, and it doesn't try to promote any services, like Ubuntu does with Canonical's services.
Regardless of the Linux distribution, it's still way to early to get good support for RISC-V. Current products are good enough for general development, but RVA23 was ratified less than a year ago, and I don't know of any designs started after the ratification that have hit the market yet. Once that's the case, it should be easier to get universal support, but until then, every RISC-V SBC is a one-off development board.
Funny how software matters I bought a specific eink tablet and the lag and choppiness in drawing made me immediately sell it for something more well known.
Sucked too as I lost $300 but hopefully helps open source community anyway
Not bad at all, but the OpenWRT image still didn’t have Wi-Fi support a week or so ago, so I don’t know how good software support is going to be in the long run.
The big ecosystem of SBCs confuses me a bit. Who is buying these?
The work required to build an actual secure, maintainable product on top of an SBC is so big that you'd surely never use one of these. The hard work is all in software. You need a supplier with product lifetime guarantees and a known SoC manufacturer.
If you're a hobbyist, unless you really don't value your time you'd be much better served buying an x86 PC or a Raspberry Pi for whatever project you've got. Any money saved buying one of these would be completely negated by the extra time taken to maintain it.
So who's the target market? Are there products out there built on these? Or are they mostly just shipped straight into desk drawers? How many of these do they actually ship?
> The hard work is all in software. You need a supplier with product lifetime guarantees and a known SoC manufacturer.
We need easily accessible documentation on the hardware so anyone can maintain the software. The problem I see is most SoC makers either have poor or no publicly available documentation. Their poor excuse for "open source" are undocumented Linux drivers for their black box hardware and a near obsolete kernel that they may or may not update.
These chips are made by big companies for big companies and not hobby people. They expect that you're going to buy in volume, sign NDAs and possibly license IP from them. The chip goes into a product that is likely never going to be updated and thrown out in 5 years.
You don’t need that many customers to design and ship an SBC. The hobbyist market is enough to sustain these companies. They can sell accessories to increase their margins. Their contract manufacturer can incrementally build batches of the boards.
Many of these boards serve as development boards for the parts they include. If you want a dev board to try this part or you need a cheap RISC-V system to test RISC-V things on, buying one of these is an easy choice.
> You don’t need that many customers to design and ship an SBC.
Right. An SoC takes five years and tens of millions of dollars.
An SBC using that SoC can be designed and made in three months for the cost of a small team's salary -- possibly even just one person e.g. see Paul Stoffregen of Teensy fame, or Jonathan Oxer at Freetronics. I'll bet the teams at Sipeed, Banana Pi, Orange Pi etc are pretty small too.
There are many products that are only expected to sell in low volumes and high margins. Also, prototyping, proof of concepts, one-offs, etc. Those are all areas that I've seen them used.
Most of the SBCs are supported by fairly standard Linux distributions: Debian/Armbian or Fedora; you just boot from the approriate image on an SD card. Some SBCs have eMMC storage and/or a M2 connector, so you can either keep running off the SD card or transfer the image to that storage.
The value proposition of using SBCs is in their embedded connectivity; in addition to standard USB/network/HDMI ports they tend to have built-in connectors for:
* MIPI input, for video cameras
* MIPI output for LCD panels
* i2c and SPI for weird peripherals (accelerometers, temp sensors, etc)
* i2s for sound
* GPIO and timers/PWM for custom peripherals like motors/servos, contacts, programmable LEDs etc.
I mostly use them for things specifically designed to not need software maintenance. I use them in test fixtures and other embedded systems. Anything that is connected to a network isn't connected to the internet, and regardless of a network connection, I always set the filesystem as immutable and create backups of the SD card, so if anything goes modified, restarting usually fixes it, but if not re-imaging the SD card for sure will.
> So who's the target market? Are there products out there built on these? Or are they mostly just shipped straight into desk drawers? How many of these do they actually ship?
Hobbyists, mainly.
I could see this being a hit with people who want to work with RISC-V, which still needs a lot of low-level stuff built out for its ecosystem. You don't need it to be a screamer, just for it to run predictably.
Remember, the Raspberry Pi was mainly a hobbyist curiosity when it came out. There's definitely a market.
Maybe for hobbyists it's more of practicing how to play around with embedded systems? Over a decade ago I sorta got burned by Odroid dropping support for one of their early units despite being much faster than the 2Bs that were selling at the time (plus the annoying cost of emmcs or whatever), so I absolutely agree with your point.
So actually one of the thing that attracted me to the RV2 is the PCI Express slots. I was able to add a SATA controller and an NVMe drive to the same unit.
I can accept the performance issues for now- it's an emerging platform, and hasn't had the huge amounts of resources poured into it like ARM. There's a chicken-and-egg problem here, since those resources will be limited since people aren't buying RISC-V equipment, which limits the incentive to commit resources, etc.
But what I cannot accept is the truly awful documentation and software support from the vendors. This is where Raspberry Pi shines, and is IMO one of the most significant factors in its success. I'm excited that RPi is dipping their toes into the RISC-V pool with the Hazard3 cores in the RP2350- perhaps they will be able to release a Raspberry Pi RISC-V edition board some day.
But I'm hesitant to buy one of the current RV SBCs, so I guess I'm part of the problem.
I'm also surprised that there aren't any startups producing small, simple CPUs and SOCs outside of China (as least, none that I'm aware of). Is there no investment available in India, N. America, Japan, Europe, Israel (* not bringing the current situation into this, just noting they have chip fabs)? Fabricating chips is not cheap, but the first ones don't need to be the top-of-the-line TSMC 3nm process.
Poor vendor software support is also expected, because it's too early into development of the RISC-V architecture for software to have universal support on different hardware variations.
If you're not working on RISC-V-specific development, than it isn't a product for you, but with step-by-step guides for building uboot and the linux kernel, as well as running various LLMs on the TPU, it is very well supported for the current RISC-V audience.
And I think a language like golang can be a really really nice fit given how it can be compiled really fast towards risc-v as well
Maybe java also runs in risc-v I am not sure, surely people are working on java support I suppose.
People buy risc-v to support an open standard and to not worry about licensing fees.
Isro (india's nasa basically) uses some risc-v chips to not license arm chips etc. because of either better national security (to have less arm influence) or because they don't want licensing fees given how rudiculously price efficient isro is.
It wouldn't shock me if performance doubles over the next couple years. This chip is one of the first supports the RiscV Vector extension, so compilers have a lot of years of catch up to vectorize effectively.
Don't suppose there's actually documentation for the CPU anywhere?
(I mean more than a tiny "datasheet" with a very high level overview and/or a pile of random Linux/uboot patches)
RISC-V going forward, one of the only beacons of hope in the silicon world.
I need one of such devices for my self-hosted services. And it will be time to port from C to assembly, really, because we have finally a CPU ISA which is 'sweet spot' balanced, standard, global, pushed forward with significant resources and without IP locks anywhere. No more developer/vendor lock-in via "the only compiler able to generate correct machine code", extremely hard to do planned obsolescence, etc, we need mainstream adoption NOW :)
The main blocker: how do I buy such device with a noscript/basic (x)html browser? And no way I use a credit card on a web site: would require well identified bank swift account, or wallet codes bought from local and physical currency terminals. I don't know of any local retailers I can buy such device from. Yep, the "web geniuses" at amazon (which supports wallet code) broke noscript/basic (x)html support a few years ago.
I use Privacy.com to get temporary credit cards, to avoid the hassle of compromised credit card numbers, although it doesn't stop various governments from tracking my transactions.
I've bought RISC-V SBCs from both Pine64.com and Arace.tech, and neither required I make an account. Arace.tech does require JavaScript to checkout, but Pine64.com does appear to work without it, although I didn't complete a transaction without it. Pine64.com also accepts USDC payments, but no other cryptocurrency.
I used to occasionally buy a single board computer ~10 to ~5 years ago.
Then I waited until SBC's reached the 32GB RAM level. The first such affordable and performant board was an Orange Pi 5+ or 5B (I should double check, may add a comment later).
I believe I wanted the Orange Pi 5+ but it was sold out, so I ordered and paid for the Orange Pi 5B (the 2 ethernet port variant) which was a bit more expensive but was still for sale. Both had 32GB RAM, my main requirement. There were multiple "flavors", with power adapters or with case or with memory card or eMMC etc. I chose the memory/eMMC version.
I sent a message to ask them to give me heads-up when they are about to ship mine, and then I was patient.
Then I waited, and waited, and waited.
Too patient, after a long while I start looking up on forums if other people are also waiting. I discover I am not the only one. So I take up contact again and ask when the board will be shipped.
They inform me the SBC is no longer manufactured, and offer me inflation-devalued currency.
I check which single board computers they still sell, and indeed they no longer sell the 5B variant, but now the 5+ is back in stock.
I ask them if they can just ship me the 5+ instead of the 5B. They refuse.
OK, I ask them how much I can pay extra so they ship me the 5+ instead of their unilaterally discontinued 5B.
They refuse.
A few months pass by. I ask again if they intend to ship the 5B as agreed, or the 5+ as a substitute.
So here comes the orange Lie:
They claim they shipped it, and provide a DHL link.
I first name (same as my father's) is German, even though I live in Belgium.
Their DHL link, is a shipment to somewhere in Germany, with a weight far below the weight of a single board computer, and which was delivered just minutes before their sending this message claiming shipped delivery.
I confront them that the weight of the SBC is advertised on their own site, and their DHL delivery link lists a value far below it, that I live in Belgium and the shipping address was in Belgium (I have 0 links to Germany), and that the timing of the delivery and their response message is so close it suggests people at customer support (presumably without arbitrary access to deliveries outside of the case) asked colleagues to let them know if a case pops up with a German delivery, so they can manually copy and paste bluff delivery of product X to customer Y as if it was my order to me into the message.
I confront them and ask them to answer a numbered list of questions.
They refuse to answer the questions (they can't without incriminating themselves), instead they offer me my money back.
Don't buy into the orange lie.
If you work for Orange Pie, feel free to msg me with a way to contact you, if you positively resolve my case I will remove this message.
I was wondering what was going on, because Orange Pi only sells to distributors, not direct to customers.
Amazon only has a single listing for any given product, and all distributors, whether authorized or not, are under the same listing. When there's more than one distributor, you can choose which one to order from, but the buy now button follows a default.
If it isn't too late, I would highly recommend disputing through Amazon, as disputes are the only way Amazon addresses problem distributors. Many know how to game the system, so if you communicate directly with them, they will try to delay you until after the dispute period has ended.
It was the only amazon distributor as far as I know.
they were directly linked from the Orange Pi 5 website.
Its their responsibility they direct their customers to dubious distribution systems (IF THAT IS THE CASE).
We should never reward orders or preorders in any fast devaluating area (like electronics) to be reneged after inflation: this would allow manufacturers to post-select a financial path: if less inflation than expected fulfill order, if more inflation than expected renege the deal.
Amazon can be rather dubious, but there's not a lot of competition to chose from. Orange Pi is far too small of a company to set up their own international payment and delivery system. I have used Ali Express, and they've been pretty good at protecting me from attempted fraud.
Orange Pi should fulfill my order, which I paid for, and still patiently await to see the board. If they feel it is the fault of Amazon (the platform of their choice) or the distributor of their choice, they can complain there and ask them for remuneration for reputational damage and cost of unfulfilled orders.
All I know, is that I paid via the Orange Pi endorsed channel and never received the board.
I have the right to warn others.
Hiding behind front company distributors to pull of inflation postselection tricks or whatever malice or incompetence it turns out to be should not be rewarded.
Yes, lots of Intel and AMD based SBCs have SO-DIMM slots, but you'll have to accept the 3.5" format.
Regarding RISC-V SBCs, there was serious consideration to release the Milk-V Oasis with SG2380 and LPCAMM2. But this didn't work out as the SG2380 was held up by geopolitical issues.
Not really. Most in the ARM space, at least, are soldered on or you need to switch out the entire compute module. Intel ones (not mini-PCs, but industrial gear and things like the LattePanda) also tend to have soldered RAM.
Pine64's claim to fame is that they have really good documentation, surpassing even the Raspberry Pi. That's what creates a good community around their products.
Most customers couldn't port software to an SBC, but for the ones who can, having all of the documentation makes it trivially easy, and having any that share their work makes it available to the whole community.
For my use case, and most Pine64 customers, I'd rather have the hardware documentation than off-the-shelf official support for a software stack. Raspberry Pi has an entirely different user base.
Orange Pi fills a similar niche, and really anyone releasing RISC-V SBCs at this point does too, as it's too early into the development of the architecture for microprocessor-level products.
There's a RV64 port of Debian and the RV2 and R2S are on the list of compatible hardware. No guarantee it'll be easy getting it loaded, it was like pulling teeth to get it on the SiFive U74 board, but that was 7 years ago, so it's GOT to be better by now.
I caught Orange Pi doing misleading advertising earlier, trying to make people hear that the CPU has an NPU. Does this actually have any AI hardware or not?
That looks quite similar to what I read before. For me those performance figures don't necessarily suggest it being any faster than just running on CPU so I'm still suspicious.
Might research this a bit more. Of course even if there's no perf improvement but perf/watt is better, that be valid too. But sounds very marketing speaky
It's not faster than running on an x86 CPU, but it's faster than an unaccelerated RISC-V CPU.
The entire platform is more for development of RISC-V software, especially at the operating system level, than it is an end-user product. There's still a lot of development needed to get RISC-V perf/watt, even with acceleration, to match ARM and x86 systems. I would expect at least a couple of years until then. RISC-V microcontrollers are already getting there, competing well with M0, but that's an entirely different market.
Four of the eight cores have matrix accelerators. If you are using the full 2 TOPS of accelerated performance, you have four more otherwise symmetric cores free to run regular workloads, otherwise you have eight. I'd rather have this setup than 4 regular cores and a separate asymmetric accelerator that provides 2 TOPS to matrix/tensor calculations but otherwise sits idle.
Hopefully Adafruit or someone could get these[1] out of "Contact Us" jail. 16GB 30TOPS BF16 in M.2 2280 at $369. PCIe 2x8 low profile "Duo" configuration available at $799. Supposedly. I believe the theoretical performance is Strix territory if these could be clustered, but only if they mass manufacture these.
Heh, interesting you call this "Contact Us Jail". I met sales and engineering staff from EdgeCortix at DSEI Japan 2025, and was very enthusiastic to pick up some of their products as alternatives to the Hailo-8L RPi cards.
So far....nothing. Not even an ETA on when they will be shipping hardware.
DSEI! I wish I had some official reasons that I'd be obligated to go and have fun in events like that :p
My fear is that they could be in endemic Japanese death spiral disease of hand-wavingly waiting for massive orders while not doing much to secure one.
There are so many cool stuffs and perfect plans from Japanese large companies and government agencies that are pushed forward up to one step before volume production or commercial rollout and then stalled until funding dries up and everyone moves on. Which, if I were to explain by speculating, might be due to people prioritizing own job security over corporate revenues, due to low psychological safety and perceived invincibility of large organizations, as one among many consequences of employment ice age phenomenon.
But whatever the mechanism is, it'll be sad if this one follows that path. Hopefully they make and ship them in substantial numbers. That should be their goal anyway.
for tasks like face recognition and object detection, would this type of hardware have good performance in real world cases? or, what is the standard hardware that devs use for tasks like that?
On Raspberry Pi, the GPU is the only thing that makes a responsive GUI or web-browser feasible, and is the primary reason most people use the HDMI LCD screens for games etc. It also took a large effort to bring up a v4l2 kernel driver for the camera modules etc.
For example, on the CPU one may pin all cores to stream a USB camera or software decode h264. With the SoC GPU decoding or streaming with the v4l2 interface might take up 30% on one core (mainly to handle the network traffic.)
The Raspberry Pi are not the fastest or "best" option (most focus on h264 or MJPEG hardware codecs), but the software/kernel ecosystem provides real value. Also, the foundation doesn't EOL their hardware often, or abandon software support after a single OS release.
A cheap RISC-V SBC is great, but ISA versions are generally so fractured (copied the worst ideas of ARM6)... few OS will likely waste resources targeting a platform that will have 5 variants a year, and proprietary drivers.
A Standard doesn't even need to be good, but must be consistent to succeed. =3
Low-power processors rarely have the AI accelerated instructions in the GPU, instead opting either for dedicated matrix/tensor cores, or as is used in the case, adding the acceleration instructions directly to the CPU core.
This results in a higher performance per Watt, but doesn't scale well to higher-power applications.
Part of the problem is that every ASIC manufacturer (and indeed each fabrication process) has a different toolchain with a different set of primitives for circuit design. Yosys and other open tooling for FPGAs has helped a great deal in lowering the barrier to chip design and by association reuse of circuits. But every ASIC, at the moment, is tied to some vendor's PDK. Here's the one Google open sourced for Cypress Semi's SKY130 process node: https://github.com/google/skywater-pdk
It is at least theoretically possible to build a headless "GPU" from RISC-V processors that have the vector extension (RVV). RVV had been designed to be able to run programs compiled for the SIMT execution model that most GPUs use.
This Orange Pi RV2 has a small vector unit in each core, and could be used for at least prototyping the software until more powerful chips are available.
BTW. There have also been a couple hardware startups that have been working on commercial GPUs based on RISC-V's vector extension, with their own GPU-specific instruction set extensions for texture lookup and the like.
The vector instructions are only in four of the eight cores. There's also extra cache in those cores, but they are otherwise symmetrical to the other four.
Hardware patents are orthogonal to open source software. If a patent covers the hardware then someone who wants to manufacture the hardware needs to license the patent, but you were never going to get free-as-in-beer hardware anyway, and a hardware patent is independent of whether the hardware is fully documented or has firmware with published source code and a license that allows users to make changes to it.
NDAs in the context of a patent license should legitimately be considered grounds for patent invalidation. Patent literally means "open to public inspection" and the entire premise is that you get a temporary monopoly in exchange for openly documenting how your invention works in the published patent. If there is something about the invention for an NDA to cover it means you left it out of the patent application, which is essentially an admission of wrongdoing.
A patent is an alternative to a trade secret. You can't eat your cake and still have it.
Patents do exist on popular IP cores, but in general standards compliant Verilog libraries are just vendor specific IP products. Thus, the copyright, NDA, and license agreements already keep the IP fairly locked down. For example, figuring out dram timing and DDR bus control yourself is nontrivial.
Some groups have attempted open IP cores, and made some progress:
However, the effort involved in getting standards compliant ASIC built puts folks in a Fabless manufacturing sector. Most firms that survive, will choose to stay with a generic FPGA option to avoid custom silicon unless absolutely necessary.
Patents are often useless/vague in many places, but on occasion may prevent platform decay for a few years. One can be sure a unique/new design will not go to fab unless such protection is in place. =3
Publish the software that does this for free so that more customers come to you instead of using FPGAs or just not making the attempt. Make it easier to design new chips so that more people do it and you get more customers.
The point isn't to help your competitors, it's to help your customers. Qualcomm or Apple doesn't have the incentive to help their competitors design chips, but TSMC or Samsung does have the incentive to help Qualcomm's competitors design chips, because then there are more companies making chips and the fab gets more business and prevents any one of their customers from getting too much leverage over them.
Samsung actually had a technological lead in many areas, and it translated into real revenue. They are a business, and exist to provide utility to consumers in exchange for investor profit:
Many incentives to push technology forwards are high-risk/expensive investments, and expecting the public/customers to willingly help pay that cost is naive:
I assure you academic funding does not cover such large costs, government grants are only a fraction of expected taxes in late stage Technology Readiness Levels, and competitor/cloner fractured markets erode fiscal returns needed to pay for the total incurred project cost.
Apple makes minimal utility products, but relies on intangible branding to maintain perceived value. Thus, only Apple could get away with selling zero chip designer handbags, and would still make absurd revenue (not a real product yet.) Steve Jobs observed very early, that selling raw motherboards was a low margin business. Which it was why the company shifted into consumer products.
For almost every other brand, consumers have shown they prefer the near material cost in opportunistic China/India factories, and thus simply ignore most firms products that include 10 years of R&D costs to pay back the investors.
There are many shelved technologies that will never see a Patent or the public markets. This is because the conditions are not ready for advanced products yet, and competitors irrationally nurture the lowest value volume market sectors. Thus, everyone gets a 15% value boost at regular intervals, and people remain excited about 3 decade old technology.
Qualcomm cellular chip product lines essentially lived off iPhone sales. Like any loyal dog, they will unlikely bite the hand that feeds them...
Rule #3: popularity is not an indication of utility.
I bought the 32 GB emmc module for it, for the root filesystem. I have a 500 GB nvme drive for everything else. I believe an nvme-to-sata riser will work, but I don't have one to test with (plus you'd need to power the sata drive with something else).
There's M.2 cards with PCIe to SATA bridges, which are great for a high-capacity low-speed NAS setups, especially with mechanical hard drives. Be sure to research the chipset before buying one, as some are notorious for poor driver support.
If this is a practical application, I would recommend sticking with ARM for this generation, as RISC-V is still in early development. In a year or two we should start getting more standardized RISC-V processor implementations, with broader support.
Ah, cool. I'd like the same, but with ECC RAM, so the options are even fewer.
Have you looked into what someone suggested here, splitting the NVMe to some SATA ports? It didn't sound like a horrible solution, but I have no experience.
It's pretty much the minimal cost for the form factor and accessories. It would be cool if they could make a $20 version with a lot less connectivity in a smaller form factor.
Check out the the Bouffalo Labs BL808 or Sophon SG2000 series. They are a system-in-package products that contain a RISC-V CPU, RAM and flash memory all in a few mm square package, for a few dollars. Pine64 and Milk-V both have breakout boards for under $10.
They only have RAM and flash in the hundreds of megabytes, not gigabytes, but they can run full Linux and are as capable as many household Wi-Fi routers. They include USB, Bluetooth, Wi-Fi, and wired Ethernet ports, and even have MIPI CSI and DSI interfaces for cameras and displays.
The SG2000 is pretty interesting. I do wish they had gone with a more homogeneous architecture (4 identical cores like the pi 2 zero), and it's RVV 0.7, but the next generation of this could be really interesting.
Depends what you're using it for. A lot of people tend to buy pi-likes as servers which is absolutely bonkers. If you time eBay right, $50 would get you a fairly powerful intel NUC with much more performance and peripherals
I don't think it's bonkers. For running a true home server sure, there's more powerful things out there. But for hosting something like a ZigBee and Z-wave coördinator a Pi makes much more sense. Electricity is expensive, yo
Ubuntu will not support RISC-V CPUs without RVA23 support going further, so this is stuck on 24.04 forever. There is no official Debian image either, and the Ubuntu version uses fixed kernel 6.6.47 with no further updates (it is not even installed from a repo).
I own one since a couple months ago but I already regret the purchase.
>I own one since a couple months ago but I already regret the purchase.
This is me with every single SBC
I owned every generation of Raspberry Pi and I still use the 5, 4, 3, and even a 2 Model B. Their software/ecosystem support is unmatched by any other SBC line.
My pair of RockPro64s refuse to go to the drawer. Every time they are about to retire something comes up that is the perfect job for those machines.
One is currently running Scripted to integrate my unifi cameras to HomeKit. Zero fans.
I've equipped the other one with 2 beefy 22TB HDDs, Tailscale and Minio. I'll send to my parent's house to act as remote backups.
It's actually worse than that, even getting updates kills the image. I did manage to work around it. but overall I don't consider it a platform for anything I would depend on. It's fun for experimentation, that is about it
My notes from that little project are here -
https://www.hydrogen18.com/blog/orange-pi-rv2-first-look-ker...
> this is stuck on 24.04 forever
But as a basis for IoT projects that's perfectly fine. They're meant to be install and forget.
only if you find it acceptable to have your IoT projects participate in botnets down the line, I guess.
Distributed computing is a perfect fit for IoT devices which mostly sit idle. Put those cycles to use!
Kinda tricky. Last time this came up, the consensus was that approximately nothing commercially available supported RVA23 at the time.
No. EXACTLY nothing. The RVA23 spec was ratified 11 months ago, today. There is zero reason to expect commercially available hardware in that time.
However there should be next year, and quite likely a couple of different choices by the time Ubuntu 26.04 LTS is officially released.
Waiting until 28.04 for RVA23-optimised code would be far too late.
Ubuntu is the worst of the Debian based Linux distributions. Give MX Linux a try. It's rarely evangelized, unlike other Linux distributions, but it has a surprisingly high user base, and for good reason. Everything it has added to Debian enhances it, and it doesn't try to promote any services, like Ubuntu does with Canonical's services.
Regardless of the Linux distribution, it's still way to early to get good support for RISC-V. Current products are good enough for general development, but RVA23 was ratified less than a year ago, and I don't know of any designs started after the ratification that have hit the market yet. Once that's the case, it should be easier to get universal support, but until then, every RISC-V SBC is a one-off development board.
Building Debian for this board is pretty easy nowadays. You can also find prebuilt images.
Want to share any links, particularly about building one's own Debian image? I'm especially curious about the bootloader.
https://romanrm.net/rv-debian
A good example. There is also a link to builscripts.
Funny how software matters I bought a specific eink tablet and the lag and choppiness in drawing made me immediately sell it for something more well known.
Sucked too as I lost $300 but hopefully helps open source community anyway
I had a go at it a few months ago: https://taoofmac.com/space/reviews/2025/05/12/2230
Not bad at all, but the OpenWRT image still didn’t have Wi-Fi support a week or so ago, so I don’t know how good software support is going to be in the long run.
The big ecosystem of SBCs confuses me a bit. Who is buying these?
The work required to build an actual secure, maintainable product on top of an SBC is so big that you'd surely never use one of these. The hard work is all in software. You need a supplier with product lifetime guarantees and a known SoC manufacturer.
If you're a hobbyist, unless you really don't value your time you'd be much better served buying an x86 PC or a Raspberry Pi for whatever project you've got. Any money saved buying one of these would be completely negated by the extra time taken to maintain it.
So who's the target market? Are there products out there built on these? Or are they mostly just shipped straight into desk drawers? How many of these do they actually ship?
> The hard work is all in software. You need a supplier with product lifetime guarantees and a known SoC manufacturer.
We need easily accessible documentation on the hardware so anyone can maintain the software. The problem I see is most SoC makers either have poor or no publicly available documentation. Their poor excuse for "open source" are undocumented Linux drivers for their black box hardware and a near obsolete kernel that they may or may not update.
These chips are made by big companies for big companies and not hobby people. They expect that you're going to buy in volume, sign NDAs and possibly license IP from them. The chip goes into a product that is likely never going to be updated and thrown out in 5 years.
You don’t need that many customers to design and ship an SBC. The hobbyist market is enough to sustain these companies. They can sell accessories to increase their margins. Their contract manufacturer can incrementally build batches of the boards.
Many of these boards serve as development boards for the parts they include. If you want a dev board to try this part or you need a cheap RISC-V system to test RISC-V things on, buying one of these is an easy choice.
> You don’t need that many customers to design and ship an SBC.
Right. An SoC takes five years and tens of millions of dollars.
An SBC using that SoC can be designed and made in three months for the cost of a small team's salary -- possibly even just one person e.g. see Paul Stoffregen of Teensy fame, or Jonathan Oxer at Freetronics. I'll bet the teams at Sipeed, Banana Pi, Orange Pi etc are pretty small too.
> Who is buying these?
All people who want to optimize software for the RISC-V Vector extension, as this is the most afordable SBC which supports it.
There are many products that are only expected to sell in low volumes and high margins. Also, prototyping, proof of concepts, one-offs, etc. Those are all areas that I've seen them used.
The thing cost as much as two cocktails at my local bar.
I might just buy one to play with it for two or three hours and then throw it in a closet. Cheaper than going to a movie and more entertaining!
Most of the SBCs are supported by fairly standard Linux distributions: Debian/Armbian or Fedora; you just boot from the approriate image on an SD card. Some SBCs have eMMC storage and/or a M2 connector, so you can either keep running off the SD card or transfer the image to that storage.
The value proposition of using SBCs is in their embedded connectivity; in addition to standard USB/network/HDMI ports they tend to have built-in connectors for:
* MIPI input, for video cameras
* MIPI output for LCD panels
* i2c and SPI for weird peripherals (accelerometers, temp sensors, etc)
* i2s for sound
* GPIO and timers/PWM for custom peripherals like motors/servos, contacts, programmable LEDs etc.
I mostly use them for things specifically designed to not need software maintenance. I use them in test fixtures and other embedded systems. Anything that is connected to a network isn't connected to the internet, and regardless of a network connection, I always set the filesystem as immutable and create backups of the SD card, so if anything goes modified, restarting usually fixes it, but if not re-imaging the SD card for sure will.
> So who's the target market? Are there products out there built on these? Or are they mostly just shipped straight into desk drawers? How many of these do they actually ship?
Hobbyists, mainly.
I could see this being a hit with people who want to work with RISC-V, which still needs a lot of low-level stuff built out for its ecosystem. You don't need it to be a screamer, just for it to run predictably.
Remember, the Raspberry Pi was mainly a hobbyist curiosity when it came out. There's definitely a market.
>Who is buying these?
If only you knew the ecosystem in China and Taiwan. Even a fraction of the Chinese domestic market alone is enough.
Maybe for hobbyists it's more of practicing how to play around with embedded systems? Over a decade ago I sorta got burned by Odroid dropping support for one of their early units despite being much faster than the 2Bs that were selling at the time (plus the annoying cost of emmcs or whatever), so I absolutely agree with your point.
> The big ecosystem of SBCs confuses me a bit. Who is buying these?
They make decent home servers, unless you need fast storage.
So actually one of the thing that attracted me to the RV2 is the PCI Express slots. I was able to add a SATA controller and an NVMe drive to the same unit.
Absolutely smoked by rpi5, often by rpi4. To make matters worse, a radically unsupported core with no mainline support. https://www.phoronix.com/review/orange-pi-rv2-benchmarks/2
I can accept the performance issues for now- it's an emerging platform, and hasn't had the huge amounts of resources poured into it like ARM. There's a chicken-and-egg problem here, since those resources will be limited since people aren't buying RISC-V equipment, which limits the incentive to commit resources, etc.
But what I cannot accept is the truly awful documentation and software support from the vendors. This is where Raspberry Pi shines, and is IMO one of the most significant factors in its success. I'm excited that RPi is dipping their toes into the RISC-V pool with the Hazard3 cores in the RP2350- perhaps they will be able to release a Raspberry Pi RISC-V edition board some day.
But I'm hesitant to buy one of the current RV SBCs, so I guess I'm part of the problem.
I'm also surprised that there aren't any startups producing small, simple CPUs and SOCs outside of China (as least, none that I'm aware of). Is there no investment available in India, N. America, Japan, Europe, Israel (* not bringing the current situation into this, just noting they have chip fabs)? Fabricating chips is not cheap, but the first ones don't need to be the top-of-the-line TSMC 3nm process.
Poor vendor software support is also expected, because it's too early into development of the RISC-V architecture for software to have universal support on different hardware variations.
What that leaves is a need for good documentation, but Orange Pi does have that: http://www.orangepi.org/orangepiwiki/index.php/Orange_Pi_RV2
If you're not working on RISC-V-specific development, than it isn't a product for you, but with step-by-step guides for building uboot and the linux kernel, as well as running various LLMs on the TPU, it is very well supported for the current RISC-V audience.
Don’t think people buy riscv for their performance competitiveness at this stage
People don't really buy RISCV at all at this point, there's noting less compatible you could get if you tried.
isn't there box64 that can run x64 applications?
And I think a language like golang can be a really really nice fit given how it can be compiled really fast towards risc-v as well
Maybe java also runs in risc-v I am not sure, surely people are working on java support I suppose.
People buy risc-v to support an open standard and to not worry about licensing fees.
Isro (india's nasa basically) uses some risc-v chips to not license arm chips etc. because of either better national security (to have less arm influence) or because they don't want licensing fees given how rudiculously price efficient isro is.
It wouldn't shock me if performance doubles over the next couple years. This chip is one of the first supports the RiscV Vector extension, so compilers have a lot of years of catch up to vectorize effectively.
It already did between minor geekbench versions even though not all benchmarks are accelerated yet: https://browser.geekbench.com/v6/cpu/compare/10993563?baseli...
Yeah I would wait until RISC-V SBCs have RVA23 and ideally ACPI & Unified Discovery support. Otherwise you're in abandoned software territory.
I'm not sure if any RISC-V boards support ACPI, and Unified Discovery is not even specified yet, so ... yeah probably don't buy this. We'll get there.
https://github.com/riscv/configuration-structure
Don't suppose there's actually documentation for the CPU anywhere? (I mean more than a tiny "datasheet" with a very high level overview and/or a pile of random Linux/uboot patches)
Raspberry wouldn't be the market leader if Orange, Rock and the rest could write documentation and support to save their life.
Spacemit's website is a pain to use, but the processor appears to work almost exactly like the K1, so:
- go to https://developer.spacemit.com/
- click on documentation
- click on Keystone
- click on K1
https://pan.baidu.com/s/1qbLD7j1nw3fRlNq09U8l9g?pwd=511c (from orangepi.cn) links to a Ky X1 chip manual.pdf
RISC-V going forward, one of the only beacons of hope in the silicon world.
I need one of such devices for my self-hosted services. And it will be time to port from C to assembly, really, because we have finally a CPU ISA which is 'sweet spot' balanced, standard, global, pushed forward with significant resources and without IP locks anywhere. No more developer/vendor lock-in via "the only compiler able to generate correct machine code", extremely hard to do planned obsolescence, etc, we need mainstream adoption NOW :)
The main blocker: how do I buy such device with a noscript/basic (x)html browser? And no way I use a credit card on a web site: would require well identified bank swift account, or wallet codes bought from local and physical currency terminals. I don't know of any local retailers I can buy such device from. Yep, the "web geniuses" at amazon (which supports wallet code) broke noscript/basic (x)html support a few years ago.
Noscript? No credit cards? Why these complications?
Run a VM or a container with a full-blown browser, then throw it away.
Get a merchant-locked, ephemeral credit card at privacy.com or equivalent, or buy a preloaded anonymous card.
Problem solved.
Come on... how it is possible to miss the point that much?
I use Privacy.com to get temporary credit cards, to avoid the hassle of compromised credit card numbers, although it doesn't stop various governments from tracking my transactions.
I've bought RISC-V SBCs from both Pine64.com and Arace.tech, and neither required I make an account. Arace.tech does require JavaScript to checkout, but Pine64.com does appear to work without it, although I didn't complete a transaction without it. Pine64.com also accepts USDC payments, but no other cryptocurrency.
What? This is not my point.
That said, 'privacy.com' does not deserve anymore trust than all the shaddy stuff already out there...
Don't buy an Orange Lie.
I used to occasionally buy a single board computer ~10 to ~5 years ago.
Then I waited until SBC's reached the 32GB RAM level. The first such affordable and performant board was an Orange Pi 5+ or 5B (I should double check, may add a comment later).
I believe I wanted the Orange Pi 5+ but it was sold out, so I ordered and paid for the Orange Pi 5B (the 2 ethernet port variant) which was a bit more expensive but was still for sale. Both had 32GB RAM, my main requirement. There were multiple "flavors", with power adapters or with case or with memory card or eMMC etc. I chose the memory/eMMC version.
I sent a message to ask them to give me heads-up when they are about to ship mine, and then I was patient.
Then I waited, and waited, and waited.
Too patient, after a long while I start looking up on forums if other people are also waiting. I discover I am not the only one. So I take up contact again and ask when the board will be shipped.
They inform me the SBC is no longer manufactured, and offer me inflation-devalued currency.
I check which single board computers they still sell, and indeed they no longer sell the 5B variant, but now the 5+ is back in stock.
I ask them if they can just ship me the 5+ instead of the 5B. They refuse.
OK, I ask them how much I can pay extra so they ship me the 5+ instead of their unilaterally discontinued 5B.
They refuse.
A few months pass by. I ask again if they intend to ship the 5B as agreed, or the 5+ as a substitute.
So here comes the orange Lie:
They claim they shipped it, and provide a DHL link.
I first name (same as my father's) is German, even though I live in Belgium.
Their DHL link, is a shipment to somewhere in Germany, with a weight far below the weight of a single board computer, and which was delivered just minutes before their sending this message claiming shipped delivery.
I confront them that the weight of the SBC is advertised on their own site, and their DHL delivery link lists a value far below it, that I live in Belgium and the shipping address was in Belgium (I have 0 links to Germany), and that the timing of the delivery and their response message is so close it suggests people at customer support (presumably without arbitrary access to deliveries outside of the case) asked colleagues to let them know if a case pops up with a German delivery, so they can manually copy and paste bluff delivery of product X to customer Y as if it was my order to me into the message.
I confront them and ask them to answer a numbered list of questions.
They refuse to answer the questions (they can't without incriminating themselves), instead they offer me my money back.
Don't buy into the orange lie.
If you work for Orange Pie, feel free to msg me with a way to contact you, if you positively resolve my case I will remove this message.
I've got dozens of various OrangePi models and I use them for a wide range of projects like drones, sdr receivers, home automation nodes, etc.
I'm sorry you have this experience but it's definitely not typical.
What was the website you bought it from?
amazon.nl
EDIT: adding the seller on Amazon.nl was BestMii, and they were linked as the sellers from Orange Pi website.
I was wondering what was going on, because Orange Pi only sells to distributors, not direct to customers.
Amazon only has a single listing for any given product, and all distributors, whether authorized or not, are under the same listing. When there's more than one distributor, you can choose which one to order from, but the buy now button follows a default.
If it isn't too late, I would highly recommend disputing through Amazon, as disputes are the only way Amazon addresses problem distributors. Many know how to game the system, so if you communicate directly with them, they will try to delay you until after the dispute period has ended.
It was the only amazon distributor as far as I know.
they were directly linked from the Orange Pi 5 website.
Its their responsibility they direct their customers to dubious distribution systems (IF THAT IS THE CASE).
We should never reward orders or preorders in any fast devaluating area (like electronics) to be reneged after inflation: this would allow manufacturers to post-select a financial path: if less inflation than expected fulfill order, if more inflation than expected renege the deal.
Amazon can be rather dubious, but there's not a lot of competition to chose from. Orange Pi is far too small of a company to set up their own international payment and delivery system. I have used Ali Express, and they've been pretty good at protecting me from attempted fraud.
Their decision is their decision.
I simply report what happened to me.
Orange Pi should fulfill my order, which I paid for, and still patiently await to see the board. If they feel it is the fault of Amazon (the platform of their choice) or the distributor of their choice, they can complain there and ask them for remuneration for reputational damage and cost of unfulfilled orders.
All I know, is that I paid via the Orange Pi endorsed channel and never received the board.
I have the right to warn others.
Hiding behind front company distributors to pull of inflation postselection tricks or whatever malice or incompetence it turns out to be should not be rewarded.
Orange Pi website shows for me only US Amazon Store. You should get your funds back from Amazon and order on Aliexpress.
Thats probably because of your IP address?
Also strangely, I am unable to load the https version of www.orangepi.org even though it was available in the past:
https://www.amazon.com/dp/B0CKVPWJVL?ref=myi_title_dp
I also just discovered the dropshipping policy (and necessary requirements and forbidden practices).
Either they had an agreement with Orange Pi or they didn't. Either way this gives leverage to ensure my order gets fulfilled.
If customers in EU or the west should order on Aliexpress, isn't it misleading of Orange Lie to provide Amazon links as well?
More info here from a few months back: https://boilingsteam.com/orange-pi-rv2-new-risc-v-board-revi...
Are there any SBC with memory slot so that i can plug in 32GiB or more of RAM?
Yes, lots of Intel and AMD based SBCs have SO-DIMM slots, but you'll have to accept the 3.5" format.
Regarding RISC-V SBCs, there was serious consideration to release the Milk-V Oasis with SG2380 and LPCAMM2. But this didn't work out as the SG2380 was held up by geopolitical issues.
Still so mad about that
the only one i've seen in a comparably small form factor is the UDOO Bolt https://www.udoo.org/discover-the-udoo-bolt/
but it's an x86 system and pricey, and borderline not an SBC
memory is soldered in most of these smaller systems including the Orange Pi and it's the main price differentiator
RAM is by far the most expensive thing in those systems past the very lowest specs, this one goes from 2GiB to 8GiB AFAICS (amazon page: https://www.amazon.com/Orange-Pi-RV2-Development-Ubuntu24-04... )
the Orange Pi 5 plus comes with 16GiB but the price jumps to US$160+ or 32GiB for US$ 270+
It baffles me that virtually all ARM and RISC-V boards have soldered RAM.
Orange pi 5 plus has 32gb versions. Pricey though
No memory slot though
Not really. Most in the ARM space, at least, are soldered on or you need to switch out the entire compute module. Intel ones (not mini-PCs, but industrial gear and things like the LattePanda) also tend to have soldered RAM.
[flagged]
I keep seeing suggestions that theres no software support for Orange Pi.
Whats the go there? Is there no distro like Raspbian supporting it?
Xunlong (Orange Pi) operates similarly to Pine64, throw hardware at the community and then let the community figure out the software part.
They provide official OS images at release but don't care much afterwards.
Pine64's claim to fame is that they have really good documentation, surpassing even the Raspberry Pi. That's what creates a good community around their products.
Most customers couldn't port software to an SBC, but for the ones who can, having all of the documentation makes it trivially easy, and having any that share their work makes it available to the whole community.
Check out the Oz64 documentation, (https://pine64.org/documentation/Oz64/_full/) which includes schematics and links to the datasheets for all of the major components. Raspberry Pi products, on the other hand, need some pretty intense reverse engineering to develop on: https://www.jeffgeerling.com/blog/2025/reverse-engineering-a....
For my use case, and most Pine64 customers, I'd rather have the hardware documentation than off-the-shelf official support for a software stack. Raspberry Pi has an entirely different user base.
Orange Pi fills a similar niche, and really anyone releasing RISC-V SBCs at this point does too, as it's too early into the development of the architecture for microprocessor-level products.
There's a RV64 port of Debian and the RV2 and R2S are on the list of compatible hardware. No guarantee it'll be easy getting it loaded, it was like pulling teeth to get it on the SiFive U74 board, but that was 7 years ago, so it's GOT to be better by now.
There are plenty of Orange Pi boards with Armbian or unofficial Ubuntu support, but they’re ARM based.
I caught Orange Pi doing misleading advertising earlier, trying to make people hear that the CPU has an NPU. Does this actually have any AI hardware or not?
The Ky X1 has AI acceleration on 4 of its 8 cores, creating 2 TOPS performance. There's a guide for running LLMs here: http://www.orangepi.org/orangepiwiki/index.php/Orange_Pi_RV2...
That looks quite similar to what I read before. For me those performance figures don't necessarily suggest it being any faster than just running on CPU so I'm still suspicious.
Might research this a bit more. Of course even if there's no perf improvement but perf/watt is better, that be valid too. But sounds very marketing speaky
It's not faster than running on an x86 CPU, but it's faster than an unaccelerated RISC-V CPU.
The entire platform is more for development of RISC-V software, especially at the operating system level, than it is an end-user product. There's still a lot of development needed to get RISC-V perf/watt, even with acceleration, to match ARM and x86 systems. I would expect at least a couple of years until then. RISC-V microcontrollers are already getting there, competing well with M0, but that's an entirely different market.
What are the specs in term of CPU speed?
So far, all the RISC-V SBC's I've tried were woefully under-powered compared to a comparably-priced Raspi.
Anyone have openssl benchmarks? I.e. run "openssl speed" and post the output. Thanks!
I don't know if it uses the vector instructions.
https://gist.github.com/faried/6955a992c6d68362fd1e07a1cd575...
Thanks! I'll compare it to my raspberry pi 4 when I get a chance.
2 TOPS is not a lot for AI projects.
They get this number by simply adding up all CPUs' processing speed. There is no NPU or similar: https://medium.com/@zlodeibaal/orange-pi-rv2-ai-board-scam-7...
Four of the eight cores have matrix accelerators. If you are using the full 2 TOPS of accelerated performance, you have four more otherwise symmetric cores free to run regular workloads, otherwise you have eight. I'd rather have this setup than 4 regular cores and a separate asymmetric accelerator that provides 2 TOPS to matrix/tensor calculations but otherwise sits idle.
thats even worse. 0 TOPS!
The whole point is offloading load from CPU and doing something in parallel while NPU does its thing.
It's a lot for 40 USD. And not every AI project is a Language Model project.
no its not. Just use a smaller rockchip, 1-2 actual TOPS in ~$20 as a board or $10 a piece. i am using one for my project.
Hopefully Adafruit or someone could get these[1] out of "Contact Us" jail. 16GB 30TOPS BF16 in M.2 2280 at $369. PCIe 2x8 low profile "Duo" configuration available at $799. Supposedly. I believe the theoretical performance is Strix territory if these could be clustered, but only if they mass manufacture these.
1: https://www.edgecortix.com/en/hardware
Heh, interesting you call this "Contact Us Jail". I met sales and engineering staff from EdgeCortix at DSEI Japan 2025, and was very enthusiastic to pick up some of their products as alternatives to the Hailo-8L RPi cards.
So far....nothing. Not even an ETA on when they will be shipping hardware.
DSEI! I wish I had some official reasons that I'd be obligated to go and have fun in events like that :p
My fear is that they could be in endemic Japanese death spiral disease of hand-wavingly waiting for massive orders while not doing much to secure one.
There are so many cool stuffs and perfect plans from Japanese large companies and government agencies that are pushed forward up to one step before volume production or commercial rollout and then stalled until funding dries up and everyone moves on. Which, if I were to explain by speculating, might be due to people prioritizing own job security over corporate revenues, due to low psychological safety and perceived invincibility of large organizations, as one among many consequences of employment ice age phenomenon.
But whatever the mechanism is, it'll be sad if this one follows that path. Hopefully they make and ship them in substantial numbers. That should be their goal anyway.
meh. i write a lot of PROLOG, so 2 TOPS is A LOT of horsepower.
for tasks like face recognition and object detection, would this type of hardware have good performance in real world cases? or, what is the standard hardware that devs use for tasks like that?
I don't like RISC-V unless it has a good GPU
Theoretically the vector extensions in a RISC-V could be scaled considerably, it seems to me, especially when combined with the extension proposed here: https://github.com/spacemit-com/riscv-ime-extension-spec
That said, the actual processor cores in this SBC seem to max out at 256 bit registers, which does not seem to be a lot.
GPUs are [effectively] irrelevant for many use cases (IoT, embedded, most servers, etc)
On Raspberry Pi, the GPU is the only thing that makes a responsive GUI or web-browser feasible, and is the primary reason most people use the HDMI LCD screens for games etc. It also took a large effort to bring up a v4l2 kernel driver for the camera modules etc.
For example, on the CPU one may pin all cores to stream a USB camera or software decode h264. With the SoC GPU decoding or streaming with the v4l2 interface might take up 30% on one core (mainly to handle the network traffic.)
The Raspberry Pi are not the fastest or "best" option (most focus on h264 or MJPEG hardware codecs), but the software/kernel ecosystem provides real value. Also, the foundation doesn't EOL their hardware often, or abandon software support after a single OS release.
A cheap RISC-V SBC is great, but ISA versions are generally so fractured (copied the worst ideas of ARM6)... few OS will likely waste resources targeting a platform that will have 5 variants a year, and proprietary drivers.
A Standard doesn't even need to be good, but must be consistent to succeed. =3
the title says "... AI projects". now, maybe our definitions are different, but you probably want some hardware acceleration.
Most likely comming in vector, matrix instructions or NPU like chipsets, not necessarly GPUs.
Low-power processors rarely have the AI accelerated instructions in the GPU, instead opting either for dedicated matrix/tensor cores, or as is used in the case, adding the acceleration instructions directly to the CPU core.
This results in a higher performance per Watt, but doesn't scale well to higher-power applications.
The chip (KY X1) comes with AI acceleration...
RISC-V designs typically have an Imagination Technologies GPU, some support for them is in recent versions of Linux and mesa.
they all suck. someone needs to make an open source gpu already, its been way too long.
We did back in 2007: https://en.wikipedia.org/wiki/Open_Graphics_Project
And there have been some others as well: https://en.wikipedia.org/wiki/Free_and_open-source_graphics_...
Recently https://www.furygpu.com/
Part of the problem is that every ASIC manufacturer (and indeed each fabrication process) has a different toolchain with a different set of primitives for circuit design. Yosys and other open tooling for FPGAs has helped a great deal in lowering the barrier to chip design and by association reuse of circuits. But every ASIC, at the moment, is tied to some vendor's PDK. Here's the one Google open sourced for Cypress Semi's SKY130 process node: https://github.com/google/skywater-pdk
It is at least theoretically possible to build a headless "GPU" from RISC-V processors that have the vector extension (RVV). RVV had been designed to be able to run programs compiled for the SIMT execution model that most GPUs use.
This Orange Pi RV2 has a small vector unit in each core, and could be used for at least prototyping the software until more powerful chips are available.
BTW. There have also been a couple hardware startups that have been working on commercial GPUs based on RISC-V's vector extension, with their own GPU-specific instruction set extensions for texture lookup and the like.
https://www.tomshardware.com/pc-components/gpus/startup-clai...
The 94% market dominance of CUDA GPUs will roast new competition for sure.
RISC-V has a fragmented ISA standard, and every version is a magical unicorn part (the worst facet of ARM6.)
A Standard doesn't need to be good, but must be consistent to succeed. =3
The vector instructions are only in four of the eight cores. There's also extra cache in those cores, but they are otherwise symmetrical to the other four.
It's probably a series of patent landmines...
Hardware patents are orthogonal to open source software. If a patent covers the hardware then someone who wants to manufacture the hardware needs to license the patent, but you were never going to get free-as-in-beer hardware anyway, and a hardware patent is independent of whether the hardware is fully documented or has firmware with published source code and a license that allows users to make changes to it.
Indeed, most IP used in silicon design are licensed, cost real money, and are under NDA.
I wouldn't say "never", but a clone is highly unlikely for another decade or so. =3
NDAs in the context of a patent license should legitimately be considered grounds for patent invalidation. Patent literally means "open to public inspection" and the entire premise is that you get a temporary monopoly in exchange for openly documenting how your invention works in the published patent. If there is something about the invention for an NDA to cover it means you left it out of the patent application, which is essentially an admission of wrongdoing.
A patent is an alternative to a trade secret. You can't eat your cake and still have it.
Patents do exist on popular IP cores, but in general standards compliant Verilog libraries are just vendor specific IP products. Thus, the copyright, NDA, and license agreements already keep the IP fairly locked down. For example, figuring out dram timing and DDR bus control yourself is nontrivial.
Some groups have attempted open IP cores, and made some progress:
https://opencores.org/
However, the effort involved in getting standards compliant ASIC built puts folks in a Fabless manufacturing sector. Most firms that survive, will choose to stay with a generic FPGA option to avoid custom silicon unless absolutely necessary.
Patents are often useless/vague in many places, but on occasion may prevent platform decay for a few years. One can be sure a unique/new design will not go to fab unless such protection is in place. =3
It seems like the fab companies like TSMC/Samsung/GF/Intel are missing an opportunity here. Commoditize your complement:
https://gwern.net/complement
Publish the software that does this for free so that more customers come to you instead of using FPGAs or just not making the attempt. Make it easier to design new chips so that more people do it and you get more customers.
When it comes to hardware, the OSS model stops making economic sense.
People have tried, but helping other competitors for free rapidly decays the market. =3
Rule #23: Don't compete to be at the bottom, as you just might actually win.
The point isn't to help your competitors, it's to help your customers. Qualcomm or Apple doesn't have the incentive to help their competitors design chips, but TSMC or Samsung does have the incentive to help Qualcomm's competitors design chips, because then there are more companies making chips and the fab gets more business and prevents any one of their customers from getting too much leverage over them.
Samsung actually had a technological lead in many areas, and it translated into real revenue. They are a business, and exist to provide utility to consumers in exchange for investor profit:
https://www.youtube.com/watch?v=KCWDzWG1BcI
Many incentives to push technology forwards are high-risk/expensive investments, and expecting the public/customers to willingly help pay that cost is naive:
https://www.youtube.com/watch?v=cru2bkqwSYk
I assure you academic funding does not cover such large costs, government grants are only a fraction of expected taxes in late stage Technology Readiness Levels, and competitor/cloner fractured markets erode fiscal returns needed to pay for the total incurred project cost.
Apple makes minimal utility products, but relies on intangible branding to maintain perceived value. Thus, only Apple could get away with selling zero chip designer handbags, and would still make absurd revenue (not a real product yet.) Steve Jobs observed very early, that selling raw motherboards was a low margin business. Which it was why the company shifted into consumer products.
For almost every other brand, consumers have shown they prefer the near material cost in opportunistic China/India factories, and thus simply ignore most firms products that include 10 years of R&D costs to pay back the investors.
There are many shelved technologies that will never see a Patent or the public markets. This is because the conditions are not ready for advanced products yet, and competitors irrationally nurture the lowest value volume market sectors. Thus, everyone gets a 15% value boost at regular intervals, and people remain excited about 3 decade old technology.
Qualcomm cellular chip product lines essentially lived off iPhone sales. Like any loyal dog, they will unlikely bite the hand that feeds them...
Rule #3: popularity is not an indication of utility.
Have a wonderful day =3
Attach your favourite GPU at the PCIe slot.
Without mainline support, which it won't have soon, due to the lack of RVA23, this is just another trash product.
What a disastrously bad name.
Meh. RISC-V performance is still pretty bad compared to other common market options: https://www.linuxlinks.com/benchmarking-the-orangepi-5-ultra...
No SATA :(
I bought the 32 GB emmc module for it, for the root filesystem. I have a 500 GB nvme drive for everything else. I believe an nvme-to-sata riser will work, but I don't have one to test with (plus you'd need to power the sata drive with something else).
What is your project where 2 x NVMe is not suitable for storage?
I'd like a decent RISC or ARM SBC for a NAS.
There's M.2 cards with PCIe to SATA bridges, which are great for a high-capacity low-speed NAS setups, especially with mechanical hard drives. Be sure to research the chipset before buying one, as some are notorious for poor driver support.
If this is a practical application, I would recommend sticking with ARM for this generation, as RISC-V is still in early development. In a year or two we should start getting more standardized RISC-V processor implementations, with broader support.
Ah, cool. I'd like the same, but with ECC RAM, so the options are even fewer.
Have you looked into what someone suggested here, splitting the NVMe to some SATA ports? It didn't sound like a horrible solution, but I have no experience.
[dead]
$40 is too expensive
It's pretty much the minimal cost for the form factor and accessories. It would be cool if they could make a $20 version with a lot less connectivity in a smaller form factor.
Check out the the Bouffalo Labs BL808 or Sophon SG2000 series. They are a system-in-package products that contain a RISC-V CPU, RAM and flash memory all in a few mm square package, for a few dollars. Pine64 and Milk-V both have breakout boards for under $10.
They only have RAM and flash in the hundreds of megabytes, not gigabytes, but they can run full Linux and are as capable as many household Wi-Fi routers. They include USB, Bluetooth, Wi-Fi, and wired Ethernet ports, and even have MIPI CSI and DSI interfaces for cameras and displays.
The SG2000 is pretty interesting. I do wish they had gone with a more homogeneous architecture (4 identical cores like the pi 2 zero), and it's RVV 0.7, but the next generation of this could be really interesting.
Depends what you're using it for. A lot of people tend to buy pi-likes as servers which is absolutely bonkers. If you time eBay right, $50 would get you a fairly powerful intel NUC with much more performance and peripherals
I don't think it's bonkers. For running a true home server sure, there's more powerful things out there. But for hosting something like a ZigBee and Z-wave coördinator a Pi makes much more sense. Electricity is expensive, yo