Posts

Showing posts from November 24, 2009

Multi-Core Slow Down [GENERAL CERTIFICATION NEWS]

Image
By Dan Woods Multi-core processors may slow many applications down, not speed them up. The widespread adoption of multi-core hardware is in many cases actually slowing down computing. A typical scenario: A company has a data-intensive software application and wants to make this software run as fast as possible. When its single-core machine reaches end of life, the company purchases multi-core hardware. To everyone's surprise, the application starts to run slower. For technical support staff, the first question to ask when someone calls with a performance problem is, Have you done a hardware upgrade? Often, when the application is switched back to single-core hardware, the problem disappears. How can this be? Don't more cores mean more computing power? Doesn't more computing power mean software that runs faster? The simple answer is no. Unless the software was written specifically for the multi-core paradigm, multi-core hardware may be a waste of money for increasing

The 'Core' Problem [GENERAL CERTIFICATION NEWS]

By Ed Sperling One of the biggest questions hanging over the processor world for the past few years is what to do with all those cores. This isn't a matter anyone is taking lightly. It's one that carries huge ramifications, because without a compelling answer there's no reason for CIOs to collectively spend billions of dollars on new server hardware with more cores on each chip. The basic problem is that chipmakers can't develop processors the same way they did in the past. If they continue turning up the clock speed with a single core (which is how they got most of the performance increases in the past) they'll literally melt the chip. So they've added lots of cores that run no faster than a single core (and often much slower), using the sum of the performance on multiple cores to achieve increases in performance. Unfortunately, most software doesn't run on multiple cores, so much of that gain is wasted. It runs on one core, or at best two, if some pa

Waking Up Multi-Core Processors [GENERAL CERTIFICATION NEWS]

By Dan Woods " Why Apps Can't Run Faster " explained that just putting more and more cores in chips doesn't increase the processing speed of applications. This is because applications aren't typically designed to take advantage of more than eight cores. To be able to make the most of a multi-core processor, you can run lots of applications on one computer with lots of multi-core processors. Then the operating system balances the workload across all of the cores. But this is much less efficient than running applications that are multi-threaded that balance the work across cores inside the application. This is the art of parallel programming. Read the entire article - Waking Up Multi-Core Processors Technorati Tags: multicore , processor , application , CPU

Why Apps Can't Run Faster [GENERAL CERTIFICATION NEWS]

By Ed Sperling The problem may not be evident at two, four or even eight multicore computer processors. But when an application is written to take advantage of eight cores and the next iteration of that processor has 20 cores, the vast majority of applications will continue to take advantage of only eight cores. Moreover, even if an application is rewritten to utilize all 20 cores, the performance gains will be lower than a decade ago when the clock speed was doubled on a single core processor. And that's a best-case scenario. A word processing application threaded for 20 cores will not run much faster than a version that uses one or two cores. Read the entire article - Why Apps Can't Run Faster Technorati Tags: multicore , processor , application , CPU