Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

(200x) "Never bet against x86" -> (202x) "Never bet against ARM"


> (200x) "Never bet against x86" -> (202x) "Never bet against ARM"

Citation needed. There is only 1 workstation maker with ARM: Apple.


nvidia dgx spark?

also, system76 has one: https://system76.com/desktops/thelio-astra-a1.1-n1/configure

also, dell: https://www.dell.com/en-us/shop/desktop-computers/dell-pro-m...

HP is coming soon, will be called `ZGX Nano AI Station` apparently

Also lenovo: https://www.lenovo.com/us/en/p/workstations/thinkstation-p-s...


All of those use the same nVidia chip and the same nVidia provided Linux distro. Personally I'd be wary of spending $3500+ on a system that relies on nVidia support. They previously released an AI focused SBC, the Jetson Nano, that shipped with a custom distro based on the already aging Ubuntu 18.04, then abandoned support for it after a few years. All the dependencies required to run ML/AI stuff became outdated, and the hardware is now basically useless. It's possible to run mainline Linux on it, but you can't do anything with the GPU since that requires a special one-off driver that never got upstreamed.


> All of those use the same nVidia chip and the same nVidia provided Linux distro

The system76 one uses an ampere chip, but a discrete nvidia card; so not quite the same.

Generally I agree, I do not mix nvidia/linux. The point is that arm workstations are clearly being produced by many different vendors at this point.


Thanks for pointing that out, I must have skipped over the system76 link. You're right that besides Apple, nVidia and Ampere also make viable ARM workstation chips.


You can get ARM-based PCs and laptops. Thing is so far they all suck except for Apple right now.

I sure as hell hope that if Qualcomm even comes close to parity with other platforms they change their name because I'm going to have a hard time associated Snapdragon and their other products with anything that is quality. So far these are for desktop CPUs what

https://en.wikipedia.org/wiki/Trabant

is for cars. On the other hand, Intel is doing everything it can to keep the x86 platform from advancing which will let even the laggards catch up.


> You can get ARM-based PCs and laptops. Thing is so far they all suck except for Apple right now.

This is true, but there's a consolation -- they suck much much less than RISC-V kit.

I run a bunch of nearly 15 year old laptops. I am not a performance addict. But I have not laid fingers on a RISC-V box I'd be willing to spend a morning on.


That'll be true for a few months at best.

The first RVA23 and development boards are coming.

Tenstorrent announced Atlantis in the RISC-V Summit, based on Ascalon microarchitecture and due 2026H1.


> That'll be true for a few months at best.

The thing is, I have been in this business for 38 years now, and people have been telling me this for at least 3 or 4 years now. Including face-to-face from industry chip makers.

Really, honestly, right now, next month, OK, next quarter, but definitely by next year, really, the next generation is coming and it will bring the speed. The performance is coming, really soon.

And in fact, it's dire. It's crap.

Apple has a huge lead with its Arm chips and although most of the team that made that happen went off to other companies years ago now, nobody else has caught up. Qualcomm isn't there, but it's closer than anyone else. Amazon's server chips look promising but they're huge multicore things, not desktop/single-user devices at all, and they're very expensive in standalone desktops like from System 76.

Nobody so far can equal what Arm can do in skilled hands, and I bought my first desktop Arm computer in 1989. I've known that the potential was there for over a third of a century because I was there at the start.

I have Raspberry Pi 3, 4 and 5 devices here.

Look at this thing: https://system76.com/desktops/thelio-astra-a1.1-n1/configure

It is one hundred times the price of a Pi and can it deliver 100x the single thread performance? No, it cannot.

Can any money buy a RISC-V that can compare to the performance of a $60 Pi 5?

I don't think it can, and promises of "jam tomorrow" don't cut it.


I get your pessimism, but I prefer to look at the past for guiding posts.

From ratified spec to hardware on shelves, a 3 year delay has been consistent. This time, it should be no different.

I see no reason for RVA23 chips not to show up as promised next year, particularly with concrete announcements such as the one from Tenstorrent.

RVA23 is densely packed with everything that high performance micro-architectures need.


We'll see.

I approached John Ronco from SiFive who had just come off stage after a talk at the Ubuntu Summit in Rīga:

https://events.canonical.com/event/31/contributions/199/

... and asked him, telling him that I was the Linux reporter from The Register. He swore blind that I'd be blown away by the kit that SiFive would release the next year.

That was November 2023.

It didn't happen.

At the following Ubuntu Summit, I spoke to Nirav Patel right after he did this demo:

https://www.theregister.com/2024/11/18/riscv_framework_main_...

I tried that machine.

I tried to be charitable:

« It runs, it works, and even high-definition video playback is smooth, but this is not a powerhouse CPU. The machine we tried was sluggish and not very responsive. Even playing a single YouTube video, CPU utilization in the System Monitor was pretty high, and it was working hard when we tried moving the video window around. Minimizing it did reduce the CPU burden, though. Aside from video decoding, the machine felt less responsive than our old Raspberry Pi 3, to pick a more familiar example. In our opinion, RISC-V is not yet competitive with Arm in performance. »

We are another year on, and I just recently returned from the following Ubuntu Summit. I am still waiting.


>(JH7110) I tried...

A RVA20 compliant chip, released in 2023. I have one of these (VisionFive 2, received February 2023) with a couple hard disks and ZFS, serving as home NAS. Between RPi 3 and RPi 4 in CPU speed, similar I/O as RPi 4.

Obviously not a speed demon, but it was the very first mass-produced RISC-V SoC.

>He swore blind that I'd be blown away by the kit that SiFive would release the next year.

Probably the Intel thing that never actually happened, despite there been test chips which is what your acquaintance likely was able to play with.

SiFive released something else instead, earlier this year, as stopgap. Based on their P550 microarchitecture, it is dramatically faster than JH7110, but of no particular interest to anybody not directly involved in RISC-V, as it lacks vector and thus isn't RVA23 compliant.

In 2026, it'll be 2023+3, and many usefully fast RVA23 core IPs (Above Zen1 performance) are expected to finally show up in development boards, although only the one from Tenstorrent has had a concrete announcement so far.


I hope you are right. I like the idea of free and open source CPUs. I think the industry needs this.

But so far, the promises have not come true, and I am a skeptic at heart.

P.S. what "Intel thing"?


>P.S. what "Intel thing"?

Horse Creek, a platform developed by a collaboration between Intel and SiFive.

Something happened behind the scenes, and it got cancelled.

Instead, SiFive released some development board with Eswin EIC7700 recently, to fill the gap.


Huh. I missed the Intel thing, but then, I've been so underwhelmed by RISC-V that I don't track the sector. Thanks for the info!


There are Qualcomm laptops now I believe (at least that's what I heard when I was last working for them). NXP also made some boxes (I own a bunch of them). The server market is also growing with Ampere and Cavium (now Novell) which I have both.


Also AWS Graviton and Google Axion servers & VMs on those clouds


"Never bet against Apple"




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: