Libre-SOC Still Persevering To Be A Hybrid CPU/GPU That's 100% Open-Source

[lkcl]

If you look through their website, you'll see that this is a laughable effort, run by people who have no understanding of how to actually build a modern SoC.

then you can join the mailing list - after accepting the conditions of the LibreSOC Charter, which require that you behave in an ethical and respectful fashion towards fellow human beings - and help provide the team with the benefit of your knowledge and advice on how to actually build a modern SoC.

does that sound reasonable?

if every entrepreneurial team set a goal that they only knew in advance how to achieve, do you think humanity would be where it is today?

we make brave decisions to set out to achieve something *even though* we do not know how to achieve it. that's how innovation works! you have the condidence and the strength of spirit to face obstacles as they come up, to work around them and often have to make brave decisions to abandon a particular path that's just not working out, and go back to the drawing board and try again.

it's that repeated systematic effort to reach the goal that, ultimately, makes success possible. it's certainly not achieved by waiting until you know everything there is to know! certainly not on something as complex as a 3D CPU / GPU / VPU.

which reminds me, to take the opportunity to thank both the OpenPOWER Foundation and XDC2020 for the opportunity to present at their respective conferences (and to Phoronix for this article). the conversations that i had online with the participants have been exceptionally rewarding and engaging: i was very surprised to find that "online" conferences could be actually better than "real-life" ones in that regard.

[brent]

People will be waiting for this as long as they've been waiting for their EOMA68 boards already...

[lkcl]

I think that part of the problem is that Risc-V has NDAs. Doesn't that seem a little non-transparent to you?

a bit of background, here: it's not _quite_ an NDA, i forget the exact details, however it was sufficiently incompatible with our transparency and independent audit requirements that we simply could not sign.

RISC-V's Foundation was established with the fundamental assumptions about potential participants:

* large well-funded commercial businesses
* well-funded universities
* poor, not very knowledgeable "open source" developers with zero practical resources

(of course, this latter assumption has now been completely blown out of the water with the Google-Skywater collaboration, which will first give open source hardware developers access to 130nm, then later 45nm and finally 28nm)

when we first pointed out that we had a business objective that *required* us to be entirely transparent - not just Libre Licensed HDL but the *development* processes as well, everyone laughed at us behind our backs, which is extraordinarily disrespectful.

and now with the Google-Skywater collaboration in place which empowers Libre/Open Hardware Developers to create fantastic ASICs that would normally need USD $5m and above to develop, that disrespectful attitude is costing them.


by complete contrast our interaction with the people behind the OpenPOWER Initiative has been absolutely amazing. they've been positive, understanding, respectful and listened to what we have to say. that's not to say that they're not extremely busy people, it's not to say that they're going to "jump immediately into action", because what we have been asking them about - inclusion of Libre Businesses and full transparency - takes time to negotiate and work its way through the "system" if you know what i mean.

and that's absolutely fine: we understand and appreciate how big companies like IBM work. they take the time to "get things right" because they take their responsibilities - to their customers - very seriously.

[lkcl]

RISC-V is hardly useful to the "general community" so by that measure the complete project is a failure.

yyyeah that's not entiiirely fair: i mean, companies like Trinamic (who were one of the first to do an embedded RV32 product, with one of their Stepper ICs), and Western Digital (who use a customised version of RV32 in their SSDs, HDDs and USB NAND products), RISC-V has been *extraordinarily* successful. not just in saving those companies product development and maintenance costs, but in cutting out at least USD $1.50 from the final product costs because they *no longer have to pay ARM royalties*!

and that's really, really significant.

but the key discerning factor behind the success of these examples is that they went the "hard fork" route. they took a snapshot of the RISC-V toolchain and maintain it internally, for proprietary use *only*. i.e. they are taking advantage of the cost savings of the "Embedded" Platform to reduce their product costs.

they are *not* - repeat - *not* - making any of their proprietary use of RISC-V public. Western Digital make a very specific point of saying, at the beginning of any of their talks, that they can talk *about* the modifications they made to RISC-V (in general), but that they are *never* going to provide access to the compiler, or provide end-users with direct access to be able to compile up their own firmware for any Western Digital SSD or HDD products.

so in _that_ regard, yes, RISC-V is still a failure because there simply are no *general-purpose* products - sold in 10 to 100 million units - no laptops, desktops, smartphones - where you can download an RV64GC toolchain and start compiling up and running your own games and apps like you can for Intel / AMD products and more recently even ARM-based smartphones.

i've even heard of people installing full GNU and LLVM developer environments on their Samsung smartphones and building apps native on them! hilarious

[ermo]

@lkcl

Out of nothing but curiousity: How is it that you end up capitalising Trademark Law and IBM, but not the first character in new paragraphs?

From my perspective, text is a shared medium of communicating ideas around which some best practices have emerged. Whenever I see someone disregard these practices (which typically only ends up making text slightly harder to read) I always wonder what their goal is. What is your goal?

[lkcl]

Sounds like a lot of work... Meanwhile, my Raspberry Pi 4 is doing the job pretty well. Maybe it's not open source hardware wise, but sheesh, it's so cheap and works so well, how can I pass it up?

Even if these guys were successful at creating a performant 100% open source hardware design, they've still got to build the things. Like it or not, Broadcom and the Raspberry Pi Foundation have economies of scale that will be challenging to beat.

yeah, this is a really good point. i'm still thinking about it, and would welcome anyone with any good ideas here. in the meantime, luckily this is a "blue sky" research project, funded by NLnet who believe in the idea, so we are good for 18 months at least to keep working

I don't see how these guys will be successful, but good luck to them!

that's really appreciated and encouraging, ed, thank you.

[lkcl]

@lkcl

Out of nothing but curiousity: How is it that you end up capitalising Trademark Law and IBM, but not the first character in new paragraphs?

25 years ago i got such bad RSI (known as carpal tunnel in the U.S.) that i had to minimise typing. it got so bad that one day i couldn't get into my house because i couldn't turn the key in the lock.

like the "pavlov dog", if it actually physically hurts to stretch your fingers just to reach a shift key, pretty soon you stop doing it. however when it comes to proper nouns, sometimes i find that the respect that i have for such words "over-rides" the physical pain that it causes me to type the word.

From my perspective, text is a shared medium of communicating ideas around which some best practices have emerged. Whenever I see someone disregard these practices (which typically only ends up making text slightly harder to read) I always wonder what their goal is.

it turns out that a study was done which showed that lower-case letters, being more predominant, are easier on the eyes and the brain. when i learned of this i continued the habit. however when writing articles (or books) i will do a careful review and make sure that the training i received at school is indeed properly respected.

What is your goal?

to bring ethically-developed technology and products to ordinary people.

[lkcl]

People will be waiting for this as long as they've been waiting for their EOMA68 boards already...

luckily and thanks to NLnet we have adequate funding, and because it is HDL, we don't have (apart from the 180nm ASIC costs) any high "line items". no repeated USD 5,000 PCB re-spin costs for example: no need to buy USD 30,000 worth of components, for example.

you've no doubt read the updates and seen the challenges associated with having to source extremely unusual "micro" category components. if the goals of EOMA68 were "to develop yet another Pi-style SBC" i would have been able to deliver that within about 4 months flat.

[Dawn]

(first OpenPOWER ASIC since IBM’s POWER9, 10 years ago)

Fractally wrong. OpenPower didn't exist ten years ago, and Power9 came out in 2018, not 2010. (Power7 launched in 2010, and P8 was in development; P9 wouldn't even really be on the radar, except possibly in very very very very early definition stages.)

[Min1123]

lkcl I've been following this (as a GPU) for a while, as I've been following the EOMA68 for a while. You mentioned once using multiple digital lower speed memory to decrease the issues associated with trace length in modern DDRx RAM implementations. Are you still considering this?
Also, I didn't see, I was kind of hoping that the SoC here would be multi-core with one or more cores being dedicated to the rendering as a GPU and the others being the CPU, potentially with an MMU that would allow memory bank swapping so that the CPU could load up the GPU RAM and swap, or vice-versa, to support using GPGPU-style operations with a speedup.