lets kickstart our own cpu! risc-v

[SOFTengCOMPelec]

How long do FPGAs last vs. ASICs?

In what sense do you mean, last ?

Physical device lifetime, technological obsolescence, firmware upgrades, reliabilty/MTBF etc ?

If you mean it losing its programming with age (flash memory lifetime limit), there is usually no such problem, as they are usually "ram" programmed ("ram" is part of the FPGA, and allows its configuration), where the ram is setup, each time it is powered up.
(Which itself, usually comes from some kind of flash or EE2 device, somewhere on the PCB).

 

[Maxima1]

It's all fun and games, until your software engineering needs are outsourced to India. Also, everyone and their mother knows that teachers are underpaid, so your argument is rather weak.

You know, that's just a fallacy? It's incredibly weak to appeal to the majority and leave it at that. I'm sure you would say more, but don't want to digress, though. XD

You were arguing on the premise that software engineering is already meeting supply & demand, but we know the private sector pays them well to attract smart people from other jobs. It would be even higher without the foreign influence on pay. Many know that foreigners drag down compensation in STEM while other jobs have been fully insulated, so is that fair? To the public, supply & demand isn't even a consideration, so your appeal to them doesn't make much sense to begin with (they generally wish for everyone to receive more). However, supply & demand is a rather simplistic look at it anyways. In Piketty's new book, he goes over many factors that result in workers receiving differently across countries and time.

Who is more likely to compensate less than the workers' marginal product: the government or the private sector? And which workers' marginal productivity is harder to determine? Both seem obvious. Look at private vs. public teachers. Or look at the military, for example. People say they're underpaid, but it's completely false. They have union-like benefits due to the politics. The basis is that they have a job risk, but a taxi driver has more risk (same risk as cops) of dying than the vast majority of enlisted personnel. The business board even went so far as to recommend cutting their $1.8 million pension and replacing it with a 401K type plan, which has already happened in the private sector.

Many people have argued teachers are overpaid. Bill Gates, for example, says the masters degrees inflates their salary (the degree makes no difference basically and neither does experience), and the teacher unions inflate the demand for them by advocating for smaller class sizes for middle/high school, but the studies show that it doesn't change the grades. (This has to be true because I've never had any teacher teach one-on-one, and it isn't feasible anyways.)

Although I've always thought current instruction is completely obsolete, and if you look around online, people are already asking this. You can give world class lectures on video with no need to waste more than two hours a day just computing to schools or listening to a lecture when you could cover more material in less time. There's a rockstar teacher in SK who is making loads through his videos. He has waaay more students than any teacher. So, in a way, it's like the software outsourcing to India, but it's isn't happening due to special interest groups. The same goes for tax accounting, correctional officers and marijuana/stiff sentences, etc.

The bubble's not popping tomorrow. Wage statistics aren't relevant right now.

They are relevant. They're not just throwing money away at any schmuck. They need a competitive package to attract smart people away from other jobs. If that part of the labor market pops in the future because the jobs aren't needed, then the market is simply being restructured. You can argue the same can be done with many service jobs or questionable regulations.

 

[Arachnotronic]

Software guys are a dime a dozen. Really, the entire hardware ecosystem is not conducive to the FOSS philosophy. Business stand to lose far too much by opening up.
Most software worth mentioning is also closed source, and where it is open source, it's generally because it doesn't negatively impact profits by doing so.

Hardware is hard. Software's a joke, relative to it. You don't need a college degree to program, but hardware practically requires it. Software guys are generally overpaid, and that's a bubble that will be popping soon. There's roughly 5x more software than hardware guys in the US, according to Wikipedia. It's easy to find software labor.

There's also a lot more expense involved in creating hardware. You need a lot of capital to get started in hardware, even if you have access to an open source architecture.

Bad programmers are a dime a dozen. Highly skilled software developers are worth their weight in gold.

Software guys are the difference between AMD Radeon GPUs and NVIDIA GeForces. Both have very good underlying hardware architecture, but NVIDIA invests far more on the software side of things.

 

[Headfoot]

Software guys are a dime a dozen. Really, the entire hardware ecosystem is not conducive to the FOSS philosophy. Business stand to lose far too much by opening up.
Most software worth mentioning is also closed source, and where it is open source, it's generally because it doesn't negatively impact profits by doing so.

Hardware is hard. Software's a joke, relative to it. You don't need a college degree to program, but hardware practically requires it. Software guys are generally overpaid, and that's a bubble that will be popping soon. There's roughly 5x more software than hardware guys in the US, according to Wikipedia. It's easy to find software labor.

There's also a lot more expense involved in creating hardware. You need a lot of capital to get started in hardware, even if you have access to an open source architecture.

I very much doubt it will pop soon. There are still more job openings in the US in software then there are graduates. Likely, this is due to the immigration restrictions and the fact that nearly no women go into software and women now attend college more than men.

 

[witeken]

Here's ExtremeTech's article on the subject:

RISC rides again: New RISC-V architecture hopes to battle ARM and x86 by being totally open source

 

[ShadowVVL]

So what is the aim with these, Will we be able to buy them and use it just like a pc with optional motherboards, discrete gpus and such?

 

[ashetos]

I had no idea Patterson was behind this. Perhaps with his influence, a large community of verilog writers or whatever will gather and we will have lots of cool designs that evolve like the linux kernel.

 

[Exophase]

The full ISA paper is here: http://riscv.org/download.html#tab_isaspec

It's good that there's a reasonable open alternative outside of SPARC v8 and they've put a lot of attention to detail, including some consideration to implementation efficiency. But I doubt this will get much traction outside of academia, much less the grandiose design plans for this to be a big deal everywhere. And honestly I'm not the impressed, the core instruction set is very weak and generic RISC and the extensions so far are either extremely niche (128-bit, QFP) or also very generic, perhaps with the exception of the atomic memory stuff which is on the questionable side. It's bordering on travesty that their promotional paper, just released recently, rags on ARM while completely omitting AArch64, an ISA that makes a lot more reasoned and fined tune instruction decisions.

And there's some really questionable decisions in there for the instructions they do, like encoding a dest register in the JAL instruction, wasting 5 bits of jump range for a feature almost no one will benefit from, then recommending a canonical link instruction is used anyway (for sake of return stack prediction). And the lack of compare with zero branch instructions (which is the most common type of branch before and complements the set instructions), meaning that those conditional branches also waste 5 bits of address space to include r0. The variable length encoding may be nice down the road and is good for some parts of its intended use today but it comes at a big cost, wasting like 79% of the 32-bit opcode space.

But I think the biggest issue is the total lack of SIMD as of now. They've reserved a space in the encoding for it and gave it a few words in the manual but that's it. And they have this to say about it, which is troubling to say the least:

In our opinion, packed-SIMD designs represent a reasonable design point when reusing existing wide datapath resources, but if significant additional resources are to be devoted to data-parallel execution then designs based on traditional vector architectures are a better choice.

This is just nuts, they'll never displace anyone in the spaces they think are interesting by abandoning modern SIMD in favor of traditional vector machines.