The Rise and Fall of Dennard Scaling

Published on 2022-05-02

Category: Misc

Dennard Scaling was created by Robert H. Dennard and it stated that “as transistors get smaller, their power density stays constant so that the power use stays in proportion with area: both voltage and current scale (downward) with length.”

Implications of Dennard Scaling

This would mean that for each new generation of processors:

• We could double the transistor count
• We could double the frequency without extracting power usage
• We could more than double the performance for each new generation of processors

The Decline of Dennard Scaling

Around 2004, Dennard Scaling took a turn backward and ultimately was a failure. Single-core CPU performance stopped growing with the performance of multi-cores taking the front seat. Frequency growth also stopped growing.

At the time, Dennard Scaling seemed appropriate, but with the growth of technology in this day and age, it didn’t go forward.

The Rise of GPUs

The failure of Dennard Scaling also sent Graphics Processing Units (GPU) to become huge. A GPU is similar to a CPU, but it is used for graphics applications and applications with heavy SIMD operations. GPUs have high parallelism, but poor data locality while CPU has the opposite making them better for general purpose applications.

Conclusion

Understanding the rise and fall of Dennard Scaling is essential for appreciating the current landscape of processor technologies. As we move forward, the focus has shifted from merely increasing transistor counts and clock speeds to optimizing parallel processing capabilities, as seen with the proliferation of multi-core CPUs and powerful GPUs. This shift underscores the importance of adapting to technological limitations and leveraging alternative architectures to continue advancing computational performance.