RISC-V has been in the “promising” phase for a long time now, especially for general purpose computing, never really breaking through into the mainstream in any measurable way. While I think that breakthrough is still relatively far away, we now do have newer RISC-V SoCs on the market supporting the RVA23 baseline RISC-V profile. One of them is the SpacemiT Key Stone KЗ, which promises to deliver a massive performance increase over previous RISC-V offerings. It’s exactly this chip that’s finding its way into complete, turnkey mini PC solutions, like this one from a company called Firefly.
The base model comes with 8GB of LDDPR5 RAM and 128GB of storage, at a price of about €300 or so (there’s also a 32GB/128GB model at well over €600). This is the first time I’m looking at a complete RISC-V solution where I feel like it might actually make for a good moment to jump in for us enthusiasts. No, the performance won’t rival anything Intel or AMD has to offer, but it seems capable enough for a lot of day-to-day tasks, and I’m curious to see just how far along the Linux world is when it comes to RISC-V support.
It’s not part of our current set of fundraiser incentives, but if you’d like to see this RISC-V mini PC reviewed here on OSNews, you can always donate and add a note that you specifically want to see such a review (so I can gauge interest not just from our few commenters, but also from the more than 99% of our readers who only lurk). As always, you can donate through Ko-Fi, or, if you’re European, via a SEPA direct bank transfer (Name: Thom Holwerda – IBAN: SE08 8000 0820 1684 4657 8414 – BIC: SWEDSESS).
I always have the same problem with this kind of stuff: ARM and RISC-V PC-level-boards (I am not talking about RPi and other SBCs) need to be either more powerful (at the same price point) or way cheaper than Intel/AMD options to be actually useful. What is the point of getting a slower RISC-V PC/mini-PC when you get an Intel or AMD one with less issues and more performance, at the same or at a lower price (not even considering the tons of used hardware available on eBay and such) ? Unless, of course, you like to tinker and experiment with different architectures, or have a niche case in mind, and that’s of course fine.
I obviously understand these platforms need time to grow and become popular, but they have been available for some time now, and the progress has been abysmally slow. I wanted to buy an Orion O6, but the price, though not absolutely bad, makes no sense when looking at the numbers.
This situation is classic. For how many years has ARM been around and only now it is becoming mainstream for multiple use cases? 30+?
And ARM has had the luxury (and developer mindspace) of being the defacto architecture for mobile devices since the 90s!
RISC-V doesn’t have this luxury. It will require a lot of love and effort until it can become a viable self-sustaining alternative. And, in my opinion, it’s best fighting chance is to follow ARM’s playbook: find a niche and take the leap from there.
Death can be fast, too. PowerPC had a lot of love from Apple and it basically vanished without that niche. It never became mainstream and never became self-sustaining. x86/amd64 has never had a mobile presence (excluding laptop) and could die fast once ARM reaches critical mass, too.
For me, diversity is essential. Of course, no one is going to develop for 5+ architectures, but a monoculture is never good. I am a hobbist but, whenever I write, I always test hppa, ppc, arm, etc.. It keeps me honest!
ARM has the corporate money behind it, so it can get the latest patented tech & smallest chip process nodes. RISC-V will simply never catch up to the latest commercial chips in performance.
dark2,
Even ARM got pigeonholed into low power applications and despite ARM’s advantages it took tons of apple money to make it technologically competitive with x86. You’re right RISC-V would require similar kinds of enterprise money. II don’t think it’s impossible though. Nvidia’s embeds RISC-V cores in their GPUs and nvidia are cutting edge. Alas these tend to be locked down and not user accessible. RISC-V could be desirable in massive server roll outs by amazon or other tech giants with enterprise money to make them good. However once again data center hardware isn’t accessible to normal consumers. Somehow we’d have to solve the chicken and egg problem around ordinary consumer applications and it’s not a given that a company with the needed resources would be interested.
> ARM has the corporate money behind it
SiFive just raised $400 million dollars. Tenstorrent has raised over a billion. Qualcomm paid $300 million for Ventana. Meta just bought Rivos for $2 billion. Alibaba is a RISC-V shop. Huawei is a RISC-V shop.
Just what kind of corporate money are we looking for?
> RISC-V will simply never catch up
Most of the money is in the mid-market. So it will be a few years before RISC-V challenges the very top of the market. But they certainly will do that. M1 level performance this year. Maybe tied for first 5 years from now..
As you pointed out, if something offers significantly worse performance at a similar, or even higher, cost while also introducing more issues to deal with, then what is the point?
RISC-V currently has a strong value proposition in embedded systems and in contexts where ISA licensing costs are a real constraint, such as startups, academia, or highly specialized applications. But that does not automatically translate into a meaningful value proposition for end users of SBCs, for example. And so far, it clearly has not.
For the foreseeable future, I suspect consumer-level hardware will remain largely an x86/ARM world.
There is no consumer level benefit to RISC-V today.
As you point out, there is a monetary benefit at the low end or when designing cores for internal use due to the lack of licensing fees. But RISC-V is an ISA, not a processor, and so you are probably still licensing a core design off of somebody in many cases. So the monetary benefit from licensing is really only a big deal when you are shipping huge volumes of inexpensive stuff.
The real benefit of RISC-V is not licensing but control. This is a benefit that matters more to companies than consumers. You can commit to your own technology roadmap over the long-term without having to negotiate it with a supplier or worry about their strategic direction. And you can design your own extensions if you need them. That is why RISC-V is proving so popular in AI and automotive. It is, of course, the reason that countries like China are betting on RISC-V to reduce their exposure to US technology, trade policy, and sanctions.
The benefit for consumers will come from competition. There are only two x86-64 providers. There are only a few ARM players designing their own chips. ARM is a diverse ecosystem because there are many, many chip makers but almost all of them are licensing core designs from ARM the company. In the RISC-V world, there will be dozens companies doing what ARM is doing. There will be many, many competitors and, at the end user level, they will all be compatible with each other.
But that day has not yet come. Nobody is selling RISC-V desktops and laptops. Ok, they are but they are only for devs. The performance is not there yet. The price is not there yet. But it is all coming fast.
First micro-controllers, then specialty embedded, then servers, then phones, then laptops and desktops. That is my prediction anyway. Success in the micro-controller world has already been achieved, momentum in areas like medical equipment and automotive is building nicely, and real interest in servers is building.. RISC-V datacenters are already a reality in China. SBCs are really just a proving ground at this point but, with the K3, RISC-V has become “good enough” other than price. Price in the mid-range will fix itself when somebody takes the leap into phones. That may happen in China first as well. XiangShan has ported Android 16 to their “9 series processors” and I would be shocked if they do not show up in phones in the next couple of years. Once that ecosystem exists, somebody will do it in the west.
The “WIndows laptop” world is going to get some serious competition from ARM. Today we have ARM laptops like the Snapdragon X2 Elite stuff. We also have the Macbook Neo. Qualcomm just announced the Snapdragon C for the lower end with a release date that overlaps suspiciously with Google’s Googlebook launch dates. Throw in the new RTX Spark from NVIDIA for good measure. ARM is going to soften the desktop market and weaken the hold that Intel/AMD have on it. When RISC-V reaches the performance needed to play in that space (maybe around 5 years from now), the market may be ready to let them in. But, even before that, the Googlebooks of the world could instoduce RISC-V options without regular consumers even noticing, especially if it is already happening on an Android phone.
LeFantome,
I would say the main beneficiary is the Chinese market.
All those SoC manufacturers that were depending on ARM due to lack of x86 license would be worried about the recent trade wars. And that naturally brings them to RISC-V, which is royalty free and modern.
Even if the design is not as good (there are some legitimate questions, and some “why did they do that?” points)… brute force and spending resources usually win on the long run
(VHS vs Beta Max)
> I would say the main beneficiary is the Chinese market.
Agreed. Though they have LoongArch and even Zhaoxin too so they have other options. And we benefit from Chinese RISC-V too (we are talking about the K3 here after all).
The Chinese are not the only ones interested in “sovereign” tech. RISC-V is getting a lot of interest in India, Europe, and elsewhere. Tenstorrent was Canadian (much of the team is still in Canada).
In the USA, Qualcomm and NVIDIA have both shipped billions of RISC-V cores. The following is an interesting read:
https://en.eeworld.com.cn/mp/Icbank/a394573.jspx
LeFantome,
It is a long read, but yes, nvidia has been known to include RISC-V for their on board control firmware.
And that is how they got “open source” drivers for Linux. The modern GPU is just an isolated computing device reached by RDMA.
FWIW just about every modern SoCs from APPL, QCOM, NVDA, MTEK, have included deeply embedded RISC-V cores as system controller. Before that it was mostly ARM.
So there are lots of RISC-V cores out there already, just not in the “sexy” visible to the programmer roles its fans expect. It just doesn’t have much value proposition as the high performance scalar compute for consumer devices, just as what we have discussed.
@Xanady
> Before that it was mostly ARM
We may have to get used to saying that.
> It just doesn’t have much value proposition as the high performance scalar compute
Next up is the mid-range market. ARM has left themselves quite exposed there as they have not refreshed the product-line enough in recent years while they chased the top of the market. If you are Samsung looking to release a smart TV, you may notice that the SiFive P570 RISC-V core offers similar performance to the ARM A7x core that you were considering, except the SiFive P570 has much faster AI performance. Perfect for your smart TV. I guess that is why Samsung added RISC-V support to Tizen. For the hyperscalers, the P870 offers high-end performance and the ability to add custom extensions for better power management, better AI, more efficient networking, or whatever you need.
The next 2 – 3 years will prove one of us right.
I have read that excited analysis about RISC-V taking over the world non stop for the past 10+ years from RISC-V fans.
You’re viewing things in terms of tiering, but that is not necessarily how it works.
Markets that are sensitive to software libraries are not going to move over to RISC-V, regardless of tiering. RISC-V has had zero penetration in those, and will continue to do so because there it does not have any value proposition.
RISC-V is showing up in areas that makes sense. And those are the ones already listed.
> I have read…about RISC-V taking over the world non stop for the past 10+ years
Fair enough. Not from me though. I have been predicting it for that long but knew it would take time. But you are hearing it from me now. I will note another that news site is running an article with this title today: “RISC-V CPU Performance Up 8x In Five Years”. What happens if that trend continues?
I heard the same complaints when making predictions about Wayland, that “people have been predicting that for 10+ years”. But you only saw me become vocal about Wayland about 18 months ago. Now we have articles about distros with 95% Wayland users and X11 support being removed.
I am not expecting amd64 or ARM64 to go away in that timeframe but I do expect RISC-V to enter most markets visibly enough that detractors will be more cautious.
> Markets that are sensitive to software libraries are not going to move over to RISC-V
Ok. Sure. But what markets are these?
I mentioned Samsung Tizen about. Those are are .NET based (very portable). Even the Tizen RISC-V demo system bragged that 1100 apps were already available. Android is similar as it also uses a VM. If Googlebooks will run Android on both amd64 and ARM, portability is something apps will already have to tackle. Adding RISC-V is not going to be an issue.
But you probably mean legacy amd64 desktops and servers. I agree with you there. However, ARM is hitting both of those hard and already having success. That success again reduces the “legacy” moat just in time for RISC-V to arrive as a competitor.
RISC-V has lots of advantages. Better power management and stronger AI at the moment even without invoking the obvious benefits of an open ISA. Alos, smaller code size. But the real question is what advantage does ARM have? Legacy? Legacy in desktops and servers? Are you sure?
Anything outside deeply embedded systems, IoT, research/academia, or startup experimentation is extremely dependent on entrenched SW ecosystems/libraries. That is where x86, ARM, and even some niche legacy architectures will remain relevant, and where RISC-V will have a much harder time breaking in.
That is the part RISC-V fans (or fans of any non-mainstream architecture) often fail to grasp. Architecture adoption is not driven by enthusiasm about random tech specs alone; it is driven by actual value proposition.
RISC-V has already found tremendous success where it actually makes sense: embedded systems, custom silicon, academic research, startups, and cases where licensing flexibility matters. But that success happened organically because the value proposition was real.
It does not automatically follow that RISC-V will jump into markets where the software ecosystem, performance, compatibility, and support story are already dominated by x86 and ARM. In those spaces, RISC-V does not offer a clear and compelling advantage, there is very little reason for users or vendors to switch and thus why they haven’t.
It is actually a very promising machine! One of my students is waiting for one to arrive. One of the key characteristics of this computer (or rather, of its CPU, the SpacemiT K3) is that it’s the first RISC-V chip to incorporate the Zvk* vectorial cryptographic extensions. We will probably be able to share some insight into this system soon!
> it might actually make for a good moment to jump in for us enthusiasts.
Indeed. It is also available in a laptop that we are not allowed to talk about on this site.
The K3 is still “slow” as a general purpose CPU. It is about 2010 level Macbook performance single core and about 2019 Macbook performance multi-core. So it is really for devs and fans at this point. It is “good enough” to use but not so good that most people would be excited to use it.
One interesting thing about the K3 is that it has 16 cores. 8 of them are processors and Linux runs on them by default (the X100 cores). There are also another 8 cores (the A100 cores) which have 1024 bit long vectors and are meant to act as an NPU. This gives the K3 about 60 TOPS of NPU performance. But the A100 cores are also RVA23 compliant which means that you can run regular application code on them as well. So the K3 is actually about 50% faster than what I said above if you use all 16 cores (eg. when building software).
RVA23 refers to a RISC-V “application” platform standard that mandates stuff like the extensions that must be available. It is massive for RISC-V as it provides about the same feature profile as x86-64v4 including vector, crypto, and virtualization extensions. It provides a stable target for Linux distributions and applications to build binaries for. It will be the standard RISC-V profile for several years. The K3 is the first, and so far only, shipping RVA23 processor.
> I think that breakthrough is still relatively far away
That is not my view. There are at least 3 different chips due later this year or next that are expected to have about Apple M1 levels of performance. Given the popularity of the Macbook Neo, I think that gets RISC-V into breakthrough territory. My own prediction is that the RISC-V high-end achieves performance parity with ARM by 2032 or so. Look into the Tenstorrent Ascalon, the SiFive P870, or the Alibaba C950 for example. SiFive claims that P870 silicon is already being tested by customers. Jim Keller at Tenstorrent said they will ship a dev version of Ascalon by Q3 this year. All three of these designs have been available to license for a while now (as are others) if you want to make chips yourself.
And that is the problem for ARM. There are dozens of companies emerging that mimic the ARM business model but use the RISC-V ISA instead. The success of any of them drives the RISC-V ecosystem forward for the rest. Collectively, they are projected to grow faster both in terms of shipments and revenues than ARM is. Unlike failed ISAs of the past, RISC-V is not fatally exposed to the failure of any one company. You cannot just buy your competition to kill it and putting a RISC-V supplier out of business does not put RISC-V out of business. It does not matter which RISC-V providers succeed. And some of them are effectively nation states.
MIPS, the first commercially available RISC CPU maker, now sells RISC-V cores instead of their own ISA.
RISC-V is not going anywhere. Alibaba is building data centers with it. DeepSeek v4 runs on Huawei RISC-V chips. NVIDIA has ported CUDA to it and claims to use RISC-V in “all” their products. Most Linux distros already have a RISC-V version, including RHEL. And you can get a RISC-V microcontroller for less than buck.
RVA23 marks the beginning of the true RISC-V era. The K3 is awesome but it is the very next generation that will truly catch people by surprise, including Thom it seems.
If you review a K3 based machine, be sure to check-out Felix86: https://felix86.com/
In that case I would be interested in how felix86 compares to qemu
Haiku is actually advancing quite well with it’s Risc-V port. I will share this article with the forum and see whether anyone is planning to get one.
> ..I’m curious to see just how far along the Linux world is when it comes to RISC-V support.
As a personal observation I cannot get my head round the concept of installing Linux on brand new computers, nevermind avant-garde architectures. Perhaps that is due to my thing of installing it on old intel Macs (for which installing any other alternative OS is quite hard due to the idiosycracies relative to PCs), a preference to avoid buying new as much as possible, and also me not being a power user content with yesterdays performance.
When Asahi linux was triumphantly announced a few weeks or days after the release of M1 macs, I was like “why the hurry, it’s probably ten years till MacOS is no longer supported”. 🙂
Squizzler,
I also make use of second hand hardware. However I personally am only interested in hardware that runs my OS of choice on day one. The fact that macos may still be supported isn’t compelling for me because I’m not interested in switching to Macos. The M3-M5 not being supported by linux today makes them non-starters. I also don’t like assuming that linux support will work in the future or assuming hardware acceleration will work.
Of course I get that some people don’t care as much about daily driving linux. For you it seems to be more of a fallback option, and that’s fine. However the fact that apple are blocking repairs/upgrades of things like SSDs is kind of irredeemable. IMHO practically every owner’s interest would be better served with replaceable storage. Apple’s decision not to was motivated purely by greed and I hate buying into that. I find it regressive that more manufacturers are copying these tactics, so I think it’s important to vote with our wallets. Alas, as is often the case, I find myself siding with a marginalized group who don’t carry much weight in the industry. :-/
I am personally moving away from Mac, but even before that was the case there was still the problem of what to do with the machines that would reach the end of support.
For users of alternative OS it is generally better to steer clear of Apple stuff. If your apple is still supported the answer is simple: sell it as there is a solid demand. If too old to be supported, then Linux steps up. As indeed with the machine I am running – a Mini 2011 running ElementaryOS, a real man’s distro 🙂
The world is awash with old computers that are quite adequate for most casual needs. I would much rather the Linux scene prioritized rock solid support for old hardware (even support for old intel macs has rough edges) than going for glory of supporting the newest things right out the gate. Indeed, although it is being developed well for RISC-V, Haiku still supports 32 bit architectures an BeOS compatibility.
Squizzler,
Yes, I understand that’s a problem. For me personally though I try to factor that into my purchasing decisions up front rather than hoping things just get solved later. I’ve been burnt by unsupported ARM hardware too many times. I couldn’t in good faith recommend buying hardware contingent on future compatibility work.
This has long been an advantage with x86 hardware thanks to relatively open standards. Often times ARM hardware checks all the boxes except for good mainline linux support, and I usually regret it down the line because I am stuck with unservicable hardware. “Fool me once, shame on you. Fool me a hundred times, shame on me.”.
So I place a lot of value on having strong support and FOSS code up front before clicking buy…. Even this doesn’t always work because sometimes none of the manufacturers are FOSS-friendly and proprietary is the only choice.
I can’t blame people for not sharing my values, but the economic truth is that when the masses don’t place openness and FOSS support at the top of their criteria in their initial purchasing decisions, it typically leads to manufacturers ignoring openness and FOSS, which harms my own choices. Ideally consumers could stand together and agree not to purchase anything proprietary….I know it’s wishful thinking though.
100% agree. Same of course with mobile phones, I am in the process of migrating from iPhone to Fairphone. Partly because they make the hardware open to reuse as a microcontroller etc after its service life as a mobile. Running Murena no less, because at this stage why would you ditch apple only to get captured by Google?