[ale] Really cool new hardware

[jonhall80 at comcast.net]

Most of what you say is true, but like everything else it can not be taken in isolation.

I will start with the easiest:

>Supply voltage has nothing to do with the CPU core, and >everything to do with the physical implementation (process node >of the ASICs, power supply design, I/O voltage requirements, >etc etc)

All true, but if the core is not designed to run at the lower voltage than turning up the clock boosts the power input and the core runs hotter.  To say voltage has "nothing to do with the CPU core" is a bit cavalier.

>Yep; the core ISA is royalty-free, and anyone can build a core >that implements it.   Which is great, if you can afford to >develop your own core (and very few can; you have to have some >pretty serious volume before the NRE involved in going your own >way eclipses the Arm licensing fees.. which is the secret >behind Arm's success.   That said, the continued consolidation >in the semiconductor industry has changed this dynamic as per->company volumes have grown massively)

I think the "secret behind ARM's success" is multi-fold.  They were willing to license out their designs, and in fact depended on it for survival.   So the "secret" was not only their technology but their business practices and the infrastructure they built behind it.   I tip my hat to "ARM University" which creates programs for universities to do research with ARM architecture.

Also the fact that ARM does not compete with the ARM chip licensees.   There is no "ARM ARM" that you can buy on the marketplace as you can buy an "Intel x86".

RISC-V has gone a level deeper, by loosening up on the licensing model.

>HPC

Your comments here are also true, but as memory models change you may find that the cores/GigaByte-power also changes.  As memory is redesigned to use less and less power then reducing the amount of power that the cores use is more significant.

Every little bit helps.

>Proprietary EDA tools...build it yourself or license it from >someone else

There is at least a third path, and that is the one provided by vendors such as SiFive where they develop the tools and allow you to use them (granted, for a fee) but the output is yours and they also have the expertise to help you do it.

>"Extended capabilities" are not something you want at the ISA >level for a general-purpose architecture, because then you'll >end up with incompatibilites at both the binary and source-code >level..)

I did not say that "Extended capabilities" were at the ISA level, but the openness of the architecture and the process could allow a tighter integration of extended capabilities than you would ever get from Intel, AMD or any of the "chip makers", perhaps even ARM (unless you were willing to pay a lot just to get their attention).

Finally, as you keep (rightly) pointing out that chips have yet to appear in the high(er) performance end, the same could be (and is) said of ARM architecture.   However, the world has been hit heavy with the performance of the new Apple chip, which can indicate that the performance is possible when the market demand brings about the investment.

As someone who was active in bringing out the DEC Alpha chip over a quarter century ago (for years the world's fastest 64-bit chip) from a company (DEC) that designed its own chips, manufactured them in its own FAB (along with VAX chips, ARM chips, Sparc chips, and MIPs chips) I am well aware of the need for investment at every level, and it does little good to have the world's fastest 64-bit chip if there are few applications that the customer wants to run on it.

But I am also very aware of the cost of starting a business when the first thing you have to do is talk to lawyers and license managers.

Regards,

maddog

> On 01/19/2021 1:53 PM Solomon Peachy <pizza at shaftnet.org> wrote:
> 
>  
> On Tue, Jan 19, 2021 at 12:45:20PM -0500, jonhall80 at comcast.net wrote:
> > I disagree.  I think the interest is across the board, for several reasons.
> 
> Yes, there is interest "across the board" but as I am writing this, open 
> "high-end RISC-V" cores do not really exist; all but a handful [1] of 
> the freely available designs are microcontroller-class.  The rest are 
> either proprietary in-house designs (which may or may not have tweaks to 
> the base ISA) or available to license in a manner similar to Arm's 
> offerings.
> 
> [1] BOOM and the Sodor cores out of UC Berkley, intended for educational 
>     purposes.
> 
> > The licensing and start-up costs were fairly high.  This prevented 
> > smaller companies and institutions getting into the marketplace of 
> > producing compatible ARM chips with extended capabilities.
> 
> Startup costs are the same either way; if anything they are lower for 
> Arm due to the much more mature ecosystem around it.
> 
> (and "Extended capabilities" are not something you want at the ISA level 
>  for a general-purpose architecture, because then you'll end up with 
>  incompatibilites at both the binary and source-code level..)
> 
> > Things are different than when Intel and AMD got started.  You can 
> > have easy access to FAB plants, good design tools, FOSS software and 
> > crowd-funding.  But when you are working on a shoestring, paying out a 
> > half-million to million dollars in licensing fees to get started is 
> > daunting.  A million bucks is five good engineers for a year.
> 
> To get a custom SoC built you're looking at about a million bucks a year 
> in proprietary EDA tools, plus salaries on top of that.  CPU core 
> licensing fees are a drop in the overall budget -- you also have to pay 
> for the rest of the SoC, and either you build it yourself or license it 
> from someone else, before you ever get to implementing your "special 
> sauce".  It's the classic "time to market vs money" tradeoff.
> 
> > RISC-V is allowing (encouraging? Cheering on?) an open design model 
> > that allows many companies to invest in parallel.
> 
> Yep; the core ISA is royalty-free, and anyone can build a core that 
> implements it.  Which is great, if you can afford to develop your own 
> core (and very few can; you have to have some pretty serious volume 
> before the NRE involved in going your own way eclipses the Arm licensing 
> fees.. which is the secret behind Arm's success.  That said, the 
> continued consolidation in the semiconductor industry has changed this 
> dynamic as per-company volumes have grown massively)
> 
> > The low-power capabilities of RISC-V is also attractive.  When you 
> > have 500,000 processors in a HPC system, every half-watt you save goes 
> > a long way, in both direct power and cooling.
> 
> In HPC environments, it's not the "cpu core" that is sucking down the 
> wattage, it's "un-core" stuff -- stuff like the memory and I/O 
> controllers, plus obviously everything else connected.  RISC-V won't 
> affect that at one iota at an equivalent process node and clock speed. 
> 
> This is a lesson that proponents of "high-end Arm" have also started to 
> learn the hard way; it turns out that by the time you slap a couple 
> hundred gigabytes of RAM across four memory busses and a several dozen 
> saturated PCIe lanes onto your system, your overall "transactions per 
> watt" comparison isn't quite so favorable any more.
> 
> > In a SBC that I am working on in Brazil, we switched from using 5 
> > volts to power the SOC to 3 volts, and went from 10W power to 5W of 
> > power used.  This actually allowed us to turn up the clock, since the 
> > heat generated was lower.  So lowering the voltage and power allowed 
> > the chip to run faster.
> 
> Supply voltage has nothing to do with the CPU core, and everything to 
> do with the physical implementation (process node of the ASICs, power 
> supply design, I/O voltage requirements, etc etc)
> 
> (BTW, my last gig was at a fabless semiconductor startup that was 
>  eventually acquired by Arm. I was swimming in the deep end of 
>  mixed-signal ASIC design/bringup for nearly six years)
> 
>  - Solomon
> -- 
> Solomon Peachy			      pizza at shaftnet dot org (email&xmpp)
>                                       @pizza:shaftnet dot org   (matrix)
> High Springs, FL                      speachy (freenode)