Re: [oc] Re: Merlin Hybrid System

["Jim Dempsey" <dempsey@northnet.net>]

----- Original Message ----- 
From: "Paul McFeeters" <paul.mcfeeters@ntlworld.com>
To: <cores@opencores.org>
Sent: Tuesday, December 04, 2001 3:31 PM
Subject: RE: [oc] Re: Merlin Hybrid System


> Jim,
> 
> Converting single thread program/OS to use multiple processors is
> a very interesting and long overdue technology leap.
 
The process I've envisioned certainly deserves a patent and could
be used for a thesis as such I cannot discuss any of the technical
details. I can say the concept is sound and will work. To do this
project will either take the cooperation of Transmeta or a
duplication of some of their efforts. Transmeta has cooperated with
other processor vendors (for a fee). So if they don't want to partner
a purchase of their technology may be done. Please note that the
current line of Transmeta processors are designed for low power
consumption. i.e. fewest active cells inside the processor core. And
using the smallest die they can fit the cells into. The processor extension
envisioned would likely use a larger die with the additional processors.
Other than for heat dissipation the pinouts and chip carrier size could
concievably remaine the same. I would think that using this process
that they could make a single Socket 7 chip replacement with
2, 4 or even 8 processors. Think of the marketing hype they could
make out of this "Wow! I could have had a V8". Think of this,
they could make a line of replacement chips suitable for insertion
into virtualy all desktop systems installed today. I cannot understand
why they aren't interested in taking a look at this. I guess I am not
a good enough pitch man.


> Could be an
> even bigger leap forward than when we finally dump the x86 instruction
> set for a proper RISC optimised set. 

Don't discount the x86 instruction set. After all, there is no processor
that actualy execuits this instruction set. They all repackage the stream
to an internal format that is much more RISC-like. The way I see it
is you have three complaints. 1) You don't have a Pentium-IV emulator
to run in your favorite processor core. 2) You don't have a Pentium-IV
IP core. 3) If you did have either 1 or 2 it would drive you either into
a larger PLD and cost you more. I think you should focus more on
not trying to enter the Intel and AMD (et al) market. What should excite
you more is that you do have decent high level (C/C++) software
development tools for use on some of the popular embedded processor
cores.


> Have you done any performance
> figures yet? Obviously the two things people will look at are
> performance and the price. If you could produce a testbench to
> demonstrate the theory in practice and obtain some performance
> figures I'm sure more companies would be interested.
> 

Also, as of this time I have not written a simulator that would be capable
of computing the efficiencies gained for particular processor demands.
Note the specific use of "processor demands" instead of "programs"
or "code sequences". With this system as the complexity increases
on the processor the efficiencies improve using this process.

Writing the simulator would be no easy task. This will not be a "simple"
case of writing an x86 (Pentium) emulator. Instead, an emulation of
the Crusoe Code Morphing would need to be done and then that
would have to be extended to incorporate the new innovation. For this
it would be best to start with using the development platforms and
equipment that Transmeta uses to develop their Code Morphing code
for the Crusoe processors. Why re-do that part of the work if it
has already been done.

> I think Crusoe have enough on their plate with trying to take on
> Intel and AMD in the X86 marketplace. We don't know how successful
> their processors have been so they might already be fully stretched
> which could help explain their lack of interest.
> 

Their target market is for notebooks and embedded systems (palm tablets)
desktop systems were not in their plans.

General discussion...

If you examine Evergreen Systems or Intel Over Clocked processors or
any of the other processor replacement options you will note that these
systems tended to fail if they attempted to exceed the design experience
engineered into the motherboard. A replacement chip of this design could
concievably live within these constraints yet produce a significant return
on performance. i.e. you do not need to crank the clock up and outside
the performance limitations of the motherboard in order to take advantage
of the extra processors.

Jim Dempsey


--
To unsubscribe from cores mailing list please visit http://www.opencores.org/mailinglists.shtml