NEC Regains Supercomputer Title

[nutball]

Instead of measuring the powerfullness of supercomputers per nodes or processors, why not measure them per MHz's :?
...like "how many instructions can the Blue Gene process per one MHz"

...or "per transistor"?. "How many transistors does the Blue Gene need to process ten commands?"

I think the key metric is nothing to do with transistors, nodes, processors, MHz.

Take the reciprocal of the product of the time taken to solve your problem, and the number of dollars spent. The guys with the highest score is Teh W1n!!!

 

[Guden Oden]

AFAIK the Blue Gene L's individual prototype processors run at 700MHz. The SX-8 is around there or a little lower.

The earth simulator CPU (SX-6? Don't remember the designation for that chip and too lazy to look it up. ) runs at 500MHz with some internal clock-doubling the official site says. If this chip runs at 700, it's a very minor bump. Also, power draw for the E-S CPU is only 120W peak, which isn't that much by supercomputer standards. Heck, it doesn't even need direct-die phase change fluorinex cooling! It manages with a standard heatsink/fan unit...

London-Boy, you should check out the official site by the way. It's quite impressive, and there's info there on what kind of stuff they use that overgrown HP48 pocket calculator for... It's easy to find using Google.

 

[pcchen]

It looks like SX-8 runs at 1GHz (its vector units run at 2GHz).

 

[PC-Engine]

It looks like SX-8 runs at 1GHz (its vector units run at 2GHz).

Yep you're right.

Further-enhanced single-chip, vector processor
The vector processor (vector and scalar units) is integrated into a single chip by applying leading-edge CMOS technology with 90-nanometer (nanometer: 10-9 meter) copper interconnects and the most advanced LSI design technology. Pipelines of the vector unit, the central part of a vector processor, operate at a 2GHz clock frequency, which is double the speed of the SX-6, and realize a peak vector performance of 16GFLOPS per CPU.

In addition, a 262TB/s (36.8TB/s in SX-6) high peak data transfer rate between CPU(s) and memory is realized, and it also boasts an enlarged memory capacity of up to 64TB (16TB in SX-6).

compared to the SX-6,

Each AP consists of a 4-way super-scalar unit (SU), a vector unit (VU), and main memory access control unit on a single LSI chip. The AP operates at a clock frequency of 500MHz with some circuits operating at 1GHz. Each SU is a super-scalar processor with 64KB instruction caches, 64KB data caches, and 128 general-purpose scalar registers. Branch prediction, data prefetching and out-of-order instruction execution are all employed. Each VU has 72 vector registers, each of which has 256 vector elements, along with 8 sets of six different types of vector pipelines: addition/shifting, multiplication, division, logical operations, masking, and load/store. The same type of vector pipelines works together by a single vector instruction and pipelines of different types can operate concurrently. The VU and SU support the IEEE 754 floating-point data format.


1 Chip LSI : 8Gflops
0.15µm CMOS
8Layers copper interconnection
20.79mm x 20.79mm
60 million transistors
5185 pins
Clock Cycle: 500MHz(1GHz)
Power Consumption: 140W (Typ.)

5185 pins?

 

[London Geezer]

5185 pins?

Yep, that's probably more than Pinhead in Hellraiser............

 

[Guden Oden]

5.1k pins is probably how that thing gets 36GB/s bandwidth, by a serious amount of on-chip memory channels. The photos of the mainboard on the earth simulator shows 16 DIMM sockets as I recall, that's probably the explanation.

The SX-8 probably uses DDR memory to reach such a high bandwidth...

 

[PC-Engine]

NASA's Columbia Supercomputer Is World's Fastest
Tuesday, Oct 26 @ 12:59 PDT
Silicon Graphics with NASA confirmed that NASA's new Intel Itanium 2 processor-based Columbia supercomputer is the most powerful computer in the world. Only days after NASA completed installation of Columbia and using just 16 of Columbia's 20 installed systems the new supercomputer achieved sustained performance of 42.7 trillion calculations per second (teraflops), eclipsing the performance of every supercomputer operating today.

Built from SGI Altix systems and driven by 10,240 Intel Itanium 2 processors, Columbia's 16-system result easily tops Japan's famed Earth Simulator, rated at 35.86 teraflops, and IBM's recent in-house Blue Gene/L experiment, rated at 36.01 teraflops. Columbia's record results were achieved running the LINPACK benchmark on 8,192 of the NASA supercomputer's 10,240 processors. Columbia also achieved an 88 percent efficiency rating on the LINPACK benchmark, the highest efficiency rating ever attained in a LINPACK test on large systems.

While LINPACK is popular as a yardstick of supercomputing performance, NASA is primarily interested in how the Columbia system will revolutionize the rate of scientific discovery at the Agency.

"Benchmarks are useful for confirming that Columbia is meeting our performance expectations, but the numbers we find most significant are something else altogether," said Walt Brooks, division chief, Advanced Supercomputing Division, NASA. "For instance, we find the number five to be significant. This is because with Columbia, scientists are discovering they can potentially predict hurricane paths a full five days before the storms reach landfall ? an enormous improvement over today's two-day warnings and one that may present huge advantages for saving human life and property."

"Also significant is the number one," added Brooks, "because with just one of Columbia's 20 Altix systems, we've reduced the time required to perform complex aircraft design analysis from years to a single day."

NASA's Columbia Supercomputer is World's Fastest/2
"Unlike other recent supercomputer speed announcements, the Columbia world speed record was attained on a system that is already fully in use at a customer site," said Dave Parry, senior vice president and general manager, Server and Platform Group, SGI. "We're delighted to see the efforts of NASA, SGI and Intel deliver such remarkable results, not only in terms of benchmark superiority, but in the creation of a system that is changing the very nature of science."

Shattering long-held assumptions about supercomputing deployment, Columbia was built and installed in fewer than 120 days, and was available to scientists throughout its installation. In fact, scientists from NASA centers and universities throughout the US used new Altix systems within days after they arrived at NASA. Columbia is already having a major impact on NASA's science, aeronautics, and exploration programs, in addition to playing a critical role in the Space Shuttle Return to Flight activity.

NASA unveiled new details of its Columbia supercomputer in a dedication ceremony today at NASA Ames Research Center in Mountain View, Calif.

Signaling a new era in deployable supercomputing technology, the Linux OS- based Columbia system is built from the same industry standard, commercially available Altix systems that have been in widespread use throughout the world since SGI introduced Altix in January 2003. Leading automakers, consumer product manufacturers, energy companies, pharmaceutical companies, national laboratories, government agencies and research institutions have adopted the SGI Altix platform, which can scale from 4 to 512 processors in a single system.

http://www.supercomputingonline.com/article.php?sid=7327

 

[PeterAce]

If your interested in these type of systems you might want to see the Stanford Computer Systems Laboratory Colloquium Presentation called:

petaFLOP/s systems and changing the nature of science, How I Learned to Stop Worrying and Love the petaFLOP.

Abstract :

http://www.stanford.edu/class/ee380/Abstracts/040331.html

Here it is online:

http://stanford-online.stanford.edu/courses/ee380/040331-ee380-100.asx

 

[DemoCoder]

Just wait till the NSA's RSFQ system is finished (if we ever hear about it)

 

[Polarbear53]

But will it run Doom 3?!

*realizes doom 3 is out, and things can run it*



*leaves thread*



lol, just remembering when lots of people use to say that

 

[pcchen]

The Columbia's full LINPACK result (with 20 nodes) is said to be about 51 TFLOPS. The 16 nodes test results is probably made to meet the deadline of TOP500 list