Slashdot Mirror
World's Fastest Supercomputer To Be Built At ORNL
Homey R writes "As I'll be joining the staff there in a few months, I'm very excited to see that Oak Ridge National Lab has won a competition within the DOE's Office of Science to build the world's fastest supercomputer at Oak Ridge National Lab in Oak Ridge, Tennessee. It will be based on the promising Cray X1 vector architecture. Unlike many of the other DOE machines that have at some point occupied #1 on the Top 500 supercomputer list, this machine will be dedicated exclusively to non-classified scientific research (i.e., not bombs)."
Cowards Anonymous adds that the system "will be funded over two years by federal grants totaling $50 million. The project involves private companies like Cray, IBM, and SGI, and when complete it will be capable of sustaining 50 trillion calculations per second."
Personally I'm happy to see Cray still making impressive machines. Not every problem can be solved by "divide and conquer" clusters.
As usual, there should be a qualifier as to what is meant by fastest. According to their definition they are, but not according to NEC's, for example.
50 trillion calculations per second.
Wow, that's darn fast.
I wonder if that processing power could be used for rendering like was done by Weta and how the performance could compare to their renderfarm.
This is the sig that says NI (again)
> ...capable of sustaining 50 trillion calculations per second.
:D
Hmm...I wonder if I could borrow it for a few days to give my dnet stats a boost
Wow, 50 trillion calculations per second. Thats almost fast enough to finish an infinite loop in under ten hours.
at an Impresive 67fps on this baby...
Still a whole year until they have a full machine, but the 512-way prototype reached 1.4 TFlops (LinPack). The complete machine will have 128 times the nodes and 50% higher frequency. So even with pessimistic scalability, this will be more than twice as fast.
The article mentions that the new supercomputer will be used for non-classified projects. Does anyone have more exact details of what these projects may involve? Will it be a specific application, or more of a 'gun for hire' computing facility, with CPU cycles open to all comers for their own projects? It would be interesting to know what types of applications are planned for the supercomputer, as it may be possible to translate a raw measure of speed like the quoted '50 trillion calculations per second' into something more meaningful, like 'DNA base pairs compared per second', or 'weather cells simulated per hour'. Are there any specialists in these kinds of HPC applications who would like to comment? How fast do people think this supercomputer would run apt-get for instance? Would 50 trillion calculations per second equate to 50 trillion package installs per second? How long would it take to install all of Debian on this thing? Could the performance of the system actually be measured in Debian installs per second? I look forward to the community's response!
Unfortunately we haven't heard much from them lately (Notice the "last updated" date). I suspect they're still waiting on their G5 xServes.
ScienceSeeker.org
There are still a few computing problems that can't be efficiently split into a large number of subproblems that can be executed in parallel. For those cases, a cluster of small machines won't help.
...because a day later Palm users will massively interconnect to form the World Fastest Clustered Computer Environment. The OS? Linux, of course. .}
Help end the use of Sigs. Tomorrow
That's correct, it's the Department of Energy.
I don't know why they would need it, but that's just because I don't know anything about the work of the DOE (not being an american and all that)
This is the sig that says NI (again)
Yes, DOE is the Federal Government's Department of Energy. Oak Ridge is a large federal govt. lab.
or it certainly seems like it (reading the specs of the thing)
I don't think Crays that were build 5 years ago are considered obsolete by anyone's standards.
Clusters solve different jobs than supercomputers. Sometimes they bleed into one another, but there are some things supercomputers will always be better at (because of higher memory bandwidth for one thing).
They were listed as part of the solution.
Oak Ridge has done extensive evaluations of recent IBM, SGI and Cray technology. Though I am still looking forward to data on IBM's Power5.
Cray X1 Eval
SGI Altix Eval
If you care to, read the pdf on their early impressions of the X1. The Army High Performance Computing Research Center (www.arc.umn.edu) did an analysis of their application and found that the X1 was actually MORE cost effective than a commodity cluster.
Firstly, the X1 was greater per-processor performance by a factor of 4. Then you add an interconnect that has half the latency, and 50 times the bandwidth of myrinet or infiniband. It also has memory and cache bandwidth enough to actually fill the pipelines, unlike a Xeon which can do a ton of math on whatever will fit in the registers. Some problems just don't work real well on clustered PCs, they need this kind of big iron.
Secondly, some problems cannot tollerate a failure in a compute node. IF you cluster together 10,000 PCs, the average failure rate means that one of those nodes will fail about every 4 hours. If your problem takes three days to complete, the cluster is worthless to you. A renderfarm can tolerate this sort of failure rate, just send those frames to another node. Some problems can't handle it.
Oak ridge is very concerned with getting the most bang for the buck.
I think ORNL and PSC know a lot more about supercomputing than you (or Internet rag pundits) do. As others have noted, there are real reasons for Big Iron.
Clusters are great for certain problems but for heavy computation -- think simulating two galaxies colliding or earthquake modeling -- off the shelf clusters don't cut it.
They're not wasting tax-payer money unless you consider basic researcher a waste.
So each node is directly connected to six ajacent nodes. Contrast this with the Thinking Machines Connection Machine CM2 topology, which had 2^N nodes connected in an N dimensional hypercube. So each node in a 16384 node CM2 was directly connected to 16 other nodes. There's a theorem that you can always embed a lower dimensional torus in an N dimensional hypercube, so the CM2 had all the benefits of a torus and more. This topology was criticized because you never needed as much connectivity as you got in the higher node-count machines, to CM2 was in effect selling you too much wiring.
Thinking Machines changed the topology to fat trees in the CM5. One of the cool things about the fat tree is it allows you to buy as much connectivity as you need. I'm really surprised that it seems to have died when Thinking Machines collapsed. On the other hand, any kind of 3D mesh is probably pretty good for simulating physics in 3D. You can have each node model a block of atmosphere for a weather simulation, or a little wedge of hydrogen for an H-bomb simulation. But it might be useful to have one more dimension of connection for distributing global results to the nodes.
--- Often in error; never in doubt!
Remember, DOE is a tax-payer funded agency. For my money, the G5 solutions looks better!
Self Defense - A Human Right www.a-human-right.com
Didn't Cray make some comparison about supercomputers vs clusters being like a tractor trailer vs a fleet of honda civics?
The civics might be fine for couriers, but if you need to move - say - an elephant they're useless.
Analogies suck, though, and I'm pretty sure I got that one wrong.
I don't need no instructions to know how to rock!!!!
Clusters are not the be-all end-all of supercomputers. Clusters are really only effective if you have a problem that can be paralellized -- or split into multiple parts that can each be worked independently of one another and then merged into a single result. Factorization, rendering, etc. are all examples of easily paralellized operations.
Certain operations, though, are highly dependant upon each previous result. Physics and chemical simulations are a good example. When you have situations like this, clusters don't do you a lot of good, since only one iteration can be worked on at a time -- leaving most of your cluster sitting there idle.
Wow, that's the first Beowulf cluster comment I would mod as interesting.
Sigs? We don't need no stinking sigs!
I worked in Instrumention and Control for the Free Electron Laser project at the Thomas Jefferson National Accelerator Facility. We also host the CEBAF (Concentrated Electron Beam Accelerator Facility), which is a huge ass particle accelerator.
the DOE does a lot of basic research in nuclear physics, quantam physics, et cetera. the FEL was used to galvanize power rods for VPCO (now Dominion Power) and made them last 3 times as long. Some William & Mary people use it for doing protein research, splicing molecules and stuff.
The DOE does a lot of very useful things that need high amounts of computing power, not just simulating nuclear bombs (although Oak Ridge does taht sort of stuff, as does Los Alamos). We only had a lame Beowulf cluster at TJNAF. I wish we would have had something like this beast.
I want to know how it stacks up to the Earth Simulator.
It seems to me that as long as multiprocessor machines qualify as supercomputers, then the Google cluster counts as the fastest right now, and will still count as the fastest long after this new DOE computer is built.
Certain operations, though, are highly dependant upon each previous result. Physics and chemical simulations are a good example. When you have situations like this, clusters don't do you a lot of good, since only one iteration can be worked on at a time -- leaving most of your cluster sitting there idle.
Umm, bwah?
It's only going to be sitting there idle if you're not properly scheduling and qeueing jobs. Also, you -CAN- do the kind of simulations (Physics, chemicals) on a cluster *points at clusters at Chrylser and Shell*. The caveat is that you need to write out the result for the appropriate job to handle (in practice - job run 1 contains step 1, job run 2 step 2, etc). And a cluster is perfectly fine for this.
That all said - a supercomputer like this -IS- generally a better tool for the job if you've got the money. Money, in most places, -IS- an object, so we get the best bang for our buck.
*shrug*
Supercomputers usually run some flavor of UNIX -- Unicos, IRIX, I think even Linux. In any case, they are specially built and designed for the supercomputer. Supercomputers are used for highly specialized scientific applications, and as such the programs would be specially written in Fortran, C, or Assembly, and often specially optomized for the architecture.
-James
The SGI altix runs a hacked up version of linux that's part 2.4 with a lot of backported 2.6 stuff as well as the Irix scsi layer. They are migrating to a pure 2.6 OS soon. The IBM system runs AIX 5.2. The Cray runs Unicos, which is a derivative of Irix 6.5, though they seem to be moving to Linux also. I'm gonna geuss that they run totalview as their debugger. They use DFS as their network filesystem. They have published plans to hook all these systems up to the Stornext filesystem which does Heirchical Storage Management. MPI and PVM are likely important libraries for a lot of their apps.
For these sorts of machines, one can by utilities for data migration, backup, debugging, etc. However, the production code is written in-house, and that's the way they want it. Weather forcasting, for example, uses software called MM5, which has been evolving since the Cray-2 days, at least. A lot of this code is passed around between research facilities. It's not open source exactly, but the DOD plays nice with the DOE, etc.
The basic algorithms have been around for a long time. In the early 90's, when MPPs and then clusters came onto the schene, a lot of work was done in structuring the codes to run on a large number of processors. Sometimes this works better than other times. Most of the work isn't in writing the code, but rather in optomising it. Trying to minimize the synchronous communication between nodes is of great importance.
The number of processors isn't as important as the memory architecture. Clusters of workstation-class machines have isolated memory spaces connected by I/O channels. Many non-clustered supercomputers have a single unified memory space where all processors have equal access to all of the memory in the system. This can be important for algorithms that heavily use intermediate results from all parts of the problem space.
Even so, for a given number of FLOPS, a vector machine would generally require fewer CPUs than a cluster of general-purpose machines. This reduces the amount of splitting that has to be done to the problem in the first place.
There are still a few computing problems that can't be efficiently split into a large number of subproblems that can be executed in parallel. For those cases, a cluster of small machines won't help.
(Score:-10, Wrong)
I'm sorry dude, but this macine is going to have more than 1 CPU in it, and the work will have to be split among the processors and ran in parallel.
(Score:-100, Wronger)
Sorry, but you have it all wrong. The parent is right. The parent stated that there are problems that can't be split in smallest problems for being handled by a cluster of computers. A cluster is a set of computers that work independant of each other and have the ability ro comunicate at ethernet speeds (10 - 100 - 1000 Mbits / Sec). There are problems that cant be solved using this approach, for example calculations where all processes must reuse the same data; with really big data sets the network connections become bottle-neck.
For those kinds of problems (the usual example is a simulation of a nuclear explosion, a star system, etc) you need a single machine with loads of processors sharing the same memory space. That's where supercomputers come to play.
Life isn't like a box of chocolates. It's more like a jar of jalapenos. What you do today, might burn your ass tomorrow.
Big Mac was tested in a small 128 node configuration as a prelude to the full 1100 nodes.
The 128 node cluster was benchmarked at ~80% efficiency, or ~1.6 Teraflops. The final cluster achieved a RMax of 10.28 TFlops, ~60% of the 17.6 TFLOP theoretical peak.
A 6000 node cluster would be very difficult to manage.
The important part of the statement is "Sustaining". There are a lot of computers out there on the top500 list that get peak numbers way ahead of their sustained numbers. An Army reseach center (www.arc.umn.edu) published a comparison of a xeon cluster and the X1. For their codes (weather simulation, material sciences, air flow, etc) the Xeons sustained performance was 5% of peak. The Cray was about 30% of peak. (this is probably due to the really awesome memory bandwidth of the cray)
You're correct that these are just numbers so lets talk about a real problem. The AHPCRC reported that a 32 processor cray X1 (peak 400 Gigaflops, 66 gflops realized) was able to simulate a weather model of the entire US with 33 vertical levels at 5Kilometer resolution in just under 2 hours. Today these models are done at 10KM resolution with 20 levels. IF you take this theoretical ornl system and assume (peak 60-80TF, 40 sustained on easy codes, 15 sustained on hard codes) then they might do a 2KM simulation with 45 layers in 1 hour.
In the long run one would like to be able to get such simulations from the 10,000 atom level up to the billion-to-trillion (or more) atom level so you could simulate significant fractions of the volume of cells. Between now and then molecular biologists, geneticists, bioinformaticians, etc. would be happy if we could just get to the level of accurate folding (Folding@Home is working on this from a distributed standpoint) and eventually to be able to model protein-protein interactions so we can figure out how things like DNA repair -- which involves 130+ proteins cooperating in very complex ways -- operate so we can better understand the causes of cancer and aging.
Thank you for your understanding in this matter,
Your friendly neighbourhood IRS agent.
Sorry, it looks like the URL has changed. The home page for Folding@Home is here.
I believe the speed was due to many factors. Here are a few.
--- Often in error; never in doubt!
well, 0-4 are all true.
comparing this to early crays is a little difficut though. For the early crays one advantage was vectors and the other was pipelines.
vector processors are cool, because they tend to be much more tolerant of the latency. You issue a load command, and it does loads until the vector-register is full. Equivalent to dozens of loads (and dozens of round trip latency to memory) on a scalar architecture. The same thing applies to the execution units. You tell the CPU ADD R1 R2 R3, and it pumps the first elements of R2 and R3 registers through the ALUs and into R1 and keeps working until it gets through all of the elements in the vector. Later models supported chaining, which allowed the output from one of these operations to feed into the input of another operation. Vector CPUs are very good at keeping the ALUs busy.
The other advantage of the early crays was pipelining. YMP designs, for example, had multiple integer, FP, load/store, and reciprical devide units. All of these (and the dispatch unit) were pipelined, allowing a munch higher clock rate than traditional designs. Multi-pipeline designs are now the norm, (powerPC, Pentium, MIPS, etc.) but were pretty amazing at the time.
The cooling, incidently, was necessary at any clock rate. Early Crays. (well right on through to the T90) used bipolar transistors, rather than CMOS. In this sort of logic you switch current rather than switching voltage. The net result is that the early crays used a TON of electricity and needed massive cooling systems.
This project claims many big improvements. First, programmers will be available to help parallalize code of scientists, who may be experts at, say, weather or protein folding but may not be experts at parallel code. Further, the facility is supposed to be open to all scientists from all countries and funded by any agnecy. CPU cycles are to be distributed on a merit-only basis, and not kept witin DOE for DOE grantees to use, as apparently has happened within various agencies in the past.
_ ___
The idea is to make it more like other national labs where - for example in neutron scattering - you don't have to be an expert on neutron scattering to use the facility. They have staff available to help and you may have a grant from NSF or NIH but you can use a facility run by DOE if that's the best one for the job.
I attended this session at the American Physical Society meeting this March and I'm assuming this is the project referred to in the talks - I apologize if I'm wrong there, but this is at least what is being discussed by people within DOE. I'm essentially just summarizing what I heard at the meeting so although it sounds like the obvious list of things to do, apparently it has not been done before.
The prospect of opening such facilities to all scientists from all nations is refreshing during a time where so many problems have arisen from lack of mobility of scientists. For example, many DOE facilities such as neutron scattering at Los Alamos (LANL) have historically relied on a fraction of foreign scientists to come and use the facility and this helps pay to maintain it. Much of this income has been lost and is not being compensated from other sources. Further, many legal immegrants working within the Physics community have had very serious visa problems preventing them from leaving the country to attend foreign conferences. The APS was held in Canada this year and the rate of people who could not show up to attend and speak was perhaps ten times greater then the APS conferences I attended previously. Although moving it to Canada helped many foreign scientists attend, it prevented a great deal of foreign scientists living within the US from going. Even with a visa to live and work within the US, they were not allowed to return to the US without additional paperwork which many people had difficulty getting.
Obviously, security is heightened after 9/11, as it should be. I'm bringing up the detrimental sides to such policies not to argue no such policies should have been implemented, but to suggest the benefits be weighed against the costs - and the obvious costs such as to certain facilities should either be compensated directly or we should be honest and realize we are (indirectly) cutting funding to facilities which are (partly) used for defence in order to increase security.
I mention LANL despite it's dubious history of retaining secrets because I have heard talks by people working there (this is after 9/11) on ways to detect various WMD crossing US boarders. Even though they personally are (probably) well funded, if they facilities they need to use don't operate any more this is a huge net loss. My understanding is that all national labs (in the US) have had similar losses from lost foreign use.
___________________________________________
a war on terrorism? How can we end a war on a method?
"Any tin-foil hats should be directed at Y-12. That's the DOD plant; X-10 is just DOE."
You're right, but let me clarify something:
The biggest weapons labs in the country are DOE, not DOD facilities. These are the "tri-labs": Los Alamos, Lawrence Livermore, and Sandia. They are operated by the DOE's NNSA (National Nuclear Security Administration).
The other major DOE labs (including ORNL) are operated by the DOE's Office of Science. These are non-weapons labs. For you conspiracy theorists out there, its pretty obvious that these are non-weapons labs. No guys standing around with M-16's etc., as you would find at a place like Los Alamos. Much, much less security.