VisionFive 2 RISC-V SBC - A Raspberry Pi Killer?
Vložit
- čas přidán 20. 08. 2024
- The VisionFive 2 is a quad-core RISC-V single board computer with an integrated GPU and up to 8GB of RAM. The board's biggest competitor isn't another RISC-V board but in fact the Arm based Raspberry Pi. Can the VisionFive 2 compete? Let' s find out.
---
Portable HDMI monitor: • 5 Ways to Use a Portab...
Twitter: / garyexplains
Instagram: / garyexplains
#garyexplains
It has one major advantage over the Raspberry Pi... You can buy one.
Underrated comment right here :D
Well you can buy either RK3399 or even 3588 SBCs there as well... And both works better than this RISC-V ...
@@JeffGeerling hope theres a video of it coming out from you soon :P
I know I haven't exactly been your biggest fan before but credit where credit is due, Gary - this was your most unbiased RISC-V video so far, and a far more balanced take. Nice work!
What's this Raspberry Pi you're talking about? At this point I think it's a mythical creature nobody has seen in 2 years. If there's anyone killing RPi, it's RPi's lack of availability. I wanted to get my hands on a CM4 4GB RAM, 32 GB eMMC for what seems like ages. Or a mere Pi4 with 4GB of RAM. Unofficially, RPi 4GB's price is $140. This is what I would have to pay for one on eBay. No thanks.
This isn't a video about stock shortages or availability. It is about features and the technical aspects.
@@GaryExplains I know, Gary... I had no intention of being rude. I was just venting my frustration about the shortage. And I'm fully aware that this is not something the RPi Foundation has much control over.
VisionFive becomes the Pi savior by taking up all of its demand, lol
You can technically get a computing module and directly use that as well.
Wow, here I am using one to run octoprint
Interesting format with two of you having a dialogue. Works pretty well. Seems like this board is going to be interesting for people who like solving puzzles for our amusement and this provides a low-cost entry point for people who want to play with RISC V.
Thanks for this excellent unbiased review. I've been watching you for many years now and I value how you have managed to maintain integrity and excitement on your channel
Great review. I came to the exact same conclusions you did and had the same firmware trouble you did. Thanks for mentioning the minimal SD card image as well. Someone had to tell me about that one on Twitter because it's not in the official StarFive documentation as a firmware update method.
It's not a board for beginners. If you are experienced with boards though it's an incredibly powerful RISC-V board (the most powerful one I know of that doesn't go inside a server and is a SBC). If you are wanting to get in or develop on RISC-V it's worth fighting through the firmware update. If you are maker looking for an easy board to use absolutely not.
Thanks Gary!
Most of these boards are not really for beginners. Hopefully after some time they will get there. I think there is a huge benefit to releasing an early bird so the hardcore fans and experts can tinker and work out bugs and those who want something more polished can wait a few months. The problem is that most open source projects never get the polish and documentation required for a trouble free experience.
It's pretty much spot on what I want in a SBC. Two 1Gbs ethernet (for running a router), 4 X USB 3 ports, a nvme connector to plug in large storage and so on. I've want an arm SBC with these features. Just needs a faster cpu.
HardKernel Odroid...
When making the point that only the ISA is open it's important to note that there ARE open core designs being worked on. This is significant for people who want to own their hardware in every aspect and for security. Hardware backdoors are one of the biggest unmitigated security concerns and open core designs (built on an open ISA) is an important step in knowing what is going on under the hood. Ensuring that the core design sent to the manufacturer is indeed what the manufacturer produced for you is still a problem. Some day it is my hope that makers will be able to manufacture microprocessors themselves (no back doors!) but semiconductor manufacturing is incredibly complex.
I forgot to add that like many open source projects open core designs may need community support to flourish. Consider donating to your favorite open core project if you find value in it.
I cover all those points in other videos, I can't repeat everything in every video ad infinitum.
You might want to look at the video on CZcams by Bunnie: Wednesday 11 00am Keynote Address Impedance Matching Expectations Between RISC V and the Open Har
He knows the ecosystem well, so he can give talks like: BlueHat IL 2019 - Andrew "bunnie" Huang - Supply Chain Security: "If I were a Nation State...”
Everyone forgot about SPARC 😢
This is not a replacement for the Raspberry Pi, rather a demonstration platform for new architecture with a lot of potential for practical use for the advanced tinkerer. One warning for early adopters is that if the board is not widely adopted, then it will end up with limited support by familiar projects with finite support.
bound to have very high adoption, as it's currently the only riscv board in this price bracket. also pine64 will release a board with the same SoC. there's really only the earlier, a bit cheaper but much slower mangopi with the D1 SoC and shortly there will be the sipeed licheepi with the TH1520 SoC.
We need one with a great development community behind it. Then you can officially call it a raspberry pi killer. Your talking about having to recode almost everything I have done to switch to this.
it seems that the only advantage is the lack of licensing fees of the instruction set architecture for manufacturers. but the manufacturers still create proprietary IP like GPUs. so we will still have binary blob drivers. and here i was hoping to see fully documented hardware interfaces. that's really the only thing for me that would outweigh the performance, price etc. good video
GPUs is hard to do and very specialized. So far it hasn't happened yet.
You nailed it. As a user, you do not benefit from RISC-V in any way. As a user, the x86 is the platform with the most freedom.
I am really looking forward for the HiFive Pro P550 MicroATX RISC-V development board which should give us finally a little bit more computing power.
More power yes, but the price hasn't been announced. I expect it will follow previous SiFIve boards using low volume prototype chips (MPW / shuttle) and be $500 to $1000. Lower cost boards with mass-production SoCs (such as the VisionFive in this video) follow a year or two later.
What is interesting: Intel is involved.
@@BruceHoult Less expensive, mass-produced boards like these SoCs won't find buyers if they perform so poorly compared to a 7-year-old Raspberry Pi 3.
@@OpenGL4ever They absolutely cane the Pi 3 in anything not using SIMD (which the JH7110 doesn't have). Of those kinds of things are better done on the GPU, of which it has quite a powerful one. Also in the real world having 4 / 8 / 16 GB of RAM works far better than the maximum 1 GB on the Pi 3.
@@BruceHoult If you need more RAM, then you can still buy a Rasperry Pi 4.
On the SIMD vs. GPU thing, it really depends on what you want to do. SIMD support is much easier to implement on the software development side, because it just needs a better compiler that does support these SIMD instruction. If you have to do all the work on the GPU, then a simple recompile with a current compiler won't do it, you will have to rewrite your code to make it GPU aware. And that is a huge untertaking and thus a big difference.
So to sum it up, a GPU does not replace the SIMD on a CPU in most cases.
In addition, a SIMD unit also helps to speed up ordinary programs where humans would not even think of optimizing the program using the SIMD instructions (or the GPU). In this case, the optimization is then performed automatically by the compiler, because it cannot distinguish between computationally intensive sections of a code and normal code; it will simply optimize with SIMD wherever it sees a chance of achieving optimization with it.
I.e. also with every ordinary forest and meadow code.
But that only applies to the SIMD, it doesn't work with the GPU because the compiler doesn't optimize for the GPU. Using the GPU requires the use of APIs such as OpenCL. And that has to be used extra for the code.
The GPU of the JH7110 therefore remains unused in most cases, while the SIMD unit of the Raspberry Pi will be used in many cases due to a better compiler. A SIMD unit is therefore missing in the JH7110 and this is also a big disadvantage.
Software support is always the key
Thx Gary👍 I hope that most of viewers will understand where RISC5 chips stand right now in the industry and where they’re going…
I think you should try measuring power consumption while doing the benchmarks and compare the star5 vs the raspberry
Thank you for your detailed review! I'll hold buying a RISC-V computer until it became more performant
Thanks for the detailed review! I've been interested in this board, just as a programmer intrigued by a new processor architecture, and it's nice to see how it stacks up to existing Raspberry Pi offerings.
Now this is a massive improvement compared to the first one! Tho i wonder if sifive makes comple custom boards if ordered as. If they do it'll probably be a mass order of thousands of boards haha still a minimalist board with nothing but the soc, ram, usb c for power, m.2 ssd slot, microsd card slot and ethernet port is something I'd like to have as a simple personal server
I'm a software developer. I already have to test on Windows on X86, Android on X86, Android on Arm, multiple flavours of Linux on X86, multiple flavours on Linux on Arm, iOS on Arm, Mac on X86, Mac on Arm. The combination of CPU, GPU, OS, driver versions goes on and on and on. I do not want yet another ecosystem to worry about.
I've got one of these on order at no point have a thought this is a RPi killer, it's something new to hack around with which will find a niche.
My loose idea is to use mine for an automation project that needs Linux over a esp32
its early but only way to get a community is to build a plattform to get a community to form around,its great to see it coming out.
Tq Gary. I wonder how many of these sell (and where are they being used). I fear that if they don't sell many then the manufacturer may just give up. I hope not as I'm really warming up to RISC V
Thanks Gary! :)
Had this board for quite some time sitting on my desk - guess I should crack on and boot it one of these days... If I recall Starfive was showing some benchmarks with an 'optimized' kernel that would make it score higher than a rpi4b but I haven't seen anything about it in the real world
I don't recall the same claim. StarFive claimed that the CPU is comparable to a Pi 3, but the GPU is faster than the one in the Pi 4. Unfortunately the GPU driver is currently a BLOB, but Imagination Technologies has expressed intention to support open source better (without telling when they are going to do what).
@@LivingLinux ah my bad I got mixed up with the sipeed lichee pi 4a RISC-V announcement which touted significant speed improvements when compiled with an unnamed optimized toolchain rather than the default RV64GC toolchain
The open source could matter to the chip makers in the long run. Over years it could amount to millions.
Really liked the benchmarks that was eye opening.
Over years ? The licensing for chip EDA software is in the millions
@@mattweger437 both of you should watch the many videos Gary had made explaining the misconceptions surrounding "open source" hardware. It's all in the risc-v playlist
@@psd993 I'm working with the esp32c3 right now and loving it. I've been sold on riscv for years. So it probably won't sway me much
@Matt Weger What advantages are you personally getting from using the ESP32-C3 compared to the ESP32-S2 or even the S3. How is RISC-V making your project better that the previous models of ESP32 chip didn't?
@@GaryExplains well the s2 has a riscv ulp co processor as well I was going to make the argument that it's more power efficient but....
I'd buy into this just because it has all the ports on one side.
Cute title. But no one in their right mind thinks this is a replacement. It's a stepping stone to making RISC-V a serious contender. And those of us who have a board know that there's plenty of work left to be done.
The board isn't sold or marketed as a stepping stone. Where does it say that? My job is to inform people about what it actually is. As you said, "people with the board" know the truth, I have the board and told it like it is. You shouldn't make bold assumptions about what you think people do or do not know.
@@GaryExplains if the underperformance is because of immature software. Particuly in the compilers then it will get better over time. If it is the hardware , then it is what it is.
@@GaryExplains I’m very sorry if you interpreted the marketing as a replacement for the RPi. I don’t think that was their intent. The first line of the Kickstarter page states: “A *pioneering* development board that combines performance with low-cost and full open-source RISC-V single board computer.” The video states that the previous board was just for developers, but that they are attempting to “open it up further” and to make the board “as approachable as possible for developers”. They then said that they hope to “bring the VisionFive 2 to as many developers as possible” to “kickstart the open source revolution”.
The page does have comparisons to the Raspberry Pi, but you’ll note that it’s more comparative in nature and that the CPU performance is significantly below that of the RPi4. StarFive does not advertise this as a “replacement” for the Raspberry Pi. Only an attempt to create a competitive RISC-V board in this space that can be afforded and accessed by more developers than ever before.
As you experienced, the actual board that has been released is still very much underbaked. The hardware is all there and wired up, but the software simply can’t utilize it. Build 55 only has one of the two Ethernet ports working, supports almost no USB dongles, only 4GB of the 8GB chip are accessible, 3D support is… not quite there, and NVMe performance is rather abysmal. (I’m convinced it’s a software problem and that the hardware is likely capable of speeds of 300-400MB/s.) Build 69 requires a major firmware upgrade and then only fixes some of the problems. I’m not sure if they fixed it, but it wasn’t even booting without a manual U-Boot start at one point.
StarFive is working with their partners to understand how the hardware is interacting with the RISC-V CPU to get all these problems sorted. But even once the software reaches a high level of support, you’re still looking at a board that’s not that competitive with the RPi4. It will be _accessible_ though, which is the real goal of this board. Mango Pi is great, but that’s not really a board that pushes RISC-V toward the mainstream. This board hopefully will, but is very unlikely to be the final say before RISC-V is competitive in the 64-bit processor space.
StarFive's PR department couldn't have written a better reply!
@@thewiirocks the CPU power is close enough to Pi 4 (about 80% for general purpose code) that you'd be hard pressed to tell the difference without a stopwatch. The GPU, once properly supported, will give a MUCH better desktop experience, CZcams etc move playback compared to the Pi 4. Give it six months and this will be a better desktop replacement than a Pi 4 is for most people. The better-than-SD storage is the icing on the cake.
The main exciting thing about RISC-V is that it is the same leap in competition vs ARM that ARM is vs x86. The sheer number of companies developing RISC-V cores and associated devices is outrageous.
The secondary exciting thing is that RISC-V's ISA design is actually turning out to enable significant improvements in power, die area (which means money), and microarchitectural flexibility... what does this mean to consumers? Even ignoring the licensing opportunity cost (and the actual fees, however small) and market size advantages, the actual design of the ISA enables smaller cheaper chips to run faster, with equal R&D effort.
Obviously right now, the core designs are less mature in the RISC-V world, having received a tiny proportion of the investment, but SiFive's licensed cores are definitely catching up. The U74 is a several years old design by now, and as the P500 and P600 cores start coming to SoCs like this, and as the architecture-specific software tuning comes through, it'll become clearer.
Did you watch my video on the power efficiency of RISC-V? It didn't turn out well RISC-V.
@@GaryExplains power efficiency depends way more on lithography than chip design. If the Die size for RISC-V for the same size transistor is smaller than an arm processor, it will be more power efficient. However, risc-v processors are not being manufactured with the same lithography of the arm cores. Arm processors have smaller transistors, and therefore are more efficient. 7-12nm vs 14-25nm is not a fair comparison.
@@vladlu6362 I guess you didn't watch that video then... nevermind... In that video the RISC-V chips were on a better process node than the Arm ones!
Thanks for the nice video, it's a bit confusing that, ARM itself is RISC where RISC-V based on, unlike RISC vs CISC that are quite different, some viewers may not know about that.
How can you kill something that is not available?
Huge pi ecosystem vs high performance upstart?
I think its a neat tinker device. But especially if you want something stable now its clearly way to early. Not to mention that Compilers, Runtimes, Libraries,... all that needs to be ported an optimized for RISC V which will take at least a decade.
I myself even recently was thinking about getting an ARM board for my homelab, but still the Software isn't 100% there yet with for example Proxmox not supporting any ARM chips. So I went for an low powered Pentium instead. Just goes to show how long it is to switch over to an entire new hardware platform.
This board is aimed at developers with it's low cost. We need boards like this to improve RISC V adoption over the coming years.
I will buy at least one risc-v board. not to use it though. but to support the trend...
Btw, does RAM/ROM has any type of "licensing" or is it open?
Also, when will we have ssd faster than RAM/ROM? enough already.
Exactly
Open source software would not be where it is today without people supporting it in its infancy
Ask the folks who use qualcomm chips when the manufacturer decides not to release any further firmware, you effectively are left with a device (typically Android phones) with no further upgrade path. This happens a lot with Android TV too
@stultus es You can support open source software easily, you just need a pc and an internet connection to use it and/or to contribute. I downloaded my first version of Linux with floppies. How do you do that with open source hardware? It costs millions of dollars to make a chip.
@Roman Ash I am impressed that you are passionate enough to spend real money on something you won't use, but because you want to support a trend. So, please tell me why it is important to support a Chinese company that is making a closed-source SBC? You know the CPU and GPU aren't open source, right?
Why?
Yes competition is a good thing. But all Risc-V really gives us is an incompatible alternate instruction set to that provided by ARM, which will no doubt fork in to multiple semi-compatible versions sewing further confusion and frustration as is the nature of unfocussed open source development.
For what? For manufacturers not pass on to consumers the few bucks they saved in licencing while passing on the consequences of all the above.
Imteresting, thank you Gary / Gary. Obviously a bit underpowered but worth watching. For me it was only when we got to the Pi 4 that it finally became ‘good enough’ power wise and so I am prepared to be patient and waitfor RISC SBCs to mature.
The real magic of the Raspberry PI isn't the Arm Architecture or performance, the magic is the support and documentation.
There are various SBCs that beat it on specs, but there software support SUCKS! Raspberry PI has a much better support community.
Risc V is attractive to companies who don't want to pay license fees to arm and/or want to completely own the design.
Since the *standard* is the RPi, and it is published & popular, why can't they keep pinouts AND names the same? If there was a copyright reason I could understand, but that ISN'T the case, right?
I've never seen a pi sbc in Canada ever for under 90. Even when there was no "component shortages".
In the end the conclusion is reasonable, if oddly-stated. If you are a developer/maintainer of some software package and you want to get it running on RISC-V (maybe people are already raising support issues asking for RISC-V) then this is the board for you. It's 5 or 6 times faster than the Allwinner D1 boards that were previously the only thing available for under $100, and it's got up to 8 GB RAM vs the 0.5 or 1 GB on the D1 boards. Night and day. You could get the same performance for a couple of years already on the HiFive Unmatched (which has 16 GB RAM and 4x PCIe vs 1x here, and can use high end Radeon or Nvidia video cards), but costs 10 times more. BUT if you are a casual user who just wants to buy a board and every software package you can imagine already works then no, this is not for you right now. Gary says to wait a year. Maybe. Or, more likely I think, three or six months will be enough.
Pretty much.
There's some new RISC V architectures on its way out that I'm excited about.
But I haven't found any products with those cores yet...
@@Luredreier It is the natire of the electronic industry that you always know what is coming three years ahead. There are ALWAYS going to be fancy new cores announced that will take two years until someone announces working test chips of an exciting new SoC, another year to be in mass-production, and another six months after that to be on a board in a store where anyone can buy it. Right now, the VisionFIve 2 is by far the best price/.performance RISC-V board that you can buy and have delivered to you within a couple of weeks. The Star64 will soon be as good, better for some people. In three or four months the Sipeed LM4A with some dock may take over the mantle (it will be a little more expensive, but also 50% to 100% faster). HiFive Pro will be maybe 50% faster again, but cost maybe ten times more -- which will be well worth it for anyone developing on RISC-V professionally, just as companies give employees $100, Windows and Mac computers, not Raspberry Pis.
@@BruceHoult Some things are worth waiting for.
The P670 cores from SiFive has *2* vector pipelines unlike this core that has non.
And it can run 4 instructions pr clock cycle.
This...
Not so much...
It's a *significant* upgrade.
Sometimes performance *matters*.
@@Luredreier P670 will be great. The core was announced in November, so expect a $1000 dev board two years from now in the first half of 2025, and maybe affordable boards late 2026 or 2027. The T-Head C910, by the way, also has two vector pipelines, each with a 256 bit ALU. A $100 Sipeed board with it is promised for the end of Q1 *this* year i.e. maybe 3 or 4 months away. The only problem? It implements RVV draft 0.7.1, not the final ratified version. Mildly annoying but, hey, three year head start...
@@BruceHoult Hum, true.
But I'd still *really* would like 4 integer pipelines.
I'll have to read more about the c910 though.
Honestly it would be kind of fun to implement a homebrew RISC V CPU sometime.
Oh crap. Gary's cloning himself now. Gotta shut down his cloning operation before the clones go all Vault108 on us.
I'd be interested to know which compilers were used on the ARM and RISC-V systems. I'm assuming GCC or maybe clang. With RISC-V being a rather new ISA, how much optimization can these compilers do right now? Were these benchmarks compiled -O0 across the board? That being said, I'm not surprised by these results. New ISA on new silicon means SLOW. I'd expect to see much more comparable results in 2 or 3 years.
The first company to get it together and use PI4 ready software... survives when PIs return to market 2nd quarter of 2023. I can do EVERYTHING with Twister OS and /or Android 13. Both overclocked to the limit and I can run them on batteries for power tools.
I couldn't really get mine to work with my 4K screen. No picture at all with HDMI 2.1 enabled, 640x480 with 2.1 disabled. Hooked it up to my TV and got 1080p. Fortunately the serial monitor is reliable, so I could see what was happening. In the end I resorted to doing things the old-school way and writing a few mode lines in the Xorg.conf file, and that gave me working 1080p on my 4k monitor. This was also not without problems though, some modelines I wrote would just completely lock up the board, or cause various unrecoverable kernel errors. And they weren't even lines that should be particularly taxing on the hardware, it was just totally random. Also, any 4k mode I tried would just totally lock the board.
Awesome video, I'm personally really excited about RISC 5. I actually want one for a classic mac restoration project. I have a sneaking suspicion that Mac OS Classic might be a good port for RISC 5 Systems and I fugred it would fit into a Mac Classic Case 😉
What benefits does RISC-V bring to your Mac port that you don't get with a Raspberry Pi or Arm based SBCs in general?
@@GaryExplains CPU Types mainly and the ability to cross over the coding. Power PC is RISC, is there any need to say more 😏I'm also working on two Mac OS Ports over at Macintosh Repository. Also I noticed RISC 5 is open about it's CPU Calls and Scims. I also have the same sheets for the Motorolla PPC. It will require some research and mass tinkering, but I think, the RISC 5 could be used as a drop-in replacement for old power macs. Keeping them from the landfills and giving them new life. Finally, just saying, screw ARM 😝 Long live RISC 😆
Don’t forget: ARM means Advanced RISC Machine - ARM is RISC!
@@SchoolforHackers I know 😆🙂 I just like how simple the Risc 5 is. Besides, it's just an idea 😀
Are you doing the port in assembly?
I've seen more favourable benchmarks of this board, maybe they had a different installation/compiler for their benchmarks.
I'm in the excited camp. Not just that Linux desktop works, but that it's super affordable, especially compared to scalper raspberry pies. Even for a maker, if it's not about the result, but the journey and you want uncharted platform development, this is it.
I gotta say this board still sounds better than a MIPS board I used before.
Nothing to do with compilers in the benchmarks Gary showed. One was a hardware SHA instruction vs software -- next year's RISC-V boards will have SHA and AES in hardware, but this one doesn't yet. The other thread test is about OS internals not compiler. Someone is doing something stupid such as not implementing lazily switching FP registers, or needlessly clearing the entire cache or TLB on task switches. It looks bad on a test that does nothing except task switches, but would barely affect normal use, with task switches only maybe 100 times a second.
Bruce, sorry mate, don't try to explain away my testing. I give single thread and multiple thread scores. The multi thread scores are consistent, relative to the single thread scores. The low performance of this board has nothing to do with the OS and some bad task switching. Please stop trying to make excuses all the time.
That's a pretty sweet tribal marker there, Gary.
9:45 the question here is this slowness the underlying hardware , or is it more the fault of an immature compiler etc?
The only real way to compare "fairly" would be something like comparing the speed of certain similar assembly instructions, and even then it would differ based on how much better the maker of the chip wanted those instructions to run, or if they even support it in the case of extensions to the instruction set
It's early days for RISC-V, people are just getting their feet wet with the tech, developers need to learn the low level stuff. Once it matures with optimisations then there will be the economies of scale that ARM processors enjoy. Long run, being able to stitch silicon without messing your code up too much, and on a core that's guaranteed to be around for a long time is a safe long-term bet. This is why I believe Tesla is using RISC-V for their Dojo chip; they can optimise the silicon to their absolute needs and stamp them out without any licensing concerns.
To expect RISC-V to win out over ARM at this early stage of the game is a bit of an ask...
Hahahaha… i remember seeing your guest somewhere but cant remember where 😅
i cannot help but think the processor market is beginning to fragment in the same way we had multiple processor types in the 80s.
8:15
The lower cost has more to do with RISC V being smaller and simpler then ARM for a given amount of capability due to less legacy bloat.
On the other hand you lose the mass production scaling benefits of arm...
But yes, if mass produced this *should* in theory be cheaper and more energy efficient.
As for the licensing...
You'll still have licenses with RISC V, it's just that there's potentially more competition as *anyone* can make a compatible processor without investing a ton in a new software ecosystem with compilers etc to go with it...
Nonsense. Armv8 doesn't mandate legacy support and the latest Cortex-A processors are 64 bit only with no legacy support. And none of that applies to the Cortex-M series.
If you install gentoo on the visionfive2 you will find gentoo runs faster on it than it does on the raspberry pi 4. I did so and compared the results of genlop -t to check the compile times as well as simple benchmarks like
time echo "scale=5000; 4*a(1)" | bc -l
I have my make.conf compile flags set to
"-mabi=lp64d -march=rv64imafdczbb0p93_zba0p93 -mcpu=sifive-u74 -mtune=sifive-7-series -O2 -pipe -fomit-frame-pointer --param l1-cache-size=32 --param l2-cache-size=2048"
Exciting stuff and yes this is not for people who just want todo things, the RISC-V software stack isnt mature yet, neither is the toolchain, Imaginations open source drivers also need some serious banging on them, they arent doing AMDs approach on open source GPU drivers not very long either. But this platform will likely mature fast, interrest from normal users and tinkerers who can finally work on real affordable hardware now is huge and industry backing of RISC-V is also very big.
I ordered this board only to get vkQuake running, we will see 😉
Lets wait for another decade before we should use risc v soc in SBC.
Intel is going to release Horse Creek this summer. Alibaba is pumping out new RISC-V chips. Google announced Android support for RISC-V. Things are moving way faster than you want us to believe.
@@LivingLinux I hope RV uses modern OSes, and then it's a race to always be on the cutting edge. This SBC is remarkable.
@@BeeRich33 Once we have mainline support, Debian and Ubuntu will probably release the RISC-V images together with the other architectures. And probably more distros will follow.
@@LivingLinux Not much else on the market re: this new segment. I'm so looking forward to these flooding this current ARM void.
@Mahesh Yes, there are general RISC-V images available, but they don't work on this specific SBC. I saw reports that people tried with the Ubuntu image for the original VisionFive, but it seems that doesn't work.
Looks like we have the same issue as ARM, where you need specific images per SBC.
What I wonder if this board will be good enough in a year or maybe even 2 so people working on creating products based on RISC-V as embedded and microcontrollers can easily use it. Because that's where it all starts, right ? Which sounds to me like: in 1 year, having a bunch of these in a rack for building stuff will probably be a good option.
The performance compared to the 7 years old Raspberry Pi 3 is really poor. If i want learn RISV-V assembly, there are better and cheaper options available. The only reason to buy this SBC would be if you need a Linux machine and are a developer of some kind of Linux application that also contains assembly code in order to then optimize your own software. In all other cases, a Raspberry Pi 4 is the better choice.
RISC-V FTW!!!
Thanks for the review. As a relative newby, I just want to play with the gpio, and not spend hours trying to boot a new board.
The raspberry pi's are getting rare and expensive, and I hoped for a cheaper option.
Maybe this Risc-V is not the answer. Cheers.
Just buy a Raspberry Pi 3 or even cheaper a Arduino if playing with the GPIO is the only thing you want to do.
Jim Keller is a journeyman computer architect. Patterson has been slapping the monkey in academia his whole life. Patterson has been working on an ISA for forty freakin' years.
What has Jim Keller got to do with the Raspberry Pi?
Did I miss something? What's the big deal here? Why would anyone switch to this "miraculous " RISC5 troublesome slow device when many other solutions are available? Maybe I need to revisit this video with a strong coffee.
There's no 'Raspberry Pi' killer. Even if that thing runs at 7 GHz, has 64GB of RAM, offers every interface under the sun, and costs only $5. Raspberry Pi is a religion. People use it not because it's so great, but because it represents fulfillment for them. There's nothing that would ever separate the Raspberry Pi devotee from their beloved company.
Any chance for an update video on how far the eco system has come
❓What's the best price / performance board for Linux?
❓ For ease of use for GPIO SPI ports etc what's the best board / firmware / OS / software/ development software combination
Has gotten a Pine64 Ox64 SBC? Haven't seen much coverage on it
Eh?
RISCV - super easy barely an inconvenience
Thanks for a cool video on this board, interesting format, my take, is nothing could ever be a RPI killer, wouldn't matter if it was half the price and twice the power, RPI just has too much community support and momentum, what is nice, is we have yet another option for us to tinker with, competition is king, and hopefully the platform improves rapidly.
Check out the Orange Pi 5. It's a surprisingly good board that gives Pi a run for its money on the high end. Software support was *surprisingly* good out of the gate, and the community has been jumping on the thing to build it out. I don't know if it will _replace_ the RPi, but it may turn this into a two competitor race.
orange pi 5 : best sbc right now
rpi 4 is best right now if you can get it
Has Patterson ever taped out a chip? Has he ever produced an actual product?
Super-selective benchmarking there Gary. I'm sure you know very well SHA and AES instructions were added to the RISC-V ISA in December 2021, long after this chip was designed, so you're comparing a hardware instruction on the ARM chips vs a software implementation on the RISC-V. Yes, sure, if you spend all your time calculating SHA hashes then stick with ARM until next year's RISC-V boards are out, but who the heck other than Bitcoin miners does that? If you want to show how this lack of SHA and AES instructions affects normal users then show us how long it takes to ssh into a server, or copy a file over the network using scp or https. I'm sure the Pis will win that too, but not by 500% or 700%. The same goes for the thread test. Sure, that's a useful benchmark for people optimising an OS, but how relevant is it to the average user? Not very! If you want to show how thread creation and switching affects (somewhat) normal users then show running autoconf and make -j4 on some large software package. What you've shown here is accurate, but very misleading.
You know that a make -j4 will be more about I/O than CPU. If you want to offer alternative testing at least be honest about what will be a good measure. But in the spirit of openness and "giving it another chance". I wrote a quick single-core benchmark that does software SHA256 hashing in C using a function from Brad Conte. It loops round doing multiple SHA256 on a small input. The Pi 4 finishes the test in 8.9 seconds, the Pi 3B in 16.32, and the VisionFive 2 in 18.6.
@@GaryExplains that’s much closer to what I’d expect, and a good result as the VF2 has micro architecture and clock speed similar to the Pi 3B. Please provide your source code. And you’re wrong about compiling. If you look att the total data read and written and divide by the time taken you can easily see that compiling isn’t stressing any modern “disk” at all.
"Disk" as in a class 4 Micro SD card?
@Bruce Hoult The code for my new and shiny softSHA256 threadtesttool is here: github.com/garyexplains/examples/blob/master/threadtesttool_sha256.c
@@GaryExplains I don’t know what “disk” you used, as the board also supports eMMC, NVMe M.2, USB3, and of course things such as NFS over gigabit Ethernet, all of which are faster than SD. However I’ve found that on similar boards (both ARM and RISC-V) as long as you have at least NFS over 100 Mbps Ethernet (I.e about 12 MB//sec) and enough RAM to cache binaries and header files, additional disk speed makes little or no difference.
Looks pretty slow vs even RPi3 let alone RPi4... but if for whatever reason someone wants/needs RISC V system then I guess it should be fine. Not sure about security aspect of it... probably best to stick to something more proven in this area.
This is a prerelease board isn't it and thusly unoptimized?
Sorry for the repeated comments.
Your a scream. Love RISKVy Gary. ❤
4:50 The word "ON" printed on a dipswitch means the switch is closed, that word has nothing to do with the function of the switch. Says nothing about what the switch will do when it is turned on or off.
What does the 'file' command returns when you inspect a binary? Like "$ file /bin/bash". I am super curious about the reported architecture string!
Got docker? Run "docker run --platform linux/riscv64 -it drujensen/riscv-ubuntu bash" and play to your heart's content. Works fine for me on both x86 Linux and M1 Mac. To answer your question I get "/bin/bash: ELF 64-bit LSB pie executable, UCB RISC-V, RVC, double-float ABI, version 1 (SYSV), dynamically linked, interpreter /lib/ld-linux-riscv64-lp64d.so.1, BuildID[sha1]=78af774fdd526282745b82bd902a9bba4d1083a8, for GNU/Linux 4.15.0, stripped."
nvme/mmc, so better storage and 2 gigabit ports, power usage ?
Silicon pen, nice
If anything kills the Raspberry Pi, it won't be this single-board computer, but rather the Raspberry Pi Foundation. Raspberry Pi SBCs have been in short supply or completely unavailable for close to a year, because they're shipping everything they build to industrial customers, with nothing left over for experimenters and students. I wouldn't waste my time or money on something that is so hard to get. The Raspberry Pi Foundation should have licensed the designs to those large-scale customers and let them deal with procurement on their own.
Does Gary know he’s got the “early bird” version I wonder?
Mine looks the same, as I could see on the periphery of the "non" platted hole near the USB-C socked are two SMD whatever. So it is kinda platted. And I got it not as early bird; or they were already scalped for just selling.
If you are interested in CPU microarchitecture, and you have ideas, want to try them out, want to develop new tech, then you can work for Intel, AMD, ARM. But what if you don't want to work for these companies? What if, like many Linux developers, they don't want to work for Apple or Microsoft?
Step in RiscV. You can now develop architectural enhancements as a freelancer at home, or just as a hobby. This is very powerful. We have an army pf people who are intelligent and who can now start thinking about the architecture of the CPU. Their ideas can be integrated into RiscV.
Ok and where will you find the $100 million to make the chips at 5nm?
@@GaryExplains You can validate and verify your ideas and tech on a simulator and on an FPGA for essentially zero cost. Not so much the case for ARM and X86!
If you get to rolling a chip out then you are entering the world of commercialisation. As a freelancer, you can either make your design public and free, or you can license it to someone who is going to roll out a 5nm design.
@@nickrhill OK, so what is the world going to do with an FPGA design of your RISC-V chip? Put it in a laptop? Please be serious. I designed my own CPU instruction set and wrote an assembler for it, all here on his channel. It isn't rocket science and I didn't need RISC-V to do it.
@@GaryExplains Nobody is putting an FPGA in a laptop. An FPGA is suitable only for evaluating a design by running real-world code, such as the linux kernel to carry out performance evaluation.
Making an *optimising* compiler and making a CPU instruction set target for it is somewhat complicated. Having a design which you already have a compiler target for, and an optimised operating system binary for should remove many hurdles to developing mainline performance CPU technology.
And then even with the optimized compiler all you have is a hobby CPU on an FPGA. So what is the point?
If it's Risc, why does it have an Instruction cache? O_o
such a discrepancy in performance might be a result of poor optimization by compilers
Rasperry Pi can't be killed if doesn't exist - out of stock long time.
So how much would the ARM licence be on say the Raspberry PI 3 or 4? Something like $1?
Yeah something like that maybe even less.
Imagination designed the soc for the sega dreamcast if I'm not mistaken.
The PowerVR2 was paired with the Hitachi SH-4 in the Dreamcast, with the SH-4 as the T&L geometry engine and the PowerVR2 as the rendering engine. Also there was a MIPS CPU in the PlayStation, and Imagination owned MIPS for a while, but much later.
@@GaryExplains thanks for explaining. Hah! ;)
if it can run linux, pihole and unbound and is cheaper than a pi(which i still can't get) i might buy it
Lucky that I got a raspberry pi 4 before they were so hard to get. It is ridiculous
Found the video informative. But you know the Arm processors are actually RISC processors. ARM stands for Advanced RISC machines. But calling it ARM vs. RISC is misleading as both RISC V and ARM are RISC style professors. Also. PowerPC, M1, Sparc, DEC Alpha, and SGI R Series processors.
But there is a difference between RISC and RISC-V. I have plenty of videos here on this channel covering most of the things you mentioned.
What are benefits of RiscV if it isn't cheaper?
That is a very good question and one I find hard to answer. There are lots of zealots who will give you theoretical answers about the specification being open, but not much else. I have a series of videos about what RISC-V really is, if you are interested.
@@GaryExplains I've watched most the videos already. Actually, there is one good reason, some big companies are interested to successfully compete with Arm and eventually get a lead there. We will see soon.
The reason why super cheap microcontrollers use in-house instruction set is because the licensing fee would cost more than the entire microcontroller.
So in some cases RISC-V indeed matters to the pricing!
So you think that the license fee for a Cortex-M0+ is greater than the price of paying an engineering team to design a RISC-V processor? 🤦♂️
@@GaryExplains It is fairly simple to design an ultra low price microcontroller. It is sometimes given as a semester project for students.
Modern VLSI synthesis tools can translate it to circuits very easily and completely automatically. The only hand designed parts are the pad interfaces, but that usually a given by the manufacturer.
I have once designed a simple microcontroller with an UART, in an afternoon. It was only synthesized for FPGA, but if you do not need bleeding edge clocks and optimizations, the synthesis for VLSI is just as simple.
Even for one engineer it is costing you tens, if not hundreds of thousands of dollars per year in salary, office costs, taxes, heating, benefits, etc. You still think that is less than the few cents you pay Arm. Even if you license a RISC-V core from SiFive or Andes the maths is the same. That is why the licensing model exists in the first place. Then there is the cost of getting your chip verified so that you can use the RISC-V logo, etc.
Is there anything about the RISC-V protocol that makes it decidedly better than ARM? In other words, after decades of time and billions of $ have been spent in R&D, can we expect RISC-V processors to be the architecture of choice for most applications or will they roughly reach parity with ARM?
Pardon my lack of understanding for computer science; I think that amidst the hype for RISC-V I'd made some assumptions about its qualities.
It's better because it's open source, meaning any company can come along and design their own chip around it. The potential market for that chip will know it's compatible with software designed for RISC-V. Makers of RISC-V chips will have the option to differentiate based on how they optimize their design, or if they support certain instruction extensions, etc. without worry.
It's pointless asking something like "after decades and billions", as any company worth that much would make sure that they make it worth it, not to mention this assumes the equivalent hasn't been put into competing architectures, other variables like consumer choice, etc...
RISC V doesnt have bloat to be compatible with old systems as modern ARM and even more so for x86. So it could in theory have more performance. But as Gary has talked about in his comparisson video of various processors, the productiong of the CPU matters way more than the instruction set.
Big issue for RISC V is it barely has any software compatibility where ARM and x86 got way more, which hinters its practical performance.
Not really, it's just a different way of expressing the same thing. Think of ISAs (Instruction Set Architectures) as a kind of language, some languages are richer and more complex, some simpler, some have concepts that don't map exactly to other languages, but at the end of the day they're all moving bytes about in memory and transforming them in various ways. All modern ISAs are quite similar in their base assumptions. I think the main benefit of RISC-V is that it's free for anyone to implement, which will create a richer ecosystem of CPU vendors compared to ARM (very expensive architecture licence) or x86 (impossible to licence, patent duopoly).
Also since all the hard work of porting open-source operating system, compilers and software is already done and shared with all vendors, it's much easier and cheaper to just use RISC-V for your CPU rather than inventing a new ISA and having to do that all yourself.
RISC-V also comes at an interesting time patent-wise, since patents from 2003 and earlier are now expired, so CPU designers can use these techniques without having to licence these patents. They can at least match Pentium III or 4 which were released before 2003.
😂😂😂 You’re so sour about the “open source” label and I’m sure the entire community is just confused about it because you’ve never actually explained what “open source” means in RISC V. RISC V is an ISA. Just as x86, x64, and ARM are. The “architects” come up with the networks of transistors that perform specific functions. These functions are what are called “instructions”. A group of them is called an “instruction set”. Ah processor is made up of loads of these networks and controllers and ASICS and I’m sure I’m missing some other stuff because I’m not an expert, but they all are designed and catalogued into the ISA. Apple designs their own SoCs. In order to be able to do so and access the catalogue, they had to purchase a license from ARM. What instructions the license permits is entirely up to ARM. It is a legal agreement. With the catalogue, some software, and some people who know how to use them, you can then design the image of a chip and have it sent to TSMC to be forged. If you are caught using any instructions not in your license, or using the ISA without a license, you can get sued you into homelessness by ARM. With RISC V, the ISA is open sourced. You can literally go to their website, access open source software to draft (3rd party), access the entire Instruction Set Architecture with VERY detailed explanations of how to use them , and design your own RISC V processor, ASIC, etc. Due to the license they share it under, they can never sue you for using it no matter how much you profit from it. But if you do so, please pay something back. If they can’t pay engineers, you’re product won’t ever upgrade from a new version of the ISA. SiFive used the ISA to design a processor and had it founded. They do have the right to protect the Instruction Set they curated and the right to license it as they choose. This is also stated in the license RISC V uses to share it. If a company doesn’t have the means or desire to design their own processor, they can purchase them or even license the design from SiFive to pay for their own to be founded. Depending on their license agreement with SiFive, uthey may be able to modify it or not and they may be able to sub license their modified chip or not. This is all covered by the license chosen. That’s how the “open source” aspect is handled.
😂
Seems you are quite hostile towards RiscV or anything not ARM considering your benchmark is specifically not geared towards measuring the chips' compute performance but utilizes the cryptographic security accelerator in the pie3. Not to mention, testing SSL hash performance is a weird benchmark to judge chip performance since it is something rarely done compared to regular math, floating point or even IO operations. This is actually something I have noticed many times in your video. Using words such as "slower/worse than they are claiming" which makes them out to be lying while "THEY" never claimed ANY of the performance that YOU are presenting. I don't know if it is subconsciously or you are really that clueless but it happens too often, that your tests or comparisons don't actually demonstrate what you claim they are testing. I have mentioned the same biased testing in your Arm vs ESP32 test.
There seems to be a bias
He seems to forget that raspberry pi initially was just as bad
Following similar comments I uploaded a software only SHA256 benchmark to my GitHub repository. The RISC-V board is still the slowest. Also, you can't just ignore my threadtesttool results, just because it doesn't fit your narrative.
@stultus es The VisionFive 2 board isn't ready, even many die hard RISC-V fans admit that here in the comments section. Just because I say that and point out why that doesn't make me biased, it makes me honest.
A rpi killer? No.
A promising little board? Definitely yes.
After reading comments. It is not a Raspberry killer.
What about after watching the video?
That other Gary really drank some koolaid!!! hahaha I can see this boils down to whether an SBC buyer views it as a tool to get something done, or as a way to further their religious beliefs over the meaning of computing :D
Raspberry what? 😂
This lists for $130. All the 'raspberry pi killers' cost much more. While there is a real need for VisionFive and similar units, they 'compete' the way a pickup truck competes withy a semi. Very different niches. What competes withy the pi is OrangePi, Banana Pi, etc. Similar price and performance, but available.
As Gary shows, it's $55 with 2 GB RAM, $85 with 8 GB. I don't know where you got $130 from. How much is a Pi 4 now?
@@BruceHoult Price from their web page.
@@mikemullen5563 I don't see it. URL?
The price probably went up when it hit the market. The kickstarter (or whatever site it was) as pretty competitively priced.
@@monkev1199 I think some of these guys think that they can get that price, but the realities of production are grim.
No Wifi+Bluetooth = No Buy
I remember installing Gentoo on the HiFive Unmatched which uses the same cores as this, took at least a couple of days to install everything :D You definitly need patience if you want to be a early adopter
you chose to suffer with gentoo...
Well that's gentoo. With Fedora or Ubuntu you simply download a disk image and copy it to your SD card and you're up and running in 10 minutes.
@@happygofishing I wonder if something like qemu would help with generating a bootable image?
@@BruceHoult where is the fun in that :)
@@Mariuspersem your previous post sounded like a complaint, not fun.
I'm worried you'll go bonkers or have a heart attack of you try to read and respond to all the "I'm so excited for RISC-V" folks, Gary. Take care. CZcams comments are CZcams comments, even on your channel.