Processor Breakthrough Could Double Speeds

A new approach to computer processing could double speeds while halving power consumption, according to researchers. The approach could work on existing hardware without needing physical modifications.

The research builds on the existing technique of multithreading. That's an attempt to overcome the major limitation of computer processors (CPUs). They are broken down into cores (often 2, 4 or 8 cores) but each core can only carry out one task at any specific split-second.

Workload Optimized

Threading means breaking bigger tasks down into a list of individual steps known as a thread. The computer then figures out the most efficient order for the processor and cores to switch between the threads to tackle the top task on the list. This means, for example, the processor can switch from one thread to another without having to wait for a response or outcome from the first thread.

Manufacturers built on this approach with variants known as simultaneous multithreading or hyper-threading. This means even more efficient arrangement of the tasks to simulate the effects of having more cores than physical exist.

The latest development is from the University of California Riverside and is dubbed Simultaneous and Heterogenous Multithreading (SHMT). This extends the approach to use the Graphics Processing Unit (GPU) that exists in most computers and is normally reserved solely for creating images more quickly.

AI Hardware Could Help

SHMT does two main things to achieve this task. First, it identifies which tasks are better suited to being assigned to the GPU. For example, as a very generalized rule, a CPU carries less risk of errors while a GPU has the edge in speed through its design. Second, SHMT accounts for, and reduces, the inherent delays in physically sending information back and forth between the two processors. (Source: techexplorist.com)

The researchers say the approach also works when extended to include specialist processors known as AI accelerators. These are used in specialist hardware and approach processing by attempting to recreate the neural networks found in the human brain.

According to the researchers, the main limitations are that the benefits are more significant with more complex tasks rather than very simple ones, and that the general SHMT approach would need to be specifically coded to work with specific hardware. (Source: techspot.com)

What's Your Opinion?

Does this sound like a gimmick or a major breakthrough to you? Would faster processing make a notable difference to your computer use? How important is cutting power consumption to the computing industry?