Global Sources
EE Times-Asia
Stay in touch with EE Times Asia
 
EE Times-Asia > Interface
 
 
Interface  

The evolution of processor cores

Posted: 16 Jul 2008  Print Version  Bookmark and Share

Keywords:transistor  microprocessor  memory 

Borkar: In history, performance was the king at any cost, but that changed a few years ago when we moved away from using frequency alone to deliver performance.

The first ever microprocessor, Intel Corp.'s 4004, in 1971, integrated around 2,000 transistors to provide an unprecedented level of computational power for the time. It revolutionized the computer world by bringing affordable computers to the masses. Today�s processor cores integrate more than 20 million transistors, supported by cache memories consuming billions of transistors and delivering almost 30,000x more performance than the 4004.

The 8bit processor cores in the early 1970s were simple machines that were not even pipelined. The next major step was the 80386, a full 32bit machine capable of addressing gigabytes of memory, suitable for a high-performance workstation. Subsequently, 80486 followed with an integrated floating-point engine.

The quest for higher performance was not over. In the early 1990s, Intel Pentium processor implemented a superscalar processor, succeeded by Intel Pentium Pro processor, which made an aggressive shift to an out-of-order and speculative processor core. Emergence of more applications and the advent of the Internet brought a voracious appetite for performance that led to deep pipelined Pentium 4 processor cores at multigigahertz frequencies. The subsequent Intel Core and Core 2 processors created a balance between performance and power consumption.

Reducing power
All of these were the result of the silicon technology evolution, from early PMOS to NMOS, then to CMOS to reduce power. Technology scaling regularly doubled the number of transistors on a chip to realize these complex architectures, yet kept power well within the limits.

In history, performance was the king at any cost, but that changed a few years ago when we moved away from using frequency alone to deliver performance. This was simply because the cost in terms of power started to escalate. Increasing transistor frequency meant that its leakage would also increase, which is not a good option. This is the reason current processor cores have lower frequency to reduce power consumption, yet they are still delivering higher compute performance.

Pushing performance
How far can you push a single-core performance? Generally, power consumption increases quadratically with performance. That is, if you double the performance, then power quadruples—not good when you are limited by power.

Hence, the advent of multicore architectures, where you add more cores to deliver higher performance instead of making a single core bigger. Both performance and power of a multicore architecture increase only linearly, making multiple cores on a chip an attractive option. This trend will continue in the foreseeable future.

It will go on as long as you can take advantage of multiple cores to increase application performance. It is true that there is a limit to parallelizing a single application over multiple cores; after some time, there is diminishing return. However, there is more parallelism that you can exploit, such as task level parallelism and application level parallelism. So the multicore trend will continue until you have taken full advantage of homogenous multicores. Heterogeneous multicores are another possibility, where large and small cores coexist, giving added flexibility to choose a core for a given application. But such a system may be a little more difficult to design.

Finally, do not rule out integration of special purpose cores as part of the multicore system. These cores will be designed with a specific task in mind and hence will deliver very high performance at low-power consumption. For example, you could imagine a special-purpose core optimized just for video encode and decode or one for network processing. With the heterogeneous integration of application-specific special-purpose cores with general-purpose cores, the processor will look more like an SoC. This will be a very powerful system for future applications with voracious appetites.

- Shekhar Borkar
Fellow, Corporate Technology Group
Intel Corp.





Article Comments - The evolution of processor cores
Comments:  
*  You can enter [0] more charecters.
*Verify code:
 
 
 
Webinars

Seminars

Visit Asia Webinars to learn about the latest in technology and get practical design tips.

 
 
Back to Top