£25 pc
 

[url]http://www.theregister.co.uk/2011/11/28/raspberry_pi/[/url]

What do people think to this?

[QUOTE=henryzz;283363][url]http://www.theregister.co.uk/2011/11/28/raspberry_pi/[/url]

What do people think to this?[/QUOTE]Yes. Can I factor on it? I looks like the power is a simple DC in. :smile: Tie a dozen into a switch and go!

Exactly. I don't suppose anybody has any sort of benchmarks for this thing? Or is there even something to run on it besides Mlucas, i.e. factoring type stuff?

Great toy...put your program on your card, plug it in, and it works -- you supply 5V (the easiest way to brick the beast!). Too bad about the proprietary graphics instructions...but might be quite a DC beast, once an OS is ported to it. I'd expect perhaps half to a quarter of a Pentium IV in compute horses, but you make that up with the 1-2 watts spent powering it and the low cost and footprint. (If I don't mind the power, or driving across town to pick them up, I can purchase working P-4 PCs right now at about $50 on Craigslist, sometimes with flat panel displays, too)

As an embedded systems developer, however, the system looks familiar....I went ahead and blew perhaps $100 on a PICKIT3, which targets any microchip device I want to use, and have the DSPIC30F4011 in a 40-pin through-hole DIP on a through-hole card. Checkout took 2-3 days of last week, mostly because of an omission/comission in the documentation (a certain mux setting was reset under a certain condition, I found this experimentally). The beast is running at 120MHz, a very small program, and I watch the outputs on an oscilloscope.

Now, if it has a good networking stack, that helps a bunch....as Modbus is probably all the networking I will ever want to support on that PIC.

It's always a struggle to balance the simplicity of the system with its power. Anyone else remember BASIC, back when it had line numbers?

To make it work, the big question will be if it supports a good debugger easily. Although I learned a lot of programming without one, I consider it an essential tool.

[QUOTE=Christenson;283404]It's always a struggle to balance the simplicity of the system with its power. Anyone else remember BASIC, back when it had line numbers?[/QUOTE]Yup. I think you'll find a good number of us old farts here.

Anyone else here program in FOCAL? A much nicer language, IMAO, than Dartmouth BASIC.

Paul

Strikes me as a dismantled smartphone without a screen. 1W for a 700Mhz CPU sounds good if we can utilize it. I can just imagine a computer farm with 100 of them:smile:.

Another link [url]http://www.bbc.co.uk/news/technology-16316439[/url]

Power of ethernet is what I'm thinking to power them.

It would cut down on cabling.


You would have to have a decent size farm for it to be worthwhile. It was single core 700MHz, you'd need at least a bank of 16 of them to beat a modern quad core.

That's a whole lot of assumptions though. But giving a rough ball park of 400quid, gets similar to just buying a modern quad core.

But still it's an interesting concept.

-- Craig

[QUOTE=nucleon;283500]gets similar to just buying a modern quad core.[/QUOTE]
Yes but those 16 would be consuming a total of ~16W rather than >200W. If software got good enough for them then they could become a easily sustainable computer farm. Worth remembering less power means less heat as well.

[QUOTE=henryzz;283502]Yes but those 16 would be consuming a total of ~16W rather than >200W. If software got good enough for them then they could become a easily sustainable computer farm. Worth remembering less power means less heat as well.[/QUOTE]

($900/i7 before monitor?) /($25/this) = 1(i7)/36(this) could 36 of this keep up with one i7, my guess is someone's interested.

I think 36 of these is closer to two i7's. On a straight clock comparison (I know, that's what nucleon's hesitant about) you'd need around 5 of these to match Intel's quads. I don't think it'll match clock per clock, but even if each ARM clock is 1/3rd of an Intel clock, that's still 15 of these to an i7 (not E series).

[QUOTE=Dubslow;283511]I think 36 of these is closer to two i7's. On a straight clock comparison (I know, that's what nucleon's hesitant about) you'd need around 5 of these to match Intel's quads. I don't think it'll match clock per clock, but even if each ARM clock is 1/3rd of an Intel clock, that's still 15 of these to an i7 (not E series).[/QUOTE]

On a straight clock comparison you'd need around 5 of these to match [B]one core of[/B] Intel's quads. So that's roughly 20 of these. But with a reasonable per-clock expectation of 1/3rd IPC that of i7 (on FLOPs, at least), it'll be more like 60 of them = 1 i7.

Whoops. 60=15*4, so at least we agree otherwise. Thanks.

another link , this time, done by nvidia
[URL]http://www.tomshardware.com/news/nvidia-cuda-arm-9-tegra-3-carma,14335.html[/URL]

[QUOTE=science_man_88;283503]($900/i7 before monitor?) /($25/this) = 1(i7)/36(this) could 36 of this keep up with one i7, my guess is someone's interested.[/QUOTE]

It's not that simple either. As you have way more machines - networking becomes non-trivial.

To use entry level switches (8 ports). You'd need 6x switches for 36 nodes. In a 5 + 1 arrangement. The price I'm looking at here - 24port 10/100/1000 for $165* - you could get by with 2 of those - $330. That's almost $10 per node.

That's still without power supplies and and multi-socket power boards.

It would till be an interesting exercise to do a full TCO. It's borderline whether you'll save on upfront costs - but it sounds like ongoing cost savings is a possibility.

-- Craig
*That's AU pricing.

[QUOTE=nucleon;283632]It's not that simple either. As you have way more machines - networking becomes non-trivial.

To use entry level switches (8 ports). You'd need 6x switches for 36 nodes. In a 5 + 1 arrangement. The price I'm looking at here - 24port 10/100/1000 for $165* - you could get by with 2 of those - $330. That's almost $10 per node.

That's still without power supplies and and multi-socket power boards.

It would till be an interesting exercise to do a full TCO. It's borderline whether you'll save on upfront costs - but it sounds like ongoing cost savings is a possibility.

-- Craig
*That's AU pricing.[/QUOTE]
Shame they don't have wifi. That would solve this problem.

[QUOTE=henryzz;283651]Shame they don't have wifi. That would solve this problem.[/QUOTE]

That would create other problems. 30+ radio sources in a small area, that has to have it's own set of problems.

-- Craig

I want BBC Basic on RISC OS.