Discrete Opamp Open Design

[gk7]

Originally Posted by qusp

... grab 2 of them and connect together, control as small cluster from terminal/console. ...

It takes a bit more to make a cluster than just connecting 2 (or more) machines together. (Cluster software and applications which are able to distribute their tasks on cluster nodes).

[qusp]

Originally Posted by gk7

It takes a bit more to make a cluster than just connecting 2 (or more) machines together. (Cluster software and applications which are able to distribute their tasks on cluster nodes).

yes, did I not give examples of the type of applications? math, visualisation, scientific modelling are all good candidates for having the capability for distributed tasks. the type of applications that have been under discussion for the last 10 pages or so. also they have Thunderbolt connections, not sure if the latest ones are still limited to 2 channels as mine is, but regardless thats still 20GB/s each way. it supports PCIe and display port as standard on the thunderbolt bus, which would probably allow the connection of an external graphics card as another node and use as GPU for math.

so does anyone here really need more than that? ie 2 x i7 quads, internal GPUs in each machine, thunderbolt link between them and an external GPU node daisy chained off each machine

[gk7]

You completely misunderstood me. I think you would need to read first how a cluster works (no offense). Some links:
Computer cluster - Wikipedia, the free encyclopedia
and here especially:
Computer cluster - Wikipedia, the free encyclopedia
more comprehensive:
http://cloudbus.cis.unimelb.edu.au/~...ster/v1toc.pdf

[hochopeper]

Having admittedly not read much of the recent discussion can I suggest that for diy software development python with iPython, scipy, numpy and matplotlib provides a very powerful set of scientific computing tools and iPython with pyzmq allows for powerful cluster computation and can run on linux/windoze/mac as in archtecture agnostic.

I'd also suggest that arm processors will come to the desktop faster than people might currently realise and intel/microsoft may well be left in tears over that situation!

[qusp]

umm, read your link (first one), i'm at a loss to see what clashes with what I said...

Mac can run Linux natively, can run mac os natively, can run windows natively. there is nothing preventing a cluster from being realised with the small system I described. no more than any other small collection of CPUs anyway.

read second link (skimmed) still nothing, didnt read it in detail because it includes some tech info beyond the scope and beyond my care factor =)

[hochopeper]

Originally Posted by qusp

umm, read your link (first one), i'm at a loss to see what clashes with what I said...

Mac can run Linux natively, can run mac os natively, can run windows natively. there is nothing preventing a cluster from being realised with the small system I described. no more than any other small collection of CPUs anyway.

read second link (skimmed) still nothing, didnt read it in detail because it includes some tech info beyond the scope and beyond my care factor =)

I think his point was simply this:

The software config needed to realise a fully functional cluster is far more complicated than hardware+OS.

[qusp]

Originally Posted by hochopeper

I think his point was simply this:

The software config needed to realise a fully functional cluster is far more complicated than hardware+OS.

sure, but did that point need to be made? of course it is.

you cant just connect an external graphics card, or group of graphics cards together and ask them to act as processors either but thats been discussed at length here. pretty much everything under discussion assumes some level of customisation of the environment. as do the basic simulation tools need customisation for the task at hand so the routines run efficiently

the need or desire to build something like this in the first place kind of assumes that level of knowledge.

[hochopeper]

Originally Posted by qusp

sure, but did that point need to be made? of course it is

anyone that has been through the pain of trying to set up multithreaded application that shares processes in a distributed fashion could easily get offended by trivialising the matter frankly cluster computing is in dire need of commoditisation. Amazon, Google et al have gone a long way towards making it affordable, its certainly not something any bloke (even with a reasonable level of general computer knowledge) could go and configure and do something productive with inside of an afternoon. Which is where the idea really starts to lose its shine at the moment.

[qusp]

trivialise? perhaps you should read the last 10 pages at the applications under discussion. there is nothing trivial about the applications or people in the discussion

I sure dont have the faculty or resources/time to do it, Maya already goes some way towards doing it out of the box for its rendering farms, but thats very application specific and that application costs several times the amount of the hardware system I laid out...per seat

[scott wurcer]

Originally Posted by hochopeper

its certainly not something any bloke (even with a reasonable level of general computer knowledge) could go and configure and do something productive with inside of an afternoon. Which is where the idea really starts to lose its shine at the moment.

You bet, at that point it's a job not a side task serving your hobby.