Category: IT|Jul 26, 2017 | Author: Admin

New memory technology provides up to 30 percent faster processors

• Hardware
• processor
• permalink

Share on

 Facebook |  Twitter |  LinkedIn |  Reddit

New memory technology provides up to 30 percent faster processors

New way to remember. A new memory technology from MIT turns down on how applications use the processor's own memory and can significantly improve performance in multi-kernel systems.

 

Organizes cache better In recent years, performance improvements in processors have come from adding additional cores, but the new Jenga technology from MIT instead further develops how the kernels access data in the cache - a memory type that is located on the processor itself. Today's pieces usually use four kernels and have three or four levels of cache on a few megabytes each, where the memory gets bigger the farther away from the kernel it is. It may be faster for a processor core to retrieve data from the local cache than to go to another kernel, so the way this is used is key to increasing the performance of processors, MIT writes. Configured "on the go"

 

"What you want to do is take these distributed physical memories and build software hierarchies that maximize performance for single programs," said Daniel Sanchez, an assistant professor of the Department of Electrical Engineering and Computer Science (EECS) at MIT, Which is the group that created the new system.

 

The Jenga system can be configured "on the go," according to the needs of the running program, as some applications need access to a large, flat memory base, while others need a more hierarchical organization. Jenga calculates which organization memory should have in order to find the best access time, and the benefits should be great. Increased performance by 20 to 30 percent The system increased processing speed by 20 to 30 percent in a 36-core processor simulation and reduced the energy requirement by 30 to 85 percent.

 

"Jenga is different in that it uses software to try to characterize what the workload is, thus making an optimal allocation of resources between competing processes," said David Wood, a professor of computer science at the University of Wisconsin at Madison. Of course, it is left to see if technology comes from simulation stage and into processors from, for example, Intel and AMD. It costs money to build new processors, and it takes years to develop and rebuild new technology, so it may at least take a while before Jenga comes into the hands of consumers.

Comments: