RISC-V based single board computers are getting there

https://bret.dk/raspberry-pi-zero-vs-mangopi-mq-pro-benchmarks/

by bmlw • 4 years ago

282 143 4 years ago

sylware2• 4 years ago

huho! Finally a RISC-V 64bits SOC with tons of GPIOs! But for a keyboard controller power would be provided via the usb-c from the host, it seems the mangopi expect power on a dedicated usb-c connector. Am I wrong?

magicalhippo5• 4 years ago

According to the schematics[1], both HOST and OTG ports have VBUS directly connected to the same VCCIN net, so either can provide power as far as I can see.

[1] https://mangopi.cc/_media/mq-pro-sch-v12.pdf (page 3 top left)

sylware2• 4 years ago

Those are good omens. It is way overkill (wifi/bluetooth/gpu/video) for a keyboard controller, but at least, it does exists. If they follow raspberry on the same road, we may get a mangopi nano (hopefully with a with a risc-v 64bits cpu with a good bunch of GPIOs).

bullen4• 4 years ago

They need OpenGL ES 3, and apparently the Star64 will have that, remains to be seen how performant the GPU is:

https://www.imaginationtech.com/product/img-bxe-2-32/

Anything between VideoCore 6 and Jetson Nano would make Risc-V interesting!

smoldesu• 4 years ago

remains to be seen how performant the GPU is

I think this is going to be the real make-or-break thing about RISC-V: early tests have shown super impressive SIMD/vector benchmarks compared to ARM/x86, but whether or not that will make it into production is another question entirely. I've got high hopes for RISC-V, but it's acceleration/HPC workload performance is going to determine whether it topples ARM or becomes the next Itanium.

FullyFunctional9• 4 years ago

Unlike Itanium, RISC-V is an ISA and there are many many different implementations of the ISA. You cannot conclude anything about the former based on a few instances of the latter. Patience.

csdvrx2• 4 years ago

The Jetson nano being stuck on Ubuntu 18 is quite a shame...

synergy20• 4 years ago

and it is EOL...

bullen4• 4 years ago

I think they released the drivers so you could make your own JN compatible distro, not open-source, so I guess you are stuck with an old kernel...

snvzz2• 4 years ago

There's a PCIe slot, and thus there's nothing in the way of using a RDNA2 card from AMD, should the imatech GPU performance prove insufficient.

bullen4• 4 years ago

I guess, but for consumer friendly, low power and cost; it's better if it's all included.

st3fan2• 4 years ago

Right now the only metric I care about is: does it actually ship.

bmlw3• 4 years ago

Hah, it does! Though the 2 production runs they've done sold out pretty quickly. I think some 3rd party stores on AliExpress snapped some up and are reselling them, though if you follow their official twitter (@mangopi_sbc) they post their official page when there's more stock. They recently posted a picture of new boards being manufactured so there's hope!

st3fan2• 4 years ago

Yeah AliExpress seems to have these. Only $40+ for the 512MB model :-/

Cyberdog2• 4 years ago

In this regard RISC-V has a whole lot of catching up to do if it wants to be something other than a footnote in the story of the ARM takeover of the world.

I know the openness of the RISC-V platform is ideal to hardcore FOSS fans, but they're far outnumbered by those who will be satisfied with the cheap, fast, and plentiful ARM chips currently on the market, and the cheaper and faster ones sure to come.

Macha2• 4 years ago

I would not describe the raspberry pi as plentiful right now, especially if you want the 4gb/8gb variants. Mine was recently destroyed and I can't get a replacement

aarroyoc• 4 years ago

I bought the Mango Pi MQ-Pro a month ago and while at first the software support was missing (only Tina Linux was available), later some ISOs started to appear. Right now I'm using the Armbian headless image based on Ubuntu 22.04[0] which works almost perfect and allows me to work on porting things to RISC-V.

[0] https://bret.dk/armbian-on-the-mangopi-mq-pro/

bb883• 4 years ago

I would like more SBC options, rather than the solitary RPi right now. My sdcard shield popped off the RPi I have, and I had to resolder it back on. Two years ago, I would have just trashed it and bought another.

bmlw3• 4 years ago

You have so many options! I have Raspberry, Banana, Mango and Orange Pi boards sat here. NanoPi, Rock Pi, Radxa, Beaglebone etc all have alternatives available too.

guyomes• 4 years ago

Also Olinuxino from Olimex, chosen [1] to host the solar-powered website low tech magazine [2].

[1] https://homebrewserver.club/low-tech-website-howto.html#serv...

[2] https://solar.lowtechmagazine.com/

bb883• 4 years ago

Loading octoprint on an RPi is a seamless experience today. Are other boards as seamless as the RPi?

ephbit• 4 years ago

Also Odroid

rvz• 4 years ago

So many options and alternatives from all of them, yet the Raspberry Pi has by far the most software support, technical support, documentation and hardware compatibility and updates compared to the alternatives.

From the rest of the other boards it is just one kernel release (if you're lucky, 3 releases) and they have moved on to the latest SBC and dropped support and releases. The Raspberry Pi Foundation still continues to support older boards.

It just seems that with the many Raspberry Pi users, the documentation and the ecosystem built around it seems to be its success rather than the technical specs.

throwawaymaths2• 4 years ago

I suspect the high expense/unavailability of the Rpi is a result of supply chain issues and would affect other sbcs as well, am I wrong?

bb883• 4 years ago

MCU's are basically locked to a vendor currently. You choose a chip, make your design around it, and hope there won't be a shortage from that vendor, and it's high enough on the vendor's priority list to not hurt your product.

If there was a RiscV chip that met a particular standard (x voltage range, y clock speed, same supporting circuitry and software), then we'd have multiple chip vendors creating the same part, and hopefully it would turn into a jelly bean chip.

mrandish• 4 years ago

According to public statements by the execs at Rpi, the availability issues are due almost entirely to unexpected new demand exceeding their planned production volume (which they claim was quite large and scaled up from previous gens).

They report their suppliers have been delivering their parts orders on time and RPi is shipping the full volume of units they always intended for 2022. They are just getting bought up at a much higher rate than expected. They also say some of the issues in buying single units through online consumer retail channels is because RPi is prioritizing filling advance orders from volume integrators.

I have no idea the accuracy of what they're saying but fulfilling advance commitments from volume integrators first makes sense as those are long-term customers who are counting on the orders to ship their own products. Shorting those orders would be fairly disastrous as it flag RPi as an unreliable supplier reducing future design wins.

colonwqbang2• 4 years ago

There is so much available beyond Rpi. I recently got one of these:

https://www.pine64.org/rockpro64/

It has four lanes of PCI express so you can connect a big NVMe SSD or whatever. In my testing it seems to run OK on only passive cooling.

tryptophan• 4 years ago

But do any of them have software support? Most of these rpi knockoffs only ever release 1 kernel version and you are stuck on it forever.

sircastor2• 4 years ago

Maybe this is a feature. It would be a shame to trash a computer because it’s cardslot broke.

Incidentally, I’ve been trying to repurpose a Pi Zero W I blew out the USB bus on. A bunch of people in my local group expressed interest, but each I contacted in turn said “give it to the next guy”

mark_l_watson• 4 years ago

Great to see alternatives.

A tangent, but bear with me: after finishing the really good book The End of the World is Just Beginning, I think it makes a lot of sense to continue building cutting edge tech that requires international supply chains, BUT, also having locally manufactured tech good enough to power locally sourced computers, run tractors, etc. International supply chains have enriched many areas of the planet but to assume that they will last seems very risky. Always have a Plan B.

EDIT: call this Plan B Tech

javajosh• 4 years ago

To what extent do we need computers? They feel indispensable but we could go back to more labor intensive information systems. And that wouldnt be all bad.

dragonmost• 4 years ago

We need chips in everything. Cars could work without them but would be much less efficient and bad for the environnement. Lots of very important medical devices rely on chips or full computers. I would like to see the media industry roll back to a lower tech leaving only the big and very bias information go through. Imagine how hard it would be to keep them accountable.

Is it possible to live without computers? Yes, but you don't want to.

xyzzy123• 4 years ago

It's a good question, if we didn't have small microprocessors the alternatives would tend to be more material intensive (eg mechanical governors or clockwork) or be less efficient.

We'd lose the Internet and cellphones, have to go back to mechanical telephone exchanges.

I think we would miss CNC a lot, it's how the majority of production work gets done now in many industries, the manual machines are in the corner for one-offs.

The computer controlled machines also tend to be making parts or doing QC or measurement to support the non computerised ones. So sure your injection moulder or die cutter might not need too many chips but wait until the molds and tooling wear out.

Although who can send your factory orders anyway...

Payments, payroll, inventory, invoicing. Small words but huge implications.

We'd lose basically all capacity to print (billboards, t-shirts, books, office memos etc) , which seems bad. Like we had less digital ways to do all that stuff but they went away. It's not easy to go back.

Medical imaging gone except for maybe the x-ray.

Behind the scenes all kinds of process control would disappear which would require massive rework. We'd have to retool most industrial processes to get computers out of the control loops. We'd lose the electric grid until people figured out how to decomputerise it. Probably trains, traffic lights, airlines, cars, shipping would be impacted in a variety of fundamental ways.

The postal service would need to be re-architected (current reliance on parcel sorters and scanners).

We'd have to move back to analog tv, radio and media production workflows would change dramatically.

It's an interesting exercise to try and figure out what industries would hurt the most if chips disappeared tomorrow, I'm pretty sure it would be a catastrophe but it's not easy to follow it all through.

arise• 4 years ago

Farmers use GPS and other technologies to plant crops with inch-level accuracy while maximizing land utilization and minimizing fuel/fertilizer. Take that away and somebody's going to starve (or start a war to keep from starving). This might play out in Africa soon with the collapse of Ukrainian grain imports.

retrac• 4 years ago

Something happened in the 1950s - 1960s economically. [1] It is really as remarkable in its own right, as the industrial revolution had been. The rate of economic growth, slow but steady since the industrial revolution, started to accelerate.

It's impossible to really quantify this kind of thing, probably. And the causes were numerous (it's also an era of relative peace, for example). Still. I've always suspected much, maybe most of it, is due to to a mix of telecommunications, computer automation, and computer-based knowledge-amplification.

[1] https://ourworldindata.org/grapher/exports/GDP-per-capita-in...

jonjacky2• 4 years ago

You don't need chips for computers. Until the mid-1960s no computers used chips. The early PDP-8 models (introduced beginning in 1965) were desktop minicomputers made entirely with discrete transistors and magnetic core memories. They were used in lots of automated process-control applications.

There is some confusion in this question and many of the reponses here between any electronics (computation and control using tubes dates from WWII at least), transistors (which became widely available in the mid-1950s, all-transistor computers reached the market in 1959), small scale integrated circuits that contained (for example) several logic gates or a few flip-flops (which reached the market in the mid-late 1960s) which were used to build computers but also many simpler electronic control devices, and finally microprocessors which finally reached the market about 1975.

Any electronics that reached the market before the mid or late 1970s did not use microprocessors. For example, the first commercial CT scanner used a Data General Nova minicomputer, which used small-scale integrated circuit logic gates, but did not use a microprocessor. The original Pong arcade game used around 60 small-scale integrated circuit logic gates, but no microprocessor.

If we didn't have integrated circuits, but were stuck with 1959 discrete transistors, or we didn't have microprocessors, but were stuck with 1974 small-scale integrated circuits, we could still have a pretty hi-tech world. There would be more emphasis on optimized design of special-purpose devices - that original Pong game is an example.

hardwaresofton• 4 years ago

I'd agree, and I'd say that RISC-V's rise has much to do with this sentiment.

An open source patent/trade-agreement unencumbered architecture is of great interest to many countries for precisely the reason that they can build it locally yet harvest innovation from across the pond (without paying a foreign company).

nirui• 4 years ago

It might not be this easy. The dilemma is this: if the company produces good product, then they will probably gaining market beyond just regional.

The international supply chains of today's world was grown organically, driven by the market where inferior company and their products gets replaced by better (sometime just slightly) ones.

However, I do agree that a Plan B is required in many fields, but I guess you need huge amount of government subsidies to keep them running.

AnthonyMouse• 4 years ago

One way to do this is to make domestic production a requirement for government contracts. Then the government contracts ensure domestic production exists, and pays off a lot of their fixed costs, which may make them competitive in the global market. Or maybe competitors are still cheaper.

But more than one country can do this, which ensures supply chain diversity. And if one of them falters, the others may be more expensive, but they exist.

hulitu• 4 years ago

The international supply chains of today's world was grown organically, driven by the market where inferior company and their products gets replaced by better (sometime just slightly) ones.

I would say corruption played an important part, especially in Eastern europe. MS products are expensive for poor countries. Coca cola without "exclusive deals" would not have the same market share (Hello wrigley, hello lindt, hello intel). The shitty bananas in Europe have nothing to do with organic growing of the market.

mahrain• 4 years ago

To me it seems that Risc-V is China's 'plan-B', their systems are the first to use it and SBC's do exist already in China. I've even seen IoT cores from Chinese vendors moving to Risc-V, like the popular ESP32 from Shanghai company Espressif.

throwaway4good2• 4 years ago

AFAIK Allwinner is a Chinese company and the C906 core is an open source core made by Alibaba.

throwaway4good2• 4 years ago

It is plan B if you live in China. The West is not under treath to lose access to ARM.

I doubt it will come to fruition though; in ten years our world be more globalised than it is today. Trade wars, the pandemic and even the Ukraine war are short bumps on a big fundamental trend that is here to stay.

matheusmoreira3• 4 years ago

Plan B is also vital for computer freedom. In an ideal world, we would be able to make good enough chips at home, just like we can make software at home. Chip fabrication currently costs billions of dollars, they are therefore centralized operations vulnerable to regulation. Ability to make computers at home would preempt any attempt to regulate encryption, for example.

automatic6131• 4 years ago

Nice! And in this ideal world, we could also make steel tools in our decentralised garden furnaces. The steel industry costs the world billions of dollar currently, imagine if we could all make our own iron tools without paying for patents e.g. torx head screw drivers from our back yard! :^)

FullyFunctional9• 4 years ago

The C906 is very slow, but so is, it appears, the Pi Zero.

The good news is that there are a lot of faster options right around the corner. The Pine folks is working on releasing Star64 (quad SiFive FU740 1.5GHz) [1] and I know of at least two other RISC-V SBCs in the pipeline. I'm not quite ready to declare "2022 is the year of the RISC-V desktop" yet though.

[1] https://www.hackster.io/news/pine64-formally-unveils-the-sta...

Note, the FU740 is in-order dual-issue with a comfortable L2 cache.

Narishma• 4 years ago

The Pi Zero uses the same SoC as the original 2012 Raspberry Pi, and it was already long in the tooth even then so that's not really surprising.

moffkalast2• 4 years ago

Yeah it's already legacy hardware, it should be comparing against the Zero 2 if anything, the original Zero was comically underpowered.

But the speed is all but irrelevant, the issue with these alternative boards is always in software support. No point in using them even if they're twice as fast if I can't apt get anything and have to compile shit from source wasting 10 times as much time. Is there even an arch tag for riscv yet like armhf and arm64? I'd assume there is, but I can't find it and the support is likely to be abysmal this early on.

gjsman-1000• 4 years ago

You have to admire how long Broadcom has been manufacturing that chip considering it isn’t an automotive or defense design…

MrBuddyCasino• 4 years ago

Kind of weird to perform extensive benchmarks against an outdated version. The Pi Zero 2 is a lot faster!

bmlw3• 4 years ago

It is, though these 2 boards are the closest in terms of specifications "on paper" as they both offer a single 1GHz core, 512MB of RAM (unless you go for the 1GB MQ Pro) etc. Comparing it to the RPi Zero 2 felt a bit pointless in a standalone head-to-head piece. There'll be a separate post in a few weeks comparing all of the "zero" style boards from the various vendors that may give you what you're looking for!

KerrAvon2• 4 years ago

The actual metric most customers will care about is price. Pick a Pi variant at the same price as the RISC-V board, how does the performance compare?

candiddevmike• 4 years ago

What if I don't want a SBC and want the ability to upgrade the memory?

rbanffy• 4 years ago

The focus seems to currently be on SoC. Even ARM is not easy to get in a memory-upgradable form.

ffhhj4• 4 years ago

Is there some single board computer without RAM, but that allows addressing SSD space as such?

punnerud• 4 years ago

Wouldn’t that kill the SSD after some time?

HomeAssistant running SQLite on on a small cheap MicroSD card, kill it a couple of months.

Each memory cell have a lifespan of ≈ 1000 write cycles, if I remember correctly.

magicalhippo5• 4 years ago

I've had Home Assistant running on the same 32GB SD card for over 3 years now, and the DB is several GB thanks to several very chatty Z-Wave devices. Zero issues so far. So it's not a given it'll die in a couple of months.

ffhhj4• 4 years ago

An SSD hard disk instead of SD card?

Maybe such computer should use some sort of treadmill allocation.

magicalhippo5• 4 years ago

SD card and SSD's have a latency on the order of 500us, while even a Raspberry Pi 3 has a memory latency[1] of around 200ns. That's three orders of magnitude difference in latency, which would destroy performance.

In addition, memory bandwidth is still at least an order of magnitude higher.

[1] https://browser.geekbench.com/v4/cpu/11296737

ffhhj4• 4 years ago

Sure, but old computers managed to get things done with much lower resources. What really seems missing in old computers is memory.

numpad04• 4 years ago

I get it but no, not yet. No microSD card hacked to run DOSbox either.

t-3• 4 years ago

It would probably be far more practical and useful to run one with lots of RAM and no dedicated storage.

Havoc• 4 years ago

That seems very good for a reasonably early attempt

kragen9• 4 years ago

Some of the most interesting benchmarks here:

Dhrystone 2: 253.2 on the Mango Pi MQ Pro, 202 on the Raspberry Pi Zero. I don't know what this number is; presumably it's not Dhrystone MIPS (VAX MIPS), because that should be closer to 1000 for both chips, and it can't possibly be Dhrystones per second, because a VAX MIPS is 1757 Dhrystones per second, so they ought to be about 1.8 million Dhrystones per second. Reading https://github.com/kdlucas/byte-unixbench/blob/master/UnixBe... leaves me no wiser.

File Copy 1024: 124.4 on the MQ Pro, 86 on the Raspberry Pi Zero. The UnixBench README says this is measured in the number of characters that can be written, read, and copied in 10 seconds; if this were correct it would mean the MQ Pro were copying 12.4 bytes per second and the Raspberry Pi Zero W was copying 8.6 bytes per second. It seems inescapable that Bret's results are incorrect by several orders of magnitude here.

C copy backwards: 1197.4 MB/s (not MiB/s as I previously read incorrectly) on the MQ Pro, 157.2 on the Zero W. Something went wrong here.

Standard memcpy: 1200.9 MB/s on the MQ Pro, 424.8 on the Zero W.

Standard memset: 2650.6 MB/s on the MQ Pro, 1699.9 on the Zero W. This seems surprisingly slow; you'd think an 8x-unrolled loop of SD instructions on a 1GHz RISC-V would memset about 5 gigabytes per second, if it got one instruction per clock, which it normally ought to. The XuanTie C906 core in the AllWinner D1 C906 is 64-bit, which is potentially an advantage for this; the analogous code for ARM6 can only write 32 bits per instruction. (The ARM has an STM instruction that stores multiple registers, but it doesn't actually run faster.) I'm not sure what happened to the extra factor of 2 in performance. Similar remarks apply to the memcpy results above.

Amazon Basics 64GB MicroSD card: MQ Pro reads sequentially at 11.48 MB/s, writes sequentially at 10.77 MB/s; Zero W reads sequentially at 21.36 MB/s, writes sequentially at 19.6 MB/s. (I'm ignoring the random reads and writes because he doesn't specify the read and write size, yet he gives the results in MB/s instead of IOPS.) Note that these are six orders of magnitude faster than the 12.4 and 8.6 bytes per second given earlier.

Unfortunately Weber doesn't link to the benchmark code or document his compilation and execution environment. Presumably the memset and memcpy results are largely measuring the performance of the libc functions, for example, so reproducing them would require knowing if he's using glibc, musl, or a C library he wrote himself.

Mostly I feel like these benchmarking results are not well enough specified to be useful, which is a shame. I'd like to be able to use this kind of benchmark to predict the performance of a system within a factor of 5 or so, but these results are too irreproducible for that.

pca006132• 4 years ago

For memcpy and memset, they are usually memory bound, so it isn't surprising that it doesn't get to 1 instruction per clock. I wonder if this is due to slow memory instead.

kragen9• 4 years ago

Seems likely. Or maybe they don't have any cache and so it's spending half the memory bandwidth on instruction fetch? That can't be the Pi's excuse, though.

brucehoult4• 4 years ago

I did my own memcpy() benchmark on the first D1 board back in April 2021.

https://hoult.org/d1_memcpy.txt

I got about the same speed (1100 MB/s) for huge in-RAM copies. The in-cache speed peaked at 3771 MB/s for an 8 KB copy with custom code (shown on that page) using the D1's 128 bit vector registers, 2058 MB/s using the standard glibc function.

150 to 400 MB/s on the Pi Zero is very believable. The D1 is pretty strong RAM performance. The HiFive Unmatched was about 180 MB/s when I tested it. You need top-end DRAM controller IP and also a prefetch engine to get the good read speeds, and it's no surprise if an ARM11 doesn't have that. SiFive also doesn't have it on their self-produced chips. Hopefully Horse Creek has the good Intel IP for peripherals and RAM.

I have no idea what's going on with the Dhrystone. SiFive's in-order single-issue RISC-V cores get about 1.6 DMIPS/MHz on 32 bit cores, 1.7 on 64 bit cores. I haven't measured but I'd expect the C906 to also be around 1.7 i.e. 1700 VAX MIPS at 1 GHz.

kragen9• 4 years ago

This is great information, thank you! I hope SiFive's chips have improved too; I look forward to seeing the Horse Creek results.

1 GB/s or even 4 GB/s is still pretty anemic for single-computer laptops, but an order of magnitude improvement over Unmatched is great news.

what-imright• 4 years ago

I’ve bought the Mango Pi and a few other cheap SBC’s on RISC-V. The performance is slow 1Ghz single core but enough for small hobbies. I was hoping the different ISA would make shellcode incompatible.

But IMO RISC-V like many Berkeley products, is a pointless waste of resources to make a political point. There’s no compelling reason to recompile all our code to help out fabless design houses out of a license fee. And nobody save a privileged well educated and funded few are in need of a free ISA.

lloydatkinson• 4 years ago

I will remain cautious until any of these use mainstream kernel and kernel modules/drivers.

blippage• 4 years ago

I think the problem so far is that no RISC-V chip has offered any USP (Unique Selling Point), other than it's licence-free. For example, Raspberry Pis exploded in popularity because they were incredibly cheap and had a lot of support.

adgjlsfhk1• 4 years ago

being license free is a huge selling point though. arm makes over 1 billion usd in license fees per year. is riscv reaches anything like comparable maturity to arm, there are a lot of companies that would rather not pay their slice of that billion.

matthews2• 4 years ago

I've had a lot of fun with my StarFive VisionFive board, it has mainline Linux support so I don't need to worry about them disappearing.

It does lack a GPU though, instead it has some kind of AI processing unit that I have no idea how to use...

RISC-V based single board computers are getting there

Related Stories