Help Sitemap Home Skip Navigation Contact Us Disability Statement

Endinburgh Council
 
 
Saturday, 7th November 2009 Change Date

UK's best supercomputer set to launch in Lothian

Premium Article !

Your account has been frozen. For your available options click the below button.

Options

Premium Article !

To read this article in full you must have registered and have a Premium Content Subscription with the Edinburgh Evening News site.

Subscribe

Registered Article !

To read this article in full you must be registered with the site.

Sign In
Register
Click on thumbnail to view image
Click on thumbnail to view image
Click on thumbnail to view image
Click on thumbnail to view image
Click on thumbnail to view image

Published Date: 08 January 2008
THE biggest and most advanced supercomputer in the UK is set to be launched near the Capital next week.
Hector – or High-End Computing Terascale Resources – will enable researchers to forecast the impact of climate change, project the spread of epidemics and develop new drugs.

The £113 million machine will be based at Edinburgh University's advance
d computing facility at Edinburgh Technopole at Roslin, Midlothian.

It will be surrounded by top-level security 24-hours a day to stop intruders getting in.

The system will run for six years and be operated by Edinburgh Parallel Computing Centre (EPCC) – one of Europe's leading centres for the provision of supercomputer services to universities.

Hector is capable of 63 million million calculations a second and is the equivalent of around 12,000 desktop systems. It is said that Hector's power is the equivalent of every person on Earth carrying out 10,000 calculations in a second, all at the same time.

Dr Alan Simpson, director of Edinburgh University company UoE HPCx, said: "It's great to see Edinburgh and Scotland leading the way in this technology.

"There are people looking at climate change, people looking at improved pharmaceuticals and the way that drugs interact with the human body, and people looking at acquiring quieter, more fuel-efficient jet engines."

Hector aims to provide a world-class service for UK-based academic research, as well as enabling researchers to work with their colleagues in Europe and worldwide.

Academic researchers from across the country will pay to link up with the Cray XT4 supercomputer via the internet.

It is expected that Hector will attract up to 1000 users.



The full article contains 279 words and appears in Edinburgh Evening News newspaper.
Page 1 of 1

  • Last Updated: 08 January 2008 2:22 PM
  • Source: Edinburgh Evening News
  • Location: Edinburgh
  • Related Topics: Midlothian
 
1

Fairfax,

08/01/2008 16:01:50
"It is said that Hector's power is the equivalent of every person on Earth carrying out 10,000 calculations in a second"

This is very pompous writing to summarise school-level arithmetic. Indeed, dividing the number of calculations per second (63 x 10^12) by the approximate population of the planet (6 x 10^9), we obtain (63/6)x10^4 = 10,500 calculations per person per second: no need for "It is said . . .".

Perhaps a more useful comparison would be the cost of this power: £10^8 for 63 teraflops is poor value for money. For comparison, a low-cost modern desktop, providing 2 gigaflops say, would cost a few hundred pounds only: let's say £30000 per 64 processor Beowulf clusters. The same money could provide 30000 such clusters. Of course, the Cray can provide more power, but I've often seen such machines used to provide crutches for scaled-up versions of inefficient algorithms.

2

,

08/01/2008 16:38:12
Comment Removed By Administrator
Reason:
3

alex paterson,

embra 08/01/2008 16:41:29
Will it help my spelling and punctuation,this may just please my English teacher.
4

GalacticCannibal,

Murrieta, California..ex Mexican territory 08/01/2008 16:58:07

UK's best supercomputer set to launch in Lothian

--------------------------------------------------

Yuo don't refer to computers are "Best"
"best supercomputer" WOW did they buy this from "BEST BUY" ??

If you Dudes paid £113 million for this "first generation system ".. you were ripped -off.


U Dudes should have bought processing time on one of the super multi- parallel computers, working in Austin TX,. Their speed and capability make yours look like the proverbial tortoise.
They were built for $100 million (£50 million)...

Another example of u trying to invent the WHEEL

DON'T PANIC the gullible UK taxpayer have deep pockets and will pay as usual . Don't have the guts to complain.

GC

5

Unimpressed one,

08/01/2008 17:01:52
No computer yet has been successful at modeling 'climate change'. I bet any money this one will not be the exception.
6

fred bloggs,

Edinburgh 08/01/2008 17:33:35
6. You've already admitted on another thread that you don't want to believe in climate change.
7

An English Voice,

08/01/2008 17:35:26
5. Some more facts about this system and the state of supercomputing in the UK:

The system itself will rank at about 17th in the world for power and speed and is 50% faster than the next best in the UK at the Atomic Weapons Establishment.

The UK is the second largest user of supercomputers in the world and the most in Europe, with 48 of the 500-ish systems currently in use (this will make it 49). The USA has 284 (roughly the same 'per person' as the UK), Germany has 31, Japan has 20 and China has 10.

By the way, remember that the UK invented the computer. So, when you think about it, it is the USA that is reinventing the wheel here. No?
8

Fairfax,

08/01/2008 17:52:37
An English Voice (8): 5. "Some more facts about this system and the state of supercomputing in the UK:"

I don't disagree with your comments but, much as I dislike Galactic Cannibal's phraseology, he does have a valid point, if he's referring to the Texas Advanced Computing Center (TACC). I used the jargon "Beowulf cluster" in my original post, which I should have defined: it means that many fairly standard PCs are networked to form a parallel computer. The TACC is particularly well-placed to do this because Dell is based in Austin, but we could probably have followed this route more economically. You can read about both here

http://www.tacc.utexas.edu/general/overview/

and

http://www.beowulf.org/

Anway, that's enough grinding of axes . . .
9

GalacticCannibal,

Murrieta 08/01/2008 18:40:29
9
Fairfax,
08/01/2008

Dude I am not interested in being politically correct , so i come across crassly most of the time. But i don't care. I try to stick to known and verifiable facts or evidence.

I was referring to the Sup Computers operating at UT Austin Texas.

Those Dudes at UT would be happy to sell processing time or get Dell to build a multi parallel system for ur Scottish endeavors. There is no shame in that, but it would have saved the UK Tax payer about £50 million


Yes I am aware the skeleton of a Computer was first developed in England. But multi generations later and until Wm.Shockley Bell Labs. invented the new technology of the transistor. The UK skeleton computer was going no where, just a massive elephant using valves, with macro power output, and requiring massive cooling and was unreliable.

Dude don't panic we are between ICE ages and we will not live to feel the next one Dude.

GC
10

icehawk,

Montana, USA 08/01/2008 18:42:06
The TACC system Ranger will be one heck of an interesting system. No doubt about that.

As someone who has built and maintained several 10+ Teraflop Linux clusters, and also built and maintained several 10+ Teraflop IBM clusters I have some experience.

First off, the IBM will cost you about 3 times as much money as a similar "Peak Performance" Linux cluster. This will be a similar situation for a Cray like HECToR. Sounds great for a linux cluster, more teraflops per dollar, what can go wrong with that?

Well, a Linux cluster of simular size, IE 10+ Teraflops, will take about 3 times the people to maintain and will have about 1/4th the throughput for a users code, and have a reliability that is at least a magnitude of order less. It's getting better, but the high speed interfaces just can't compete with the dedicated systems made by IBM, Cray, SGI, etc. let alone the RAS (Reliability and Servicability) of a more expensive system. Software stacks don't support 10,000+ processors, compilers don't optimize code very well, etc etc.

If someone wants to impress management they can say they bought a 10+ Teraflop Linux system for less than 500k pounds. However, if they don't care about impressing management and want instead to impress the scientists and users, they pay more and step up to a system like HECToR.

One last thing. The 113 Million pounds is for everything. The computer, the building, people to maintain it, maintenance costs, upgrades, etc etc. That's the cost for the WHOLE project, not the cost for the computer. Most of the numbers you will see for a large supercomputer purchase will show just the cost of the computer itself and nothing else. It can give a false sense of price.

Icehawk
11

Fairfax,

09/01/2008 10:53:11
Montana (11): "Well, a Linux cluster of simular size, IE 10+ Teraflops, will take about 3 times the people to maintain and will have about 1/4th the throughput for a users code, and have a reliability that is at least a magnitude of order less."

Agreed. In my post (1) I was suggesting using the money to build many medium-sized clusters, rather than one large one. There are two advantages to this: (i) less precious resources can be used by general PhD students, without the bother of applying for time: they can play with algorithms in a way that cannot be countenanced with a national resource, and (ii) I'm simply not convinced that the computer will do anything more than run poorly designed larger versions of existing computational fluid dynamics code. In my experience, greater computing power is often wasted, whilst more limited facilities can engender ingenuity.

"That's the cost for the WHOLE project, not the cost for the computer."

I don't believe the figure includes future salaries for staff, so the total cost will be higher, although this would be relatively small.
12

anchovy1,

Edinburgh 09/01/2008 12:39:01
Thankfully, the University of Edinburgh do believe it includes salaries for staff, because one of them's mine.

As my colleague Icehawk mentioned, that's a large figure because that's the grant from the Research Council for the whole project over 6 years. It pays for dedicated scientific support staff, system administrators, electricity, site improvements, technology updates including a quadrupling of capability next year, another upgrade in 2011, and no doubt other stuff I've forgotten. Feel free to peruse the HECToR website for information.
13

Fairfax,

09/01/2008 12:52:52
anchovy1 (13): "Thankfully, the University of Edinburgh do believe it includes salaries for staff"

Fair enough: my error. I'm still very much the sceptic, despite the full funding. Here in Cambridge we've had a Hitachi supercomputer which was similarly praised initially, but the actual effect on applied mathematics seems to have been small. I hope your experience is better.
14

icehawk,

Montana, USA 09/01/2008 15:22:01
Fairfax(12): "I was suggesting using the money to build many medium-sized clusters, rather than one large one"

Don't get me wrong, and rereading my original comment it looks like I'm against Linux Clusters. I'm actually not, they are a great way to get small and medium scale HPC systems built for Universities and research. (it's strange to think 10+ Teraflops as being small ro medium. Wow time flies) Linux clusters have been a great boon for HPC, and with the proper staff and talent it's possible to build a big system like Ranger.

However, they are not the only, nor the best solution for a lot of large scale research projects.

I agree that trying to get time on HECToR might be a nightmare for a Phd student. But, that's not what HECToR is for. It's for those researchers whose code needs to have a system so nightmarishly big that they can't effectively run on smaller systems.

In the US we call these projects "Grand Challenge Projects" and the Government grants time on the largest supercomputers so that these projects have a chance of moving forward. (Some of the projects were pretty cool)

If HECToR was built for just the University of Edinburgh, or for a specific problem, or (fill in the blank), then yes, a single or multiple linux clusters might have been the best way to do things, but that's not what it was designed and built for.

Just my thoughts.

Icehawk
15

icehawk,

Montana, USA 09/01/2008 15:25:51
Fairfax (12): I'm simply not convinced that the computer will do anything more than run poorly designed larger versions of existing computational fluid dynamics code. In my experience, greater computing power is often wasted, whilst more limited facilities can engender ingenuity.

By the way. There is something to this thought. I've seen some pretty poor code run on some pretty big systems. I guess that's one of the prices that get's paid to have a few really "good" projects get a chance to run.
16

Fairfax,

09/01/2008 16:59:30
icehawk (16): "I guess that's one of the prices that get's paid to have a few really "good" projects get a chance to run."

That's a good point. I agree that any facility of this type will have an inevitable waste factor. I suppose I've simply refereed too many poor proposals. for which the main problem was ignorance of modern numerical analysis, rather than lack of computing power.
17

icehawk,

Montana, USA 09/01/2008 17:50:10
Fairfax (17): "the main problem was ignorance of modern numerical analysis, rather than lack of computing power."

You solve that problem and you're going to be a very rich person. I think that's been a serious problem since computers were first built. Why have better code when I can just use a faster computer? Uhg. I think it's one of the biggest travesties in HPC.

A good friend makes a living trying to get bad code to run "better" on these large systems. I'm just glad I don't have his job. I'd be bald from ripping my hair out. (-:

If you're a good numerical person that truly understands the underpinnings of a large HPC system, then you will always have a job trying to help a researcher get better performance from his code.

 

Comment on this Story

 

In order to post comments you must Register or Sign In

 
 
RSS Feed of this article's comments
Delivery formats: RSS, Atom, Javascript & Email
 
  

 
 


Sister Newspapers:
Press Complaints Commission

This website and its associated newspaper adheres to the Press Complaints Commission’s Code of Practice. If you have a complaint about editorial content which relates to inaccuracy or intrusion, then contact the Editor by clicking here.

If you remain dissatisfied with the response provided then you can contact the PCC by clicking here.