On Sun, Sep 16, 2012 at 3:26 PM, Andy Farnell padawan12@obiwannabe.co.uk wrote:
On Sun, Sep 16, 2012 at 10:24:45AM -0300, Alexandre Torres Porres wrote:
now my question is;
spending 4k to build a Pi supercomputer can give you more power and possibilities than with a top of the line MAC for example (which will cost just as much, and be a quad core 2.7 intel i7, 1.6GHz bus, 16GB Ram).
We keep using the word 'supercomputer', and maybe a bit of perspective would help clarify matters of scale.
...
A supercomputer is, by definition, that which is on the cutting edge of feasible research. Most supercomputers are in a single location and not distributed or opportunistic, they usually have a building dedicated to them and a power supply suitable for a small town of a thousand homes (a few MW). A team of full time staff are needed to run them. They cost a few hundred million to build and a few tens of millions per year to operate. Current supercomputers are measured in tens of Peta FLOPS, ten to a hundred times more powerful than the equivalent mainframe, and are primarily used for scientific modelling.
Yeah, but when I tell people what I do, do you think I say "cluster computing" or symmetric multiprocessing or CUDA applications engineer? No, I tell them I work with "supercomputers"--It's not a term for practitioners, since there's more specific things to say, ... and it keeps people from thinking I'm going to waste time talking about nerdy shit that I don't want to talk about anyway :)
The current guise of the 'mainframe' is what we would now see as a Data Center, a floor of an industrial unit, probably much like your ISP or hosting company with many rows of racked indepenedent units that can be linked into various cluster configurations for virtual services, network presence and data storage. Aggregate CPU power in the region of 10 TFLOP to 0.5 PFLOP
At the moment, I'm (the engineer) putting together the proposal for a grant for GPU computing resources (for the researchers and scientists). We're looking to spend about $750,000 on hardware that will perform about 100 TFLOPS. Mostly it will be made up of--whatever NVIDIA Tesla is most cost/power effective--in servers that will hold 4 GPUs. Altogether, we hope this fills up 5-10 racks (in our shiny new energy efficient data center with 32 racks, that the f'ing fire marshall won't let us into for another month, when we've been postponed since June anyway).
Supercomputers are still supercomputers, by definition they are beyond wildest imagination and schoolboy fantasies unless you happen to be a scientist who gets to work with them. A bunch of lego bricks networked together does not give you 20PFLOP, so it does not a supercomputer make.
However, there is a different point of view emerging since the mid 1990s based on concentrated versus distributed models. Since the clustering of cheap and power efficient microcomputers is now possible because of operating system and networking advances, we often hear of amazing feats of collective CPU power obtained by hooking together old Xboxes with GPUs, (Beowulf - TFLOP range) or using opportunistic distributed networks to get amazing power out of unused cycles (eg SETI at home/BOINC and other volunteer arrays, or 'botnets' used by crackers) (tens to hundreds of TFLOPS).
Clustering is currently the most scalable model for supercomputers. Many expensive options exist for systems with large numbers of cores and shared memory--but year after year, more circuits get put on a single die. Generally when you think of supercomputers these days, it's a network of systems that each have a lot of x86_64 cores and a maybe nice co-processor (like the NVIDIA Tesla's).
Some of the IBM machines (and Cray, still?) use pipelined multi-core processors of a different architecture and 1000s of cores on a single system, but I don't see that as a trend that will survive.
Chuck