Comments on: Supercomputer ranking method faces revision

On the eve of a new list ranking the world's fastest computers, momentum is building for a change in the measuring process.

[crypto k. colloid]

Message has been deleted.

[olePigeon]

I wonder why...

I wonder why AMD and Intel (least of all, Intel) are always
credited with breaking away from the "MHz myth," when it was
Apple who fought tooth and nail to convince the market and
industry analysts that MHz isn't everything.

Then I remember that it's C|Net, and they'd discredit Apple even
if they discovered Cold Fusion. :/

. I say let them develop and use the HPC Challenge Benchmark.

IBM suggests that no one understands the HPC Challenge Benchmark and suggests that the Linpack should remain the test of choice. There is also mention that the customer can just simply try their software out on the system (before they buy). That would be a little hard since a lot the systems are builded to order and are still on paper (IBM has been researching the Blue Gene/L since 1999 and is still being developed). I suppose that IBM could offer a refund if, once, they build the system and it does not perform in accordance with the Theoretical peak performance although I doubt it since IBM has poured $100 million into Blue Gene's development.

Additionally, since the government does funding for these large systems, they should be free to design tests to test the entire architecture before investing millions into any one program. Research facilities are in such a rush to have the fastest system and one of the main factors used in deciding on a particular product is based on the speed of current systems in the marketplace. So, the Top500 list becomes a player in the decision, and currently the test of choice is the Linpack. The Linpack test has been used to set the bar and to judge our latest and greatest systems. At some point we stopped thinking and became focused on developing machines that have favorable markings using the Linpack ratings. While this was happening, Japan developed the Earth Simulator. In short, while we were beating our chests were getting kicked in the tail. I say let them develop and use the HPC Challenge Benchmark.

[steve_dupuis]

Its amazing how guy things always get reduced to the same thing ...

After reading the story, it slowly dawned on me that all the fuss about the newer tests was getting away from the ubiquitous machismo male thing of 'bigger is better'. The Linpack test gives out a simple number, something that marketing guys can get both hands around - that they don't have to think about. Either yours is bigger or its not. It doesn't matter how much money is involved - it always comes down to the same juvenile purile nonsense.