Having a big player as the "ARM of Risc-V" funded by VC was so toxic. It takes the oxygen out of the ecosystem.
The next step in open hardware is not having more proprietary silicon shops, it's streamlining the manufacturing process to make it look more like pooled PCB manufacturing, so that open collaborative groups can cheaply iterate their designs.
The problem is that the interesting bits in the VLSI space right now aren't in digital design.
Nobody needs another processor. Even an old-ass MIPS core is good enough.
The interesting bits are RF, ADC, DAC, SerDes, high efficiency DC-DC, low leakage designs, etc. RISC-V does not one iota of good for any of these things. Nor do these things need 5nm technologies--180nm or 250nm would be just fine though you'd probably have to use 120nm just because everything else probably has too little fab capacity left.
Which is a shame because that is precisely the path RISC-V needed to take to unseat ARM. It needed to be really good in the under 10 cents category with some decent analog peripherals such that it could expand upward and eventually eat ARM.
The under 10 cents category of microcontrollers is an absolute shitshow and has been for 10+ years. RISC-V could have brought unified tooling and architecture to that space. However, that isn't sexy. It would only let you ship a zillion chips and make reasonable profits. And that's just not VC compatible.
What's interesting to one person can be boring to another. I'm assuming you are on the analog side and find that stuff interesting. I'm on the digital side and was in a serdes team for 10 years and I'm glad to be gone.
You're absolutely right though that companies pick the process node based on the application. Everything I'm working on now is 5nm or smaller. Our customers DO want faster processors but we also have lots of serdes for 400+ gigabit networking and other stuff. I have friends doing DC power converters in 180nm. That's over 20 years old now but still useful for a lot of applications.
I worked at another startup and you're right about some things not being VC compatible. They want big investments and big payoffs. They aren't interested in investing in companies with lower risk but steady smaller profits.
It takes $30 million for masks in 5nm. EDA tools for physical design are over $1 million per instance and you may need 20 to 200 licenses for a big chip.
Total compensation for engineers is in the $150K to $400K range in the US.
This isn't getting cheaper.
PCB manufacturing is many orders of magnitude simpler and cheaper and easy to do at the hobbyist level.
Manufacturing processors is expensive. You need a big successful company to show that it's possible and profitable to invest.
Long term, yes I kind of agree, you run the risk of extend and extinguish if you get one dominant player, but you aren't going to get to the point of RISC V being successful by shunning big companies willing to invest. You just need to make sure they're more red hat or sun (pre buyout).
Not really. I'd say designing processors is expensive in engineering costs, and open designs are already pretty impressive and can be used as starting points. It's also expensive to optimize a design for a given process, which is necessary to get the best performance (probably by a factor of two?) from the design/process. So I'd expect the foundries to start offering optimized RISC-V cores to SoC designers as part of their offerings - especially as we reach the end of scaling.
If you have to spend £5million over 5 years to design something, you need to get funding until the point where you're actually selling products. If all the cost is in the actual manufacturing, you don't need the capital outlay early on.
If “healing” is the ultra wealthy holding on to their money or investing it in financial instruments instead of tech startups then god save us all. I don’t understand the animosity towards VCs that take risks on entrepreneurs like us.
Open source collaborative groups are not going to do shit in this space piddling away in their garages or community workspaces. It takes hundreds of millions of dollars to build a fab that can make competitive chips and beer money donations are not going to get us there.
SiFive is a fabless prop chip designer, they build no fabs. Yes building fabs is heavy industrial investment, and thus less amenable to open models. Drawing the masks is not.
Lot of VC work is regulatory arbitrage: how to steal the flowers from the public park without going to prison. That's why they are so proud of all these local sectoral monopolies they established while the lawmakers were asleep at the wheel, or bought, so the normal limits on profit in a market economy, through competition, are suspended.
Bulk of VC compensation is management fees, wisely based on the "head I win tail you lose" model and losses are often outsourced to ordinary folks via institutions like the Ontario Teachers' Pension Fund.
Money is just the scoring system of the economic game. If the rich play zero sum games with the points with each other it's harmless, and much better than malinvestment in Yachts or Web3 platforms where actual steel and engineering capacity is taken away from better use cases.
If it becomes a problem consensus can evolve to trim inactive data (say expiring unspent outputs after N blocks in UTXO chains, "move it or lose it" model) or explicit charging for storage per unit of size and time (decay some associated balance accordingly).
They use the occasion to get more commission, by encouraging renewals further in the future to "lock" the current price. Presumably they get the full 10 years worth of commission today when a .com is renewed for that long.
Sounds inferior to the "no cookies no banner" solution.
The GDPR does not mandate gratuitous and pointless personalised spying, which is the only case that requires consent. Normal operations (say a shop collecting payment details and shipping address to fulfil an order) do not require a consent banner.
Are these vendor kernels good enough to run some sort of hypervisor and pass through devices at the lowest possible level to (updatable) guest kernel(s)?
The idea here would be to run things like the TCP stack, USB from the lowest proxy-able level the in USB stack, etc in the guest kernel(s), as well as the entire application level, so as to reduce exposure to vendor kernel bugs and feature freeze leaving it only with minimal SOC-specific nitty gritty. For GPIO the vendor kernel could be used as a PRU of sort, passing messages to the guest kernel for actual processing.
Then the vendor kernel is just treated as a BIOS/blob getting in the way as little as practicable, it's very ugly but would allow using all these boards, also same method could possibly be used to recycle obsolete android phones.
nim with a bureaucracy would quickly turn into rust, so corporate heads can just adopt rust and let the world of niche opinionated tools be themselves.
and this is a project lead whose company has chosen to trust nim and written an ethereum client in it. Nothing is unopionated in this world, everyone makes their choices. And Status.im choices seem to have worked out.
the argument seems to be that redhat is/was playing fairly and then Oracle/AWS/Google/etc (which for legal reasons obviously cannot be named explicitly) came and started freeloading on redhat's work instead of "working together". bit of a tragedy of the commons/adverse selection issue within "capitalism" than a "capitalism vs community" thing.
maybe they should just grandfather RHEL (only support current releases for the 10 year period, no new LTS) and if clients want a security patched newer version of Linux, offer consultancy to help them switch to Oracle Linux. and then Larry will have to actually do the work lol.
maybe at a later point they can offer support for a bug-for-bug compatible rebuild of Oracle Linux :-)
Has the cost of lawyers and moderation been modelled? At $2/month spammers break even if they get like 5 clicks from all their posting on the platform, huge magnet for them, and content copypasters, if the platform gets some traction.
Sounds like the "is DJing an art form" debate. :o)
Unlike classic "hiring a musician", here it's practical to "hire" the (robot) musician 10000 times with a feedback loop between the model and the prompt writer, iterating and picking the best output(s)... which looks like a similar process to other exercises considered art forms.
Having a big player as the "ARM of Risc-V" funded by VC was so toxic. It takes the oxygen out of the ecosystem.
The next step in open hardware is not having more proprietary silicon shops, it's streamlining the manufacturing process to make it look more like pooled PCB manufacturing, so that open collaborative groups can cheaply iterate their designs.