>But after this first personal taste of RISC-V, I don't think OpenPOWER has much to worry about right now.
The D1 is far from the state of the art in RISC-V. It already was a low-end chip when it was released two years ago. What's good about it is the low cost of the modules.
Exactly! It's a Pi Zero / Pi CM kind of µarch at Pi CM clock speed and performance and (roughly) price.
This 1.0 GHz single-issue in-order CPU gets 37% of the performance of a 3-wide PowerPC G4 at 1.5 GHz (I think he probably means 1.42 GHz).
That seems like a pretty good result to me. At equal clock speeds it would be about 50% the speed.
He also says SiFive's U74 dual-issue in-order "gets by the G5, but by less than you'd think". The actual figure isn't given, and I think he still means G4 there, but that's also pretty good given the much simple µarch and similar clock speed.
And it's 270x faster than the 68030, at only 40 times the clock speed. I remember when I had a IIcx (my first personally-owned computer) and lusted after a cache-card equipped IIci. L2 cache card that is, the 68030 having L1 cache built into the chip.
The D1 isn't a powerhouse. We know that. Treat it like a Pi Zero and you'll be happy. Much faster things are on the way.
So what CPU would you have recommended for a machine like this? I hear "much faster things" all the time and this is exactly what I mean by the hype machine. What silicon is out there that people can get?
The Mac mini is a 1.5GHz G4. This was the last generation (the "secret clock bump" one).
The "much faster things" are coming at the expected rate.
Hardware simply takes many years to go from a company starting to work on a core, to having the core ready for licensing (at which point it works in an FPGA), to someone taking that and designing an SoC around it, to taping out the SoC, to getting back a few test chips (MPW, "shuttle run"), to fixing problems and taping out again, to volume production of chips, to yes another company designing a board around the chip, to getting that manufacturer, to it being available in a shop or on a web site.
It's no different in the ARM or x86 worlds, you just don't tend to hear about projects at such an early stage.
The D1 boards are the best available right now for under $30, or under $150 for that matter.
SiFive's U74 core is significantly better, being dual-issue with a limited out-of-order capability like the ARM A55 (it can issue two dependent instruction together, and execute the 2nd one in a later pipeline stage). The U74 is available (or was) in SiFive's FU-740 SoC on the HiFive Unmatched board. It's also in the StarFive JH7100 SoC which was on the 300 prototype "Starlight" boards the BeagleBoard Foundation gave away for free just over a year ago (I was lucky enough to get one) and now commercially available since December or so on the VisionFive V1 board for around $170. StarFive have been promising an improved JH7110 with four cores instead of two, and a GPU, for some time -- it was originally supposed to be out last September. Apparently it's still on the way.
The U74 core running at 1.4 or 1.5 GHz sits somewhere between a Pi 3 and Pi4 in performance.
There is presumably potential for someone to put a JH7100 SoC on a cheaper board e.g. a CM-style board.
Something else that is available right now as a chip and on a board is the Alibaba "ICE" SoC. This has three Xuantie C910 OoO cores, one with a high performance vector unit (2 vector pipes, each 1 cycle/instruction, vs the single 3-cycle pipe in the C906 in the D1). This is fully in Pi 4 territory. The "ICE EVB" board is available to buy. I have one. It's what the RISC-V version of Android is being developed on, and the board comes from the factory with an early version of Android in the onboard eMMC. It's $399, including a 7" touch screen.
Again, someone could potentially quite quickly take that SoC and put it on a cheaper board if they saw a market opportunity. The SoC exists, works, and is in production at some level.
In the "not here but coming soon" category is Intel's "Horse Creek" platform, which wraps SiFive P550 cores in Intel IP for caches and peripherals. According to SiFive, that SoC is in production at the fab right now, and boards (the replacement for the HiFive Unmatched) should be demonstrated later in the year.
Depending on what MHz that hits (it's being made as the first chip produced on the "Intel 4" process) that should be performance competitive with the Rockchip RK3588 ARM A76 boards that have come out in the last month or so. In x86 terms, that may be somewhere around Sandy Bridge/Ivy Bridge.
In the more distant future, Apple is suing a startup called Rivos for hiring away too many M1 engineers who have taken too many M1 secrets with them. Rivos have said they're making a high performance RISC-V chip. Rivos celebrated its 1 year anniversary a couple of days ago.
Very interestingly, I think, Apple is not asking for an injunction to stop Rivos, or for fixed damages, but for a royalty on their (eventual!) sales. That's of course going to be at least three years away before you can buy one. But there's no reason to think it won't be M1-class. They have the people, the experience, and the funding.
Yes, I'm fully aware that these things take time, and I even mentioned some of what you did in the article. As far as what's actually available, I appreciate the heads up on the ICE SoC. That's a new one to me and does sound like it would be a better fit for this application. But the rest of what you mentioned reads like the same sort of vague future promises that have swirled around RISC-V since the beginning.
I don't have anything personally against RISC-V, and given that OpenPOWER has all but ceded the low-end for the time being it might as well be RISC-V that fills it. What's around certainly seems a better fit for low-power and mobile.
But the point is what's around. Sure, you don't hear as much about ARM, x86 or other projects at earlier stages, but that's also because the hype around RISC-V has encouraged a lot of people to make a lot of promises early that don't go anywhere (I mentioned Micro Magic for a reason). That can't be good for the ecosystem. Cores are only as good as the ones that make it to tape-out, and real engineers ship.
If you read Micro Magic's press release it was clear (at least to me) that it was just a lab demonstration of their design tools and process technology and absolutely nothing like a product and not intended to be. If some of the press said something different that's not really Micro Magic's fault.
I'm not sure what you mean by "vague promises". As far as I can see (and I've been following them for years), the major players such as Andes, SiFive, Western Digital, Alibaba (T-Head) carry through with what they promise. Sometimes they don't meet the schedule they initially announce, but that happens with everything.
I sure wouldn't count Horse Creek as a "vague promise". Future chips from StarFive -- sure.
Companies normally announce a CPU core when it's ready for licensing by chip designers. That means the RTL design is complete and is running at 10's of KHz in software simulation (e.g. Verilator) and 10's of MHz in a FPGA (or faster in the ultra-expensive stuff the EDA houses sell). Linux is booting, Dhrystone and Coremark and SPEC have been run. The design works and the performance per MHz is known with basically 100% accuracy. The only remaining question is what MHz a customer will get on some silicon process and with how much effort at physical layout.
Well, and the question of whether they get any customers, and whether such customers make a product they announce, or something we never hear about because it is used internally in another product.
If you’re into the D1 get a $25 Mango Pi 1Gb from Alixexpress. They also have very nice minimalist metal cases if you search. Wifi and bluetooth onboard.
This Devterm product looks great largely because it copied the design of the Tandy. But, the reviews for the build quality are damning, the chip is underpowered for this task and some of the design choices make no sense. Who needs a thermal printer in 2022? Why are there two giant bezels on the sides that pop open the case?
But hey nice try. A few more revisions and I might buy one. I like that they’re trying something different, big points there. I wonder if they could put magnetic modules on the back like Motorola’s bricks (phoneblocks?) concept.
Same reason why I need a UART plug externally accessible
Its a cyberdeck, the more questionable the features the better, because then you can make up silly scenarios about saving the world with it, e.g. doing a direct audit of the network hardware in a datacenter, plug in your serial cable, download the running/stored configs, beam them up to some wifi accessable server, plus generate a paper backup at the same time
or, log into your deck at the start of the day, as you are reading the morning news, you get a printout of today's todo list and agenda
I bought one. The build is actually really good. The bezels are there for easy access for opening (no screws). The thermal printer is just a Retro gimmick which is the whole theme of the devterm. I disagree with LazyDev about the build, it's actually very good IMO, and fun to build.
It basically runs linux but I don't like it too much mainly because the UI isn't suited for the keyboard and the little mouse roller (which is terrible). A form factor like this really needs an entire GUI targetted specifically for it, instead this uses XFCE.
I would say if someone made a GUI specifically targetting the DevTerm it would be great. If you exclusively use a mouseless interface then I would say this is actually quite good in the sense that it's this portable retro thing you can bring to places. But compared to say a small linux ultrabook, the ultrabook is lightyears ahead of this.
Only get this if you want something with retro vibes.
First off the tech doesn't go that far back nobody is using a TRS anymore. And second it's more of a retro vibe not an attempt to recreate the exact tech. I get why someone like you can't understand this concept. It's mainly because you're too logical and intelligent. You're basically a savant with a mind similar to a super computer, you're too logical to fully understand stupid people like me.
Let me explain what's going on. This is just low IQ people like me having a sort of appreciation for retro vibes. High IQ people like you likely are too efficient and awesome to sink to this kind of base cave man behavior that has no logical point. Pay no head to us, we're beneath you.
And since when was RISC-V retro? Calling RISC-V fantasy might be accurate. I fail to see how everyone shifting to an open ISA benefits the people. Saves chip designers a dollar on licensing sure. But the RTL is still closed for most RISC-V that gets tape out. So you have no clue what it’s doing. “open”. Sure.
Have you heard of pixel art? It's a form of art that appreciates the retro aesthetic. I don't think it's your thing. Your brain is to big for such simple minded behavior.
Yes, but if its operation is invisible to you, and you can't install anything on it (without the use of a soldering gun and a JTAG style adapter), then it really an apt comparison...
The D1 is far from the state of the art in RISC-V. It already was a low-end chip when it was released two years ago. What's good about it is the low cost of the modules.