RISC-V CPU arrives on a tablet starting at $149 | Hacker News Reader

RISC-V CPU arrives on a tablet starting at $149

tomshardware.com

86 pointsdoener28 comments

Wonder if any of the RISC-V cpus are fast enough as a daily driver running Linux? (Or maybe in phones or servers one day I’d that’s the intention) - I was trying to read about it and saw that people’s analysis show some of them are 28% slower than a raspberry pi 4[1] and “the RISC-V U74 core had roughly 1.8 times lower performance per clock compared to the ARM Cortex-A53”[2]. I guess if the gap between non-open cpus and risc-v is too large and you can’t have the same OS experience or battery experience etc then it might be hard to be successful in mainstream devices? I have no idea how it works but would love to hear what people think; and it’s cool that we have an open processor architecture that seems relatively capable and that’s making its way into different devices!

[1] https://www.reddit.com/r/RISCV/comments/1c1ikfp/eswin_eic770...

[2] https://www.reddit.com/r/RISCV/comments/1coxdwm/whats_the_cu...

> Wonder if any of the RISC-V cpus are fast enough as a daily driver running Linux?

Short answer: Absolutely not, and you pretty much summarized it in your comment. RISC-V cpus are still at the performance level of Raspberry Pi 4, a $50 computer from five years ago.

> RISC-V cpus are still at the performance level of Raspberry Pi 4, a $50 computer from five years ago.

What do you mean "still"? Those are brand new RISC-V CPUs, or were a year ago when they came out.

The base RISC-V specification which the current cores implement (except the ones from the last couple of months) was not even formally frozen and published yet when the Pi 4 came out. Getting comparable RISC-V boards out in 4 years was a great effort. Note that the Pi 4 arrived eight years after the ARMv8-A spec was published.

That original spec doesn't contain a lot of important things such as SIMD which make a huge difference to those Geekbench scores. They are not comparing like to like.

By this time next year we'll have RISC-V boards performing better than the Pi 5, less than two years after it (possibly 1.5 years).

A year after that, RISC-V things currently well into their design pipeline will be somewhere between Zen 1 and Apple M1 performance.

The rate of progress is incredible. In the last 2-3 years many designers of current CPUs from Apple, Intel, AMD, Arm, Qualcomm have moved to RISC-V companies.

I mean maybe not many of us here, but my mom’s daily driver is a Chromebook. I SSH into my server using a Core 2 Dup laptop, that could be replaced.

When I was commuting 2 hours a day on the bus, my daily driver was an 8-inch iPad that I used with a logitec keyboard case tethered to my iPhone for internet, and would connect to my VSCode on my $10/mo droplet using Code-Server.

If you offload the computer to the cloud, and don't need to do anything graphically intense, it makes daily driving an underpowered tablet very easy.

I mean, I keep a 10 year old MacBook in my kitchen/dining room. If someone were still actually making a high-end Chromebook that would be fine. Unfortunately with Google's departure from doing that hardware, it's all pretty much lower-priced educational market stuff and if I really needed a lightweight travel device, I'd get a tablet (which I did recently) or a MacBook Air. MacBooks are low enough maintenance that a good high-res Chromebook wouldn't really be enough cheaper or lighter to really buy you much.

I don't think so, I run a CI on a VisionFive2 and it is terribly slow, takes ~8x compared to my Ryzen 3600 CPU. It's not helped by also having fairly limited IO speed on the NVMe side (board issue? unclear)

A 6 core OoO CPU with 6-wide µop issue running at 4.2 GHz is 8x faster than a 4 core in-order 2-wide CPU running at 1.5 GHz?

(6/4) * (6/2) * (4.2/1.5) = 12.6

Who would ever have guessed it might be 8x faster, without running the test?

In boards you can buy, there is spacemiT K1, the first RVA22+Vector1.0 SoC.

The boards are Banana Pi BPi-F3 (small SBC) and Milk-V Jupiter (mini-ITX).

The cores in there perform similar to -still usable- Intel Core 2, but there's eight of these cores, and presents a sizable improvement over the older VisionFive2's JH7110.

This review[0] by Explaining Computers shows how it is already usable in practice, using a launch day software image. Take note that newer board variants offer a more workable 16GB RAM.

Of course, having these boards means developers can now run RVA22+V in real hardware rather than simulators.

0. https://www.youtube.com/watch?v=GZGryhBnkV0

https://www.youtube.com/watch?v=YxtFctEsHy0

IMHO Jeff Geerling's recent video on RISC-V is a good update for those not tracking RISC-V closely, the links in the description are also useful.

  > The DC-ROMA RISC-V Pad II starts at $149 for the 4GB RAM version and is already available for pre-order on the DeepComputing store. 
  > If you get the 16GB RAM variant, the tab is compatible with the Android Open-Source Project (AOSP) Android 15 operating system.

Weird, isn't 4GB RAM enough for Android 15? Or is AOSP compatibility completely separate from Android compatibility?

Is it cheaper / faster to emulate RISC-V in ARM? I know that would be defeating the purpose of having open hardware, but it might be a useful stopgap measure (until hardware catches up) if it lets us instantly create a economically viable market of devices that run RISC-V software.

Wrong way round: there's tons of already compiled ARM software, so you'd want a Rosetta-style system for running it on RISC-V.

It isn't the first RISC-V tablet, there is also the Pine64 PineTab-V.

Sipeed had previously announced two tablets ("Lichee Pad 4A" and "Lichee Pad Max 4A"), but those are gone from their web page now.

Those would have fit Sipeed's CPU module: so they could have been upgraded to the same SoC as DC's tablet, or better, but at the expense of thickness: 10 mm.

Isn't DeepComputing also announced a RISC-V based mainboard for Framework Laptop? See: https://x.com/DeepComputingio/status/1826582426807927121. I was hoping this would enable me to get a Framework in China, since DeepComputing is a HongKong based company. Plus, a RISC-V based product-grade project is nice to try out, maybe.

You probably shouldn’t get the Framework with the RISC-V mainboard as your main machine, as that would be incomparably worse in terms of performance compared to the normal ones. That said, since mid-2023 you can reportedly get Frameworks in Taiwan—perhaps you could arrange for somebody to ferry it to you from there? Or is Chinese customs strict enough that that’d be impractical?

At the current stage, no one should be using a RISC-V machine as their main machine. That's why I mentioned Framework, because with Framework, if the investment was a failure, I can just get a X86 board and the rest is not a total lost.

However, even through Framework ships to Taiwan, importing product from Taiwan to China is not significantly easier than importing from the US. I still need to handle shipment forwarding and customs etc. Plus, my knowledge is, Framework discourages/bans shipment forwarding. So before I can get the product in an official and legitimate way, I'll just keep waiting.

The tablet mentioned in the main article while promising, it just can't interest me as a consumer. For it's starting price at $149, I can buy a secondhand ThinkPad X230t to run Ubuntu 24.04 and it's also a tablet (arguably the best tablet in the entire tablet history in the entire tablet universe so far).

Maybe the Chinese government should buy those tablets as work equipment to prevent their employees/personals from playing Microsoft Solitaire during work hours.

> This RISC-V-powered tablet is helping developers build more mobile apps for the open standard

Could you run something like Mobian (https://mobian.org/) or PureOS on it with convergent desktop apps?

If the tablet has both mainline kernel and uboot, it shouldn't be too hard for someone with embbeded Linux experience to port other distros over. However, there's quite a learning curve to doing it yourself.

Other than it being open source, what are the advantages of RISC-V over ARM?

While there are some open-source RISC-V cores and there are some easily available computers that use open-source RISC-V cores, most RISC-V cores are not open source.

For using an ARM core, you must obtain a license from Arm and you must pay a fee for every product that incorporates the licensed core. Moreover, the Arm license may have various restrictions on what you may do with your products that incorporate the ARM core.

With RISC-V, you do not have to pay for any license for RISC-V, even if you still have to either pay the development costs for your own RISC-V core or pay for a license to use a RISC-V core designed by another company. For some companies it may be more important that there are less restrictions, so they can design a custom variant of RISC-V, with proprietary extensions.

Except for the business model, there are no advantages of RISC-V over ARM.

There are also other open-source cores besides those that implement the RISC-V ISA, for example there are several open-source POWER cores.

If you are a silicon designer or a fab, RISCV costs nothing to license. People think this might cause more production-grade open hardware being introduced to the scene, but I don't think so. Anything that is worth selling to the actual customers will be closed source and may include security processors.

As a user no difference than any other architecture. You are completely bound by the quality/performance of the tooling, the compilers and the actual hardware

I think open source is really the big advantage. ARM is in the news every other day for some licensing shenanigans, possibly a lot of manufacturers are looking keenly at this hoping to one day get away from ARM.

There probably are not too many at this point.

What would be awesome is to have an open and standardized NN accelerator to go with RISC-V, but that is a dream.

https://semidynamics.com claims to be able to integrate tensor units to their design... I guess the standardized part is not there but if you can use standardized APIs...

It‘s really cool that RISC-V is becoming so real (with even Android and Ubuntu already being compatible), but it shifting processor production to chinese companies feels ... weird.

For RISC-V to succeed it needs open drivers, open specs. They can hide their IPs in the hardware and document the interface.

I'm ok with slower hardware if it is totally open. Blob free.

Reply on news.ycombinator.com