PS: for those still hesitating to tinker with RISC-V the workflow is becoming quite convenient already, to the point you can "just" boot and install Linux (as mentioned in the article) on it to get a headless server running in minutes.
What a sad world to live in, stochastic bullshit machines and exploding drones get more computers than I
It may still have a slight price advantage compared to proprietary ISAs like ARM, due to the latter's ISA licensing costs. But it remains to be seen how much of this advantage will be passed to us, the end users.
The license cost of the ISA is not that much per chip. It is not nearly enough to equal the extra cost per unit that is going to be added to any upcoming RISC-V chip due to the small production runs compared to x86-64 or ARM.
On the microcontroller side, we are seeing what you are saying. RISC-V has taken over the microcontroller world and chips are being produced in the billions. Scale, combined with the lack of licensing, means that you can get RISC-V microcontrollers for a couple of bucks.
The other thing is that there will still be license fees for RISC-V. You are not paying RISC-V International. But somebody designed the RISC-V chip and board you are buying and they will want to get paid. Instead of paying ARM, you will be paying SciFive, or SpaceMIT or Andes, or Tenstorrent, or UltraRISC (for the Titan board this story is about).
But compare this to ARM where your choices are licensing from ARM or purchasing hardware from Qualcomm or Apple. Or compare to x86-86 where there is only Intel and AMD to choose from. There will be dozens of RISC-V suppliers and competition will drive innovation up and prices down.
Of course, you can of course always design your own! Or use an Open Source design. No license fees required. But the chips you will actually want to use as a consumer will probably come with licensing fees.
But 3-5 years from now, things may be different. I hope and believe that RISC-V will be successful. As RISC-V chips with competitive performance appear, volumes will go up and we will see the same pattern we have seen in microcontrollers. Many of the niches that are currently filled by ARM will start to be filled by RISC-V. This will include the SBC space and the server space for sure. We may see tablets and phones. With luck, it will also start to fill the periphery of the laptop and desktop space. Prices will be higher at first and then come down.
RISC-V is inevitable. And the competition that will bring will pay dividends to all of us.
This is basically what I've been waiting on. Besides the mentioned Milk-V Titan, what are some other good boards people here tried out and could vouch for being good? Ideally European, but happy to receive any recommendations as long as you've actually tried it yourself :)
conda-forge/conda-forge > "RISC-V Support?" https://github.com/conda-forge/conda-forge.github.io/issues/...
Is that because the platform itself is very lite, or is just typical for a dev ITX board?
https://www.cnx-software.com/2026/01/12/milk-v-titan-a-329-o...
The closest thing is probably Orange Pi RV2, but it has an outdated SoC with no RVA23 support, meaning some Linux distros won't even run on it. Its performance is also much poorer than of the RPi5.
I mean a board with decent storage and better performance.
geekbench: https://browser.geekbench.com/v6/cpu/16145076
rvv-bench: https://camel-cdr.github.io/rvv-bench-results/spacemit_x100/...
There are also 8 additional SpacemiT-A100 cores with 1024-bit wide vectors, which are more like an additional accelerator for number crunshing.
The Milk-V Titan has slightly faster scalar performance, than the K3.
So the main difference between this Milk-V Titan and the upcoming SpacemiT K3 is that the latter has better vector performance?
It matters, as the ecosystem settled on RVA23 as the baseline for application processors.
But you are correct that once RVA23 chips begin to appear, everybody will move to it quite quickly.
RVA23 provides essentially the same feature-set as ARM64 or x86-64v4 including both virtualization and vector capabilities. In other words, RVA23 is the first RISC-V profile to match what modern applications and workflows require.
The good news is that I expect this to remain the minimum profile for quite a long time. Even once RVA30 and future profiles appear, there may not be much pressure for things like Linux distributions to drop support for RVA23. This is a lot like the modern x86-64 space where almost all Linux distributions work just fine on x86-64 v1 even though there are now v2, v3, and v4 available as well. You can run the latest edition of Arch Linux on hardware from 2005. It is hard to predict the future but it would not surprise me if Ubuntu 30.04 LTS ran just fine on RISC-V hardware released later this year.
But ya, anything before RVA23, like the RVA22 Titan we are discussing here, will be stuck forever on older distros or custom builds (like Ubuntu 25.04).
Seldom does an SBC vendor want to actually support their products. You get the distro they made at launch, that is it. They do no updates or support. They just want to sell an overpriced chipset with a fucked and unwieldy boot sequence.
Same thing with all the Android devices. Pick a version of Android that you like because that's what you'll have on it forever.
What is the difference in performance?
Titan hw docs: https://milkv.io/docs/titan/getting-started/hardware
To add a 2x20 pin (IDE ribbon cable) interface like a Pi: add a USB-to-2x20 pin board, use an RP2040/RP2350 (Pi Pico (uf2 bootloader) over serial over USB or Bluetooth or WiFi; https://news.ycombinator.com/item?id=38007967
For 99% users, the only real "benefit" RISC-V can bring to the table is the _false_ feeling that "I am different". Before you start to be excited about those a few cents risc-v MCUs - there are much cheaper MCUs, consider those risc-v MCUs are dead expensive.
Thanks for reading my honest opinions, please feel free to downvote.
I am a user, I like to tinker, I'm fairly confident there's more than 1% of people who care about these things. If you live in a country that is threatened by export embargos and the like it also makes a lot of sense to prioritize open.
in fact, such ISA is only going to fuel more closed ecosystems as it made hundreds of Chinese vendors to join the game for free, they all suddenly got the chance to build their totally closed platforms.
And if you are a country, nobody can kill your RISC-V ecosystem. Worst case, you have to design your own chips but at least all the software exists and is established. And Ooen Source cores exist and are getting better. They may not be bleeding edge but they could be good enough if push came to shove. The BOOM chip just got vector extensions.
There is (so far) nothing 'open' about RISC-V. and I wonder if there really ever was any desire for it, at this point.
This whole "Open ISA" crap appears to be a thin veneer to funnel quite large sums of investment into an otherwise completely proprietary and locked-down environment that could never harm the incumbents in any meaningful way - while still maintaining just enough of a pretense of open source, that the (regrettably myself included) shallow nerds and geeks could get smitten by it.
Where is the RTL? Where are the GDSII masks? Why am I unable to look at the branch predictor unit in the Github code viewer? Or (God forbid!) the USB/HDMI/GPU IP? I reject the notion that these are unreasonable questions.
I want my SoC to have a special register that has the git SHA ID of the exact snapshot of the repository that was used to cook the masks. that, now that - is Open Source. that is Open Computing. And nothing less!
I dont care about the piece of paper with instruction encodings - the least interesting part of any computer!
Wasn't that the whole point? We're more than a quarter of a century in and we're still begging SoC vendors for datasheets. Really incredibly embarassing and disappointing.
As you note correctly, the ISA is open, not this CPU (or board).
The important point is that using an open ISA allows you to create your own CPU that implements it. This CPU can then be open (i.e. you providing the RTL, etc.), if you so desire
I assume it will be much more difficult (or impossible?) to provide the RTL for a CPU with an AMD64 ISA, since that one has to be licensed. I wonder if you paying for the license allows you to share your implementation with the world. Even if it does, it's less likely that you will do so, given that you will have to pay for the licensing fee and make your money back
Since there is no license to pay for in case of RISC-V, it allows you to open up the design of your CPU without you having to pay for that privilege
So? You've been able to do that since...computers. Anyone can roll their own ISA any time they want. It's a low-effort project that someone with maybe a Masters student level of knowledge can do competently. When I was in school, we even had a class where you would cook up an (simple) ISA and implement it (2901 bit-slice processors); these days they use FPGAs.
So you got your own processor for your own ISA...that was slow, expensive (no economy of scale) and without a market. But very fun, and open source, at least. And if "create your own CPU that implements it" is what you want, go forth and conquer...everything you need is already there and has been for a long time.
But if your goal is "I want an open source ISA that I can produce that's price and/or performance competitive with the incumbents", well, that's a totally different ballgame.
And there are open source ISAs that have been around for decades (SPARC, POWER, SuperH). These are ISAs that already have big chunks of ecosystem already in place. The R&D around how to make them competitive already exists. Some, like LEON SPARC have even gone into something like production (and flown in space).
So, yes, an open source ISA affords the possibility that we can make processors based on our own ISAs on our own terms. It has even in extremely rare occasions produced a product. But the fact remains, the market hasn't cared in the slightest to invest what's required to turn that advantage into a real competitor to the incumbent processors.
Yes, you can create your own ISA. But to run what software?
If I create my own RISC-V implementation, I can install Ubuntu on it. Maybe even Steam.
See the difference?
And, the market has responded with a tidal wave of CPU contenders. Like in the rest of the world, not all of them target the highest end portion of the market. But some are choosing to play there. Have you checked-out Ascalon?
And why did Qualcomm pay all that money for Ventana recently? You do not expect them to release high-end RISC-V chips? I mean, they already ship many low-end ones.
with the majority players being Chinese vendors (those you can buy, not including those building RISC-V for their own in-house applications), RISC-V is far less open than ARM or x64.
expecting openness from Chinese vendors is like trying to hook up with some virgin bar girls in your favourite gogo bar in Bangkok.
High performance implementations, i.e. actual chips you can buy, are going to be proprietary and that's not going to change. Engineering hardware is expensive.
But there are different levels of proprietary. Having your entire software ecosystem impossible to lock-in means something. And competition tends to breed openness.
MIPS certainly did gain a lot of traction. It was a real force at one point and the world is awash in them. But of course MIPS (the company) is RISC-V now.
If there were an economically viable number of people who cared about those things (and it would need to be significantly more than 1%), we'd be running SPARC or POWER or maybe SuperH derived systems, all of which have open source, royalty free implementations.
For example, OpenSPARC is something like 20 years old, and covers SPARC v8 through t2. SPARC LEON is a decade older, and is under a GNU license, and has been to space.
And that doesn't consider going the Loongsoon route: take an existing ISA (e.g. MIPS), just use it, but carve off anything problematic (4 instructions covered by patents).
It's a pretty inescapable fact on the ground that in the 'processor hierarchy of needs', an open source license is of no consequence in the actual market.
you need to be totally autistic to believe that Chinese vendors are going to share anything meaningful with you. they don't hate you, they want their paying customers to be happy, but the brutal competitions in China doesn't allow them to be open in any sense. For products like RISC-V processors and MCUs, the moat is extremely low, being open leads to quick death. It is not about how much stuff they share with you as paying customer, it is about how much they are willing to share with their competitors when there are hundreds of companies trying everything to survive.
as a developer, you just need to ask yourself a dead simple question - how such risc-v platforms are going to be more open than raspberry pi.
They are pushing their RISC-V products into the Linux mainline before those products even ship.
Those autistic Chinese also contribute a rather surprising amount of Open Source RISC-V out of their academic world.
How is that feeling "false"? People running RISC-V systems are different, or at least they have different motivations than you.
this is like saying being homeless is a lifestyle choice. some people argue that with a passion, I just choose to be kind to those people.
Agreed. Boards like this are helpful for getting RISC-V to the next stage, where it could make sense for more users.
Like no UEFI, no PC architecture (every board is different), got to x86 complexity (a miriad of instruction sets and extensions) in just a couple of years, needs a special linux kernel to boot with support for newer versions not planed.
Yeah, great software, great tooling. /s
That said, this is the year.
The Tenstorrent Ascalon is supposed to be as fast as AMD Ryzen 5 (according to the guy who created that Architecture at AMD). It is aimed at servers initially but they say they will release their own silicon sometime in the first half of this year. Even if that is optimistic, sometime this year seems likely. They released the licensable IP last year.
https://m.youtube.com/watch?v=Y3rtN8TTGf4&pp=0gcJCTIBo7VqN5t...
Don't set your self up for dissapointment. Ascalon is supposed to match Zen5 performance per clock, but at 2.5GHz, so will still be at a minimum 2x slower.
Additionally, the announces Ascalon devboard is supposed to be on an older node and have an ever lower frequency due to that. (the 2.5GHz were on SF4X, the devboard may be on something like 12nm)