[Libre-soc-dev] 3d gpu business plan

Luke Kenneth Casson Leighton lkcl at lkcl.net
Wed Dec 30 12:25:39 GMT 2020

crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68

On Wed, Dec 30, 2020 at 7:59 AM Lauri Kasanen <cand at gmx.com> wrote:
> On Tue, 29 Dec 2020 18:07:07 +0000
> Luke Kenneth Casson Leighton <lkcl at lkcl.net> wrote:
> > On Tuesday, December 29, 2020, Lauri Kasanen <cand at gmx.com> wrote:
> > > A giant like Mediatek will have volume licensing and ready suppliers in
> > > place, requiring no negotiation at all. Most things they'd have
> > > in-house.
> >
> > Lauri: *it still takes 12 to 18 months*.  this is just how it is.
> Why do you think it will take as long for a company like Mediatek as
> for a small player? Their product portfolio suggests otherwise.

just like Samsung and Allwinner they have multiple overlapping teams.
samsung have two separate teams: one internal, and the other is
external.  allwinner have five or six separate business teams, each
with their own separate investors.

> If it takes too long to explain, please point me to a book or some other
> resource.

i can't: nobody can.  it's an extreme-exclusive world given the amount
of money involved.  it's taken me over 10 years to gain this
knowledge, part of it through a mentor of 10 years who worked for LSI
Logic and was the head of Samsung Research for many years.  he's over
80, now.

the associate i contacted is 60 years old: he's worked in this field
since the 80s.  both the people he's put me in touch with are over 68.
they've been in this field again since the 80s.  they're not going to
write books about this stuff.

> > the scale at which large players operate, they are simply not interested in
> > adding unnecessary peripherals in specialist markets that would cause them
> > to be unprofitable, trying maybe to take a 0.5% market that is not their
> > core business.
> >
> > we are simply starting out "too small for them to try bothering to wipe off
> > the map" even accidentally.
> It's not a 0.5% specialist market. Consider case ESP. It's a crap
> processor, but it's literally everywhere because it costs 1/10 of the
> alternatives. Because they got the design and license for cheap, and
> are now shipping billions.

that's great for them.  how many pins does it have, and what's the
maximum processor speed?  based on that: is it suited to anything
remotely similar to what we're trying to do?

no it doesn't, does it?  therefore inherently by design it's not a
threat, is it?

you get *one pin* missing based on the customer's requirements and
they'll go elsewhere.  this is why SoCs have pinmuxes.  ST Micro goes
another level of extreme on top of that by having a massive family all
the way from STM8S003 which is 8 pins 8mhz to a (gosh!) 48 pins 24 mhz
(exciting!), through the STM32 series which goes all the way up to 256

> If there now becomes available an entirely free, perfomant quality
> core, it very much will make them take notice.

see below.  this is not as true (in general) as you believe.

> RISC-V has not had any such implementation, just the spec.

it's slowly getting there.  however what you are missing is that
there's been several RISC-V cores already gone into mass-production.
Trinamic was one of the first.  Western Digital was another.  NVidia
*might* have already done so.

why have you not heard of them?

because they're all entirely closed-source implementations where the
companies behind them are simply using RISC-V to eliminate the cost of
ARM licensing.  or in WD's case, to eliminate in-house development
costs [they had their own 32-bit DSP with a custom in-house ISA, for
HDD, SSD and NAND management].

> Their target is not the "we want openness and security" crowd. Their
> target is to sell billions of widgets, and if our core lets them
> undercut the ESP by half, they will do so happily.

OpenPOWER - the ISA itself - is total overkill for them.  an OpenPOWER
ISA decoder alone is larger than the entire RISC core used in that
type of processor.  they would be utterly shooting themselves in the
foot to use it, because the die area would be increased so greatly
that it would knock the price up by a margin so high that it would
become unprofitable against their *own* current product.

does this give you some insight, here?  this is a multi-faceted,
multi-dimensional arena.  you simply cannot draw these kinds of
generalised comparisons and analogies: they just don't apply because
there will be *another* dimension - another aspect - where your lack
of experience will bite and invalidate the comparison.

> > appreciated.  it's not a problem.  the market differentiators are so large,
> > and the specialist targetting of each SoC so different, that there is no
> > overlap.
> It may be enough, but if so it needs to be phrased to convey that.
> I may not be qualified in chip design, but I very much am qualified in
> investing,

Lauri: the two simply do not automatically go together.  if you do not
know the market - whatever it is - then do you think it is wise to
invest in that market (on your own, i mean, without a group to share
the risk)?

if you did not know about, say, the Mining Industry, would you bet all
your money on a new team that claimed to have a better way of
prospecting for diamonds?

if you did not know anything about, say, nanotechnology, would you bet
all your money on that?

of course not.

it is the same thing here.  the lack of knowledge that you have in
this field means that you are simply being a *sensible investor* by
staying *away* from investing in something that you know nothing

> and it's certainly not convinced me so far.

the ASIC market is one that you just have to "know" through
experience.  that means that the investors themselves have to know the
market, through experience as well.

a non-experienced investor in the ASIC world will need
approximately.... i'm guessing here... 3-5 months of full-time
training and research before they'll have the information needed in
order to be "fully informed" and know the difference between a good
bet and a bad one.

they'll need an experienced mentor to be able to train them, walk them
through the process.  and as you can see from the age of the people
i'm in contact with, time is marching on.


More information about the Libre-soc-dev mailing list