Okay, honestly for all that I wouldn't go for it myself, $1500 doesn't seem super bad for what it is. It's got 4 cores with a decent clock speed and 16GB of RAM, and it's a completely novel system that's not mass-market. The first RISC-V laptop isn't going to benefit from economies of scale; dev boards are barely starting to get there.
VisionFive2[0] w/JH7110, U74x4 @ 1.5GHz, 8GB, 2x GbE, M.2 slot, pi-like gpio header, is less than $100, and thus a much more cost efficient way to get started with RISC-V.
UPDATE: Bruce is right. This is more like twice as fast
WAS:
I signed up for the VisionFive2 myself, but in fairness, the C910 in this laptop is vastly more powerful than SiFive's U76 in the VisionFive2. I don't have the benchmark results handy, but it's on the order of 6X as fast (higher IPC and 66% higher frequency), so it's not really a just comparison.
$1500 for a limited run laptop is actually not bad deal. My main hesitation would be what level of support we can be expecting. I have zero interest in proprietary kernels and distros (Many Arm SBCs, like Hard Kernels ODROID line, has this problem and it's a pain to deal with).
No way it's 6x. The C910 is very similar to an ARM A72.
On my own primes benchmark (https://hoult.org/primes.txt) a Pi 4 at 1.5 GHz takes 11.2 to 12.6 seconds depending which ISA you use (12.1 for 64 bit) while a HiFive Unmatched at 1.5 GHz (same U74 cores as the VisionFive) takes 15.3 seconds.
So that's only about 25% on IPC, at least for integer codes in cache. The OoO A72 and C910 will get a bit more than that if there are a lot of cache misses.
Add in 2.5/1.5 in MHz and you get about 2x overall.
The real fun with RISC-V chips will start with the Intel "Horse Creek" which should be around RK3588 performance (similar to early Core i7?). A Horse Creek chip was shown (non-running) at a conference a couple of weeks ago and will probably be demo'd on a board, running, within the next month or so, and available on the next SiFive dev board maybe Q2 2023.
It appears I have mistaken C910 with a different core. I haven't found trustworthy benchmarks but T-head's own numbers of 6 DMIPS/MHz and 7 CM/MHz (Dhrystones and CoreMark - which leave much to be desired) compared to 4.27/3.32 DMIPS/MHz (??) and 5.75 CoreMark/MHz seems to agree with you. I have U74 Geekbench results [1] and would like to see the same from C910.
With that in mind, I agree that $1500 is a bit much.
[1] GB 4.5.0 40 (Crypt 7, Int 49, FP 27), but don't compare to Arm and x86 architectures as unlike them, the RISC-V code is missing a lot of optimization.
What version of U74? It has changed over time. Not only it got faster (both per clock, and changes to allow for clock increases), but it also gained extensions like B at some point.
This is particularly important with cryptography in mind; I would assume the crypt result is about that.
This laptop is certainly not a daily driver, and certainly not something you'd buy if you're concerned about lifespan, durability, or support. It's a mobile demonstration that the burgeoning community of risc-v developers can bring along with them to conferences and boardroom tables. It's (relatively) expensive and is functionally a reference board with a battery and screen attached.
The price, OEM, and support story is a non-issue for the intended market.
Do you or anyone happen to know if one of these U74 based boards (including the Star64 from Pine) feature encryption extensions? I tried looking but I couldn't find any mention on the CPU spec sheet.
I'm a firm believer that any machine packing personal data must have the capability to encrypt data at acceptable performance levels. And while I'm pretty sure I'm going to raise comments about hardware backdoor and state backed attackers I'm more worried about a thief or burglar having access to my personal data.
I have a couple of silent Rockpro64s running in a closet NAS and other applications. While those are not very performant they can handled 200-600MB/s encrypted data throughput and thus saturate a gigabit network.
i have a couple of these boards that use u74 (hifive unmatched) and afaik they don't have any official encryption extensions (RV64GC + bit manipulation extension) [0]
VisionFive 2 looks like a great board. I'll know in six weeks or so when my super early bird 4 GB one arrives (and a regular 8 GB in February).
However it doesn't have a case, screen, battery, charger, keyboard, trackpad ...
$1500 seems entirely reasonable for any laptop that is being made in quantities of 1000s (only 100 before New Year) not millions, regardless of what processor is in it.
I've certainly paid far more money for slower laptops in the past ... PowerBook 100, Duo 230, Wallstreet G3/266, 1.0 GHz G4 17" to name a few.
In raw CPU performance, it's probably similar to a Core 2 Duo MBP ($1999 for a 2.16 GHz in 2006). But it's got twice the cores, far more RAM, far more "disk", far better GPU.
The VisionFive 2 is more like one of the last Pentium IIIs or G4s in performance, per core.
Sweet. Imagine if you could put that in a laptop, like you would change motherboards in a PC. Then when the specs get dated, you can switch it out for the next best thing while reusing the original hardware.
That'd be nice, but having a standard form factor for swappable motherboards in laptops is a very separate problem than having cheap boards with RISC-V at all.
Sounds like progress. I don't need a RISC-V laptop badly enough to spend $1500 on this one, but I'm glad to see a shrinking delta between my desires and the market's offerings.
Battery weight depends on capacity, not rated life. Kind of nitpicky, but this really depends on how the power management works on this platform. It could just be really inefficient right now. (especially since we don't have systems really optimised for it yet)
Drivers are open source for both or are we stuck with vendor kernel images or some other garbage? How easy will it be to port Open/Net/FreeBSD or another OS?
> The Roma is available from Alibaba (opens in new tab) as a basic package for $1,499
The StarFive JH7110 that is going into the Star64 board [1] also has an Imagination Technologies GPU, there have been statements that there is going to be an open source driver for it.
My problem with this laptop isn't with it's pricing but with who's selling it. If it was Asus or some other reputable OEM, I think I would have bought it.
I've had extremely poor experiences with some Chinese OEMs in the past and I'm not risking 1500 with unproven OEMs unless I have to.
The small production run coupled with the price suggests that it's basically an evaluation board in a laptop form factor. As-is, it's probably mostly of use for people that want to hack on RISC-V (operating systems, ports, application development...) with a hardware target that can also act as a standalone host computer for said hacking on the go.
I imagine it's because Alibaba does not and will never have the purchasing agreements like HP/Dell/Apple/Lenovo do. I'd guess the secondary components that complete the laptop come at a premium, jacking up the retail price.
No, the point of RISC (and RISC-V) has nothing to do with cheap.
The _point_ of RISC-V is freedom [to use, implement, etc] and a rallying point (no need for everyone to re-implement the whole SW eco-system if that's not the main point of the implementation).
The specs are better than the first 15" Core 2 Duo MacBookPro for $1999 in 2006 (when a dollar was worth a lot more). CPU probably similar (but four cores not two) and everything else ... RAM, disk, GPU ... far better.
But this is of course less about trying to create a competitive laptop and more about China attempting to become technologically independent. Having the ability to mass produce something entirely within their own soil. Pardon the politics.
Interesting to see china develop its own processors. I don't know much about RISC-V or other processor, beside the openness what are the drawbacks and advantages ?
If you're really bullish on RISC-V as a developer, it may be a valid investment. Or if you already develop for RISC-V machines and it would be convenient to test binaries locally, instead of doing some kind of remote/emulated dev.
What about the Pinebook Pro or MNT Reform or new MacBooks? I know TrustZone is a thing, but heard it's not really the same and also not used in every SoC.
Also, The Asus C201 Chromebook has Libreboot support, it's not just old ThinkPads.
That isn't bad per-se. It's about what runs in that chip. It might simply be available to you to run whatever code you want, or just leave permanently off.
Neat to see a RISC V laptop but they must be smoking crack if they think I’m paying $1500 for the privilege of recompiling my operating system really, really slowly. And the 5k option is outright predatory. Someone needs to protect the vulnerable enthusiast market (you know who I’m referring to) from this kind of pricing. Ask a fair price. It’s an $800 laptop, maxed out.
It's a small-batch release of extremely-enthusiast hardware for a very niche segment of customers. They could charge $10,000 a pop and probably have no problem selling out.
There are cheap RISC-V dev boards if you want to use them, but this is mostly evidently not that. These people wanted to see if they could build a fairly decent 'all-in' RISC-V machine, and it looks like they did.
Talos makes it harder to backdoor the machine, because it’s radically open. An average RISC-V or ARM machine still contains various blobs and design compromises that make it harder to know that your code (or at least, code reviewed by you/the community at large) is the only thing running on the machine.
I wouldn’t recommend Talos to anyone who doesn’t have the need for it (and vast pockets to match), but I wish every ARM and RISC-V machine could be as open and trustable.
VisionFive2[0] w/JH7110, U74x4 @ 1.5GHz, 8GB, 2x GbE, M.2 slot, pi-like gpio header, is less than $100, and thus a much more cost efficient way to get started with RISC-V.
0. https://www.kickstarter.com/projects/starfive/visionfive-2
WAS: I signed up for the VisionFive2 myself, but in fairness, the C910 in this laptop is vastly more powerful than SiFive's U76 in the VisionFive2. I don't have the benchmark results handy, but it's on the order of 6X as fast (higher IPC and 66% higher frequency), so it's not really a just comparison.
$1500 for a limited run laptop is actually not bad deal. My main hesitation would be what level of support we can be expecting. I have zero interest in proprietary kernels and distros (Many Arm SBCs, like Hard Kernels ODROID line, has this problem and it's a pain to deal with).
On my own primes benchmark (https://hoult.org/primes.txt) a Pi 4 at 1.5 GHz takes 11.2 to 12.6 seconds depending which ISA you use (12.1 for 64 bit) while a HiFive Unmatched at 1.5 GHz (same U74 cores as the VisionFive) takes 15.3 seconds.
So that's only about 25% on IPC, at least for integer codes in cache. The OoO A72 and C910 will get a bit more than that if there are a lot of cache misses.
Add in 2.5/1.5 in MHz and you get about 2x overall.
The real fun with RISC-V chips will start with the Intel "Horse Creek" which should be around RK3588 performance (similar to early Core i7?). A Horse Creek chip was shown (non-running) at a conference a couple of weeks ago and will probably be demo'd on a board, running, within the next month or so, and available on the next SiFive dev board maybe Q2 2023.
With that in mind, I agree that $1500 is a bit much.
[1] GB 4.5.0 40 (Crypt 7, Int 49, FP 27), but don't compare to Arm and x86 architectures as unlike them, the RISC-V code is missing a lot of optimization.
What version of U74? It has changed over time. Not only it got faster (both per clock, and changes to allow for clock increases), but it also gained extensions like B at some point.
This is particularly important with cryptography in mind; I would assume the crypt result is about that.
The price, OEM, and support story is a non-issue for the intended market.
I'm a firm believer that any machine packing personal data must have the capability to encrypt data at acceptable performance levels. And while I'm pretty sure I'm going to raise comments about hardware backdoor and state backed attackers I'm more worried about a thief or burglar having access to my personal data.
I have a couple of silent Rockpro64s running in a closet NAS and other applications. While those are not very performant they can handled 200-600MB/s encrypted data throughput and thus saturate a gigabit network.
[0] https://www.sifive.com/cores/u74
Yes, they change over time. JH7110 uses the 21Q1 version. I believe. They have more extensions and are faster / clock, relative to unmatched.
However it doesn't have a case, screen, battery, charger, keyboard, trackpad ...
$1500 seems entirely reasonable for any laptop that is being made in quantities of 1000s (only 100 before New Year) not millions, regardless of what processor is in it.
I've certainly paid far more money for slower laptops in the past ... PowerBook 100, Duo 230, Wallstreet G3/266, 1.0 GHz G4 17" to name a few.
In raw CPU performance, it's probably similar to a Core 2 Duo MBP ($1999 for a 2.16 GHz in 2006). But it's got twice the cores, far more RAM, far more "disk", far better GPU.
The VisionFive 2 is more like one of the last Pentium IIIs or G4s in performance, per core.
What the world needs is something like ATX or ITX specs, but for laptops. Then anyone can make laptops/cases/boards/psus/otherparts.
Seems more like medium weight to me; hoping that's due to a large battery.
> Battery life is rated at 10 hours.
Oops, apparently not.
Still, cool to see RISC-V in a laptop.
4X RAM? And is the RAM socketed or soldered?
> 4 TOPS NPU and an Imagination Technologies GPU
Drivers are open source for both or are we stuck with vendor kernel images or some other garbage? How easy will it be to port Open/Net/FreeBSD or another OS?
> The Roma is available from Alibaba (opens in new tab) as a basic package for $1,499
I have a better idea: Buy a mnt reform.
[1] https://wiki.pine64.org/wiki/STAR64
I've had extremely poor experiences with some Chinese OEMs in the past and I'm not risking 1500 with unproven OEMs unless I have to.
Overpriced, or better, why is the price so high?
As Star64 is a thing, I am hopeful a Pinebook refresh will be RISC-V.
My point was that you can get a board built into a nice laptop in Shenzhen for a lot less than the cost of the one in the linked article.
It turns out, there is not much competition in the RISCV Laptop market.
But still with Windows(r)(tm)(fu) logo on the meta-key? :D
The _point_ of RISC-V is freedom [to use, implement, etc] and a rallying point (no need for everyone to re-implement the whole SW eco-system if that's not the main point of the implementation).
"Freedom" at any price though, I guess.
I wonder if the user will be able to change the distro, or the drivers will be somehow fix with the distro.
And thus, the newest laptop without an Intel ME equivalent is still around 14 years old.
Also, The Asus C201 Chromebook has Libreboot support, it's not just old ThinkPads.
That isn't bad per-se. It's about what runs in that chip. It might simply be available to you to run whatever code you want, or just leave permanently off.
There are cheap RISC-V dev boards if you want to use them, but this is mostly evidently not that. These people wanted to see if they could build a fairly decent 'all-in' RISC-V machine, and it looks like they did.
https://www.raptorcs.com/
I wouldn’t recommend Talos to anyone who doesn’t have the need for it (and vast pockets to match), but I wish every ARM and RISC-V machine could be as open and trustable.