[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: More multi-core



Larry,

Occam? No.

The application is generating and checking packets where the aggregate throughput exceeds 5Gb/s. The language is still evolving (as I write the compiler) but is very specific to the purpose - although it DOES have ? and ! (and !? for an input that also immediately outputs on another channel). Concurrency is mostly hidden from the user.

    Barry.

Dr Barry M. Cook, BSc, PhD, CEng, MBCS, CITP, MIEEE
CTO,
4Links Limited,
Bletchley Park,
MK3 6ZP,
UK.
----- Original Message ----- From: <tjoccam@xxxxxxxxxxx>
To: "Barry Cook" <Barry@xxxxxxxxxxxx>
Cc: "Ian East" <ian.east@xxxxxxxxxxxx>; "Occam Family" <occam-com@xxxxxxxxxx>
Sent: Thursday, July 05, 2007 3:47 PM
Subject: Re: More multi-core



Barry,

Is there an occam compiler for this device?

Larry Dickson

Ian,

Why wait - do it yourself, we just did!

In order to provide an option customers asked for we implemented a small
processor and then dropped 16 of them onto an FPGA - actually on the side
of an FPGA already pretty full with our main product. Peak performance of
each processor is ~750MOPS so the total is about the same as claimed by a
top-of-the-line Pentium (but at far lower power - the whole box with its
display and fast interfaces consumes ~25W). Admittedly the application is
naturally fairly parallel and interaction between processors is limited
(the processors work in pairs).

Anyone wanting to do some research should not find it too difficult to
implement interesting multi-core on reasonably priced FPGA's.

       Barry.

Dr Barry M. Cook, BSc, PhD, CEng, MBCS, CITP, MIEEE
CTO,
4Links Limited,
Bletchley Park,
MK3 6ZP,
UK.
  ----- Original Message -----
  From: Ian East
  To: Barry Cook
  Cc: Occam Family
  Sent: Wednesday, July 04, 2007 5:14 PM
  Subject: Re: More multi-core


  Looks like I better dust off the old Parallel Programming text what I
rote back in the last century!


  At Brookes/Computing we now teach no concurrent programming at all.  If
it resumes, it'll probably be monitors using Java - which only 1 in 20
or so will master to any useful degree.  But I'll still be expected to
return 70% pass-rate, of course.


  It will start to get interesting once the hardware, on a general-purpose
machine, passes 8 or so cores.  Any guesses as to performance-scaling
achieved in practice?  Could be lucrative this time around.


  Ian


  On 4 Jul 2007, at 09:35, Barry Cook wrote:


    News of another manufacturer entering the multi-core arena - with yet
another architecture.

    Interestingly, the power consumption argument is getting more
prominent [Two cores consume less power than one core at twice the
speed]. We heard it at WoTUG in Eindhoven (from Philips) and it is
spreading.


http://www.embeddedtechjournal.com/articles_2007/20070626_freescale.htm

        Barry.