REGISTER | LOGIN
Breaking News
News & Analysis

U.S. Angles to Retake Supercomputer Lead

11/29/2017 00:01 AM EST
5 comments
NO RATINGS
Page 1 / 2 Next >
More Related Links
View Comments: Newest First | Oldest First | Threaded View
realjjj
User Rank
CEO
Re: Top 100
realjjj   12/1/2017 9:42:52 AM
NO RATINGS
There many more variables, for example in China some of those might be used as web servers. However, a good metric would be the Green500, so efficiency. The ones investing in new machines , the ones with more appropriate hardware, will lead in efficiency. If you look at Japan, they are clear leaders in efficiency and that's very telling. The US is about suppliers with too much legacy and shady business deals. https://www.top500.org/green500/lists/2017/11/

Les_Slater
User Rank
Author
Re: Top 100
Les_Slater   12/1/2017 8:58:56 AM
NO RATINGS
Not necessarily a U.S. win, just as Linpack is not the best metric for measuring performance, the distribution of Linpack scores probably is not a good measure of high performance computing within a country. The sum of all the top Linpack scores within the Top500 of a country may be more important than the distribution of a few exceptional machines.

luting
User Rank
Author
Top 100
luting   11/30/2017 4:50:30 PM
NO RATINGS
Just forget about Top 500. Instead we could use Top 100, US has over 30+ at Top 100 while China only has 4. US win!! 

R_Colin_Johnson
User Rank
Author
Re: High Performance Linpack
R_Colin_Johnson   11/30/2017 9:26:26 AM
NO RATINGS
Yes, high-performance Linpack was invented when we thought parallel programming could eventually automatically divide-and-conquer tasks for linear speedups. But today we know that is not possible. Top500 and other standards groups need to put together a new benchmark otherwise that Chinese will just keep throwing money at adding CPUs that they should be spending on improving their citizens lots-in-life.

Les_Slater
User Rank
Author
High Performance Linpack
Les_Slater   11/30/2017 6:49:04 AM
NO RATINGS
R. Colin,

Good article. No doubt U.S. will regain crown from China but that will not be the end of it. To me, the real issue here is the obsolescence of the High Performance Linpack benchmark. not only is a new benchmark needed but a study, even a discipline, regarding the relationship of processing, storage, I/O and communication resources to the application of real workloads.

I've been proposing, for about three decades now, that a new discipline be created, a formal Problem Space Geometry, where all attributes of problems be investigated, categorized and relationships between such attributes be systematized.

Les  

Most Recent Comments
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed