Traversed Edges per Second Measure
(Redirected from TEPS)Jump to navigation Jump to search
- AKA: TEPS.
- See: LINPACK Benchmarks, Supercomputer, Computer Cluster, Petascale Computing, Graph_(Abstract_data_type), Benchmark (Computing), Graph500.
- (Wikipedia, 2015) ⇒ http://en.wikipedia.org/wiki/Traversed_edges_per_second Retrieved:2015-5-10.
- The number of traversed edges per second that can be performed by a supercomputer cluster is a measure of both the communications capabilities and computational power of the machine. This is in contrast to the more standard metric of floating point operations per second (FLOPS), which does not give any weight to the communication capabilities of the machine. The term first entered usage in 2010 with the advent of petascale computing, and has since been measured for many of the world's largest supercomputers. In this context, an edge is a connection between two vertices on a graph, and the traversal is the ability of the machine to communicate data between these two points. The standardized benchmark associated with Graph500, as of September, 2011, calls for executing graph generation and search algorithms on graphs as large as 1.1 Petabyte. The ability of an application to utilize a supercomputer cluster effectively depends not only on the raw speed of each processor, but also on the communication network. The importance of communication capability varies from application to applications, but it is clear that the LINPACK benchmarks traditionally used for rating the FLOPS of supercomputers do not require the same communications capability as many scientific applications. Therefore alternative metrics that characterize the performance of a machine in a more holistic manner may be more relevant for many scientific applications, and may be desirable for making purchasing decisions.
- (Grace, 2015) ⇒ Katja Grace. (2015). “Brain Performance in TEPS." Personal Blogpost.
- QUOTE: We can use Traversed Edges Per Second (TEPS) to measure a computer's ability to communicate information internally. We can also estimate the human brain's communication performance in terms of TEPS, and use this to meaningfully compare brains to computers. We estimate that the human brain performs around 0.18-6.4 * 1014 TEPS. This is within an order of magnitude more than existing supercomputers. At current prices for TEPS, we estimate that it costs around $4,700 - $170,000/hour to perform at the level of the brain. Our best guess is that "human-level" TEPS performance will cost less than $100/hour in seven to fourteen years.
- (Merrill et al., 2012) ⇒ Duane Merrill, Michael Garland, and Andrew Grimshaw. (2012). “Scalable GPU Graph Traversal.” In: Proceedings of the 17th ACM SIGPLAN symposium on Principles and Practice of Parallel Programming. ISBN:978-1-4503-1160-1 doi:10.1145/2370036.2145832