Introducing a new RISC-V Mainboard from DeepComputing

Framework Team Blog
82

To my mind, the net benefit is simple and obvious: competition.

The x86 sphere is kept relatively honest by the fact that AMD needs to have a license from Intel for x86, and Intel needs a license from AMD for x86_64.

In desktop ARM (serving as the proxy for “Desktop RISC” since, today, there is no meaningful alternative), there is only the kinds of competition that ARM Holdings finds useful to their bottom line. And, besides, there’s plentiful fragmentation there too with a bazillion cores up for licensing (that may or may not have such simple things as division as part of the ISA implemented) and even whole ISA-only licenses where companies go full RISC-V style and implement from-scratch - doesn’t stop the Macbooks from being successful in the market. And looks like Qualcomm is making inroads via Microsoft.

RISC-V has the risk that specific implementations may differ dramatically in which instructions are implemented, making binary compatibility problematic. (Though less problematic where rebuilding is easy, eg FOSS.) But this is already an issue ARM is dealing with, and it is also an issue that x86 is dealing with. As an example, some projects, like simdutf, will be providing separate inline assembly depending on which exact x86 CPU you have - does it support AVX512 for example? And, after I found a bug caused by the fact that the OS (in this case OpenBSD/amd64) also needs to support AVX512 for it to function, it also has to check if the OS support AVX512 before selecting in that codepath. (Sans the fix, OpenBSD and a few other operating systems would flat-out crash any software using simdutf, for example nodejs, if executing on Intel 11th Gen or later.)

Due to this, Linux distributions (and proprietary OS vendors) already typically does things like flat-out ignore anything except the most common ISA extensions. It’s taking Arch years to reach a decision on whether to support x86_64_v3, because there’s a lot of stuff that can break if you default to ISA extensions that are as recent as 2013… (We’re soon going to have some Gentoo and/or FreeBSD poudriere fans in the room. :stuck_out_tongue: )

So as far as I see: all the problems RISC-V faces are actually the same as what both ARM and x86 is successfully navigating. There’s no reason to expect RISC-V to do worse (and also not better). So at the end, RISC-V can achieve one important thing: give ARM a competitor on the Desktop RISC ISA market.

And RISC-V doesn’t even have to be a huge player for this to be a benefit. You don’t have to use RISC-V chips to benefit from that.

7 Likes
83

Oh, JH7110, sounds tasty with the messed up bootROM and the half-broken MMU. Well, I mean, messing with it was fun on the StarFive2, so I’m not too afraid of it, but… Please tell me there is an SPI flash/EEPROM there for the firmware. If it loads the firmware from eMMC, that WILL be a massive pain.

4 Likes
84

My body is incredibly ready

85

Real world, instead of theoretical, counter-example: OpenWrt. “Same” (ARM, for example) ISA, many different platforms/SoCs. And yet the community/ecosystem is thriving.

Editing to add: Within OpenWrt, platforms that never supported (Hi there, Broadcom) or dropped support for (looking at you Marvell) open source have fallen in relative decline, you might say. But that’s just open source, if the vendor doesn’t play nice, move on to the next one. RISC-V lowering the bar for new SoCs/platforms is good in that way.

Another important point:

They are sometimes hobby or side projects maintained by one person in their free time.

That’s the “romantic” depiction of open source, and still true for smaller, userspace projects. But for system-dependent things (hardware support in the kernel, important infrastructure) it’s very much professionals, on the payroll of the hardware vendor, doing the open source work. Look around these forums for Mario from AMD as one example.

2 Likes
86

I can speak to this as he’s given plenty of advice and direction when solving issues…awesome source.

1 Like
87

Well,RiSC-V is just an open standard not open source. So the hardware is not open hardware by default. Though nothing stops a CPU manufakturer to make open hardware out of RISC-V.

It’s just removes the control that (Intel, AMD and ARM) has of CPU market.
Cool anyhow and risc-v is the future.

88

Can confirm. I am typing this on my FreeBSD install with both the base and packages built in poudriere optimized for the znver4 in my framework laptop. If only poudriere was ported to Arch Linux…

1 Like
89

Don’t worry. Even with RISC-V, atomakers will botch it somehow.

90

I would imagie that they reached out to framework with an initial prototype, instead of the other way aroud.

91

I think it is a bit of both. As far as I know, RISC-V does not license core designs (what get’s printed to the silicon), leading to the problem of potentially incompatible (even if unintentionally) core designs. We will also have the issue of how to talk to different controllers, where they are, and more. With X86(_64), that is (mostly) standardized making the development of device trees easy. RISC-V will have some of the same challenges ARM has, even if the scale is different.

TLDR; RISC-v will remove some nails, but how many more will it put in.

1 Like
92

RISC-V >> ARM, hopefully Framework in the future actually invests into RISC-V themselves like Apple does to ARM. (In the far future, of course, since it would take boat loads of money :P)

1 Like
93

I’m very happy in my OpenBSD-land, but Poudriere is one of those things that do call for me like a siren a little now and then. Might soon be time to add another NVMe drive to the collection and get a separate FreeBSD install going just to play with Poudriere and Jails.

1 Like
94

Under that light RISC-V seems like the “open source” of ISAs, but how is it going to look in reality, if microsoft and apple go with their vendor ARM, which manufacturer is going to supply RISC-V?

I think you’re ruling out that Intel and AMD may produce RISC V based chips, eventually.

If you consider this possibility, in a not too distant future RISC V and ARM could become the two main architectures around, with x86-64 becoming more and more the legacy option (consider also that emulation will keep improving).

In this scenario, of course the answer to your question would be “basically every OEM shipping Intel/AMD CPUs today”.

95

AMD Microblaze V
And, of course, AMD is a RISC-V member after the acquisition of Xilinx.

So AMD already has RISC-V products after going full spanish acquisition. Not desktop-oriented, of course, but… For now?

5 Likes
96

Realistically, probably for a long time. AMD profits immensely off of holding a duopoly on x86.

97

I am so excited for this. I have a DIY Framework 16 ordered, but once the RiscV board comes out I would love to just get the board and have an enclosure for it to tinker with on the side. I only have Windows 11 running as a virtual machine for games and occasional Win only firmware updates on some of my random devices.
My current laptop is an aging 4 year old System76 Lemur Pro, but I miss having a numpad when I’m on the go. I love that with the FW16 I will have the option for a numpad or macro pad, or just blanks. I may or may not have ordered all the options to play with when it comes in.

98

AMD wouldn’t have to stop making x86_64 chips to make a desktop RISC-V or ARM chip, obviously.

It might be too big a drain on resources they feel would be better spent in x86-land, of course, but the value proposition in desktop RISC is a lot different now than it was just 5 years ago, and the market reality is changing. Case in point how we might finally see non-crap Windows ARM machines (though reports on Qualcomm’s latest are mixed), and Apple has proven a lot of the value, especially in the laptop space.

That’s RISC chips taking market share from x86_64. Having a duopoly doesn’t help when that duopoly starts to shrink.

If forced to bet, I’d agree with you. But there’s been remarkably fast shifts before, even recently. I suspect the decider that may motivate them (or Intel, for that matter) would be if Qualcomm starts to make serious gains in the Windows laptop market. If that does not happen, why divert resources to a different ISA? But if it does happen, I’m sure they’ll at least look at it and game it out as a possibility. (And then it’ll take 5 to 10 to develop and mature, ofc.)

99

Accurate on the performance side of things, but I believe a SiFive based processor stole the crown from the Apple M1 for highest performance per watt ever.

1 Like
100

That is a very interesting, thanks for sharing! I’m definitely going to check that out.