Received: by 10.66.90.37 with SMTP id bt5mr3564123pab.40.1352098710198;
        Sun, 04 Nov 2012 22:58:30 -0800 (PST)
Path: 6ni62960pbd.1!nntp.google.com!news.glorb.com!news-out.octanews.net!indigo.octanews.net!auth.brown.octanews.com.POSTED!not-for-mail
From: Paul Rubin <no.em...@nospam.invalid>
Newsgroups: comp.lang.forth
Subject: Re: Xula-200 or Raspberry-Pi a better buy?, was [Re: RTX2000 optimization]
References: <e1479cfa-a969-40ab-b20c-096d82601657@googlegroups.com>
	<k6pljm$df7$1@dont-email.me>
	<3289bb1a-2b62-4b5e-b32f-55887176d7b9@googlegroups.com>
	<k6rht8$57r$1@speranza.aioe.org>
	<ec3f886a-0d36-4220-a67d-dfc36d02c371@googlegroups.com>
	<7x390ulich.fsf@ruckus.brouhaha.com> <k6rqio$qh9$1@dont-email.me>
	<7xehkeea8s.fsf@ruckus.brouhaha.com> <k6s1r6$cdh$1@dont-email.me>
	<7x1ugegfoi.fsf@ruckus.brouhaha.com> <k6udm9$19t$1@dont-email.me>
	<7xlielw5a9.fsf@ruckus.brouhaha.com> <k6uf5c$b41$1@dont-email.me>
	<7xip9ne0dm.fsf@ruckus.brouhaha.com> <k73i0a$kh0$1@dont-email.me>
	<7x625mtwfw.fsf@ruckus.brouhaha.com> <k73m1d$dn1$1@dont-email.me>
	<7x1ugatsdb.fsf@ruckus.brouhaha.com> <k73s3v$mho$1@dont-email.me>
	<7xlieixpmd.fsf@ruckus.brouhaha.com> <k74eak$v9v$1@dont-email.me>
Date: Sun, 04 Nov 2012 22:58:29 -0800
Message-ID: <7xy5ig1qp6.fsf@ruckus.brouhaha.com>
Organization: Nightsong/Fort GNOX
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/23.1 (gnu/linux)
Cancel-Lock: sha1:+Z8KxJnRNgHrQZpoULl4sts20SU=
MIME-Version: 1.0
Lines: 82
NNTP-Posting-Date: 05 Nov 2012 00:58:29 CST
X-Complaints-To: abuse@octanews.net
Content-Type: text/plain; charset=us-ascii

rickman <gnu...@gmail.com> writes:
> Unless you have some novel ideas I think yes, this is a bit
> unrealistic. The NRE may be high on more current technology, but even
> 90 nm would be a bit advantage over 350 nm and the costs would be
> worthwhile if there is any sort of a market I would think.

Bernd's paper on the b16 mentions the b16 uses 0.4 mm**2 in a TSMC 0.5
micron process (I think "nanometers" came into use around when feature
sizes dropped below 0.1u).  So other things being equal, that would
scale to 0.2 mm**2 at 0.35u.  The smallest chip you can order is 3 mm**2
which would be 15 processors, hmm, yeah, you probably do need either a
bigger chip or a smaller process.  Might still be feasible at a few
notches up the scale.

This was actually the first application I thought of for the GA144, but
the GA architecture is of course completely wrong for it.

> Yes, you can even connect somewhere in excess of 2.5 Gbps directly to
> FPGAs.  I'm not current on just how fast you can do this.  Many of
> them have SERDES built in, just not sure how fast they go.

I'm seeing there are some astoundingly powerful FPGAs now, including
some with 28 gigabit inputs, high end ARM cores as hard cells, etc.
They are crazy expensive (Altera Stratix V GT).  :-O.

> I would think the value of bitcoins might increase because of the
> increased costs.

That's the hope of the participants, but remember, they're a commodity,
they're not consumed, they circulate and trade on exchanges.  They're
hard to produce but they're not terribly scarce and their prices derive
mostly from crazy speculation.

>> I suspect that people who bought expensive FPGA mining rigs 
> Ok, so it's a bit like selling the stuff for custom PCs, not because
> they are faster so much, but because they are *cool*.  Also like
> hotrod cars.  Possibilities...  But in 350 nm?  Maybe, maybe not.  If
> the point is speed, then the really old technology won't cut it I
> think.  FPGAs may be faster for the $$$.  Not sure.

Yeah, old numbers were that ASIC's were an order of magnitude smaller
and faster than FPGA's that did the same thing, i.e. by using an ASIC
you get the equivalent of an FPGA that is several process nodes more
advanced.  I don't know if this still holds.  Power consumption also
matters.  A bitcoin is worth maybe $10, so if it takes 100 KWH to mine
one, at least here in California you're losing money by mining.

> [Novix-like Forth processor] A 10 year old FPGA runs my CPU at 50 MHz
> ... I wonder how fast it might run in 350 nm?

The Pentium II ran at 233 mhz in 350nm, if I'm reading Wikipedia
properly.  Bernd mentions the b16 at 100 mhz in 0.5u.

>> There is certainly a commercial market for security hardware
> They don't trust nobody!  Very paranoid about the backdoor thing.  Who
> is the market for a security chip specifically?  Do you have expertise
> in this? ... If it can be done on a shoestring, it can be copied on an
> even smaller shoestring.

I have some knowledge of the software side: I've programmed commercial
units and have implemented a software workalike.  As before I'm
near-clueless about hardware.  Besides the chip, you need
tamper-reactive packaging, EM shielding, isolated power supplies, etc.
But it doesn't have to be particularly miniaturized.

I wouldn't worry too much about copying.  The industry making these
things is very conservative and if they think you're doing something
that threatens them, they'd probably rather buy you out than copy you.
There's a saying that if your idea is REALLY good, nobody will steal
it--you will instead have to ram it down their throats ;-).

I can also see some point to a relatively low cost product with less
excessive anti-tamper stuff, for use in ordinary private business
communications.  Existing general purpose PC's are so full of backdoors
and security holes that it's time for dedicated hardware to make a
comeback.  And IMHO the Freedom Box would be more interesting with a
chip like this.

I notice recently there's a kickstarter for a FOSS SOC with some 32-bit
core and peripherals, even though the development chain isn't all FOSS.
So maybe a chip made with all-FOSS (so it's auditable) would also be of
interest.