[Beowulf] Re: Transputers vs Today doing lossless FFT - was Rackable SGI

[Nifty Tom Mitchell]

On Sun, Apr 05, 2009 at 02:19:45AM +0200, Vincent Diepeveen wrote: > Subject: Transputers vs Today doing lossless FFT - was Rackable SGI
> On Apr 5, 2009, at 12:38 AM, Nifty Tom Mitchell wrote:
>
....
>>
>> Two of the problems with the Transputer was the lack of virtual memory
>> and that the the network was fixed.
>>
>
> Hi Transputers a bit before my time,
> but wasn't it the case that transputers were like 1Mhz and a network of 
> nearly 10 mbit?
> The processors speeds that were reported to me was having a really bad 
> ugly ipc.

In its day the transputer was very interesting.
It was RISC and the clock seemed more than fast enough at
the time

My thought was that it would be too easy to place more
cores in a package than cache and memory bandwidth can
well support.  By adding a communication channel say multi
ported memory/ fifo that would let neighbors communicate
outside the shared commons of main memory and cache many
more cores could be applied to problems.

It may be that compilers could look at pairs (or more)
and pipeline data flow or even generate speculative execution
code paths...  If the programming model and patents issues
are open enough that multiple vendors could play fun stuff
can begin.

-- 
	T o m  M i t c h e l l 
	Found me a new hat, now what?