Intel Power Consumption Then and Now

By Patrick Schmid
published

Conclusion

The advances between the Pentium 4 and the Core 2 processor generation are tremendous. Not only will Core 2 processors require 1:10 to complete an entire SYSmark 2007 run as compared to 1:35 to 1:40 for the Pentium D and Pentium 4 on our reference test system, but the Core 2 processors require less power in all system states. Normalized to a 3.0 GHz Pentium 4 630 (3.0 GHz), the Pentium D 830 offers a 4% increase in performance per watt, while Core 2 Quad at 3.0 GHz and Core 2 Duo 3.0 GHz offer a 3.6x and 4.2x better performance per watt ratio in our SYSmark 2007 test environment. These numbers were collected by tracking the entire power consumed during the benchmark run.

The total power in watt-hours required to complete a SYSmark 2007 run was cut in half by replacing the Pentium D 830 by a Core 2 Duo E6850 Compare Prices on Core 2 Duo E6850 . Since SYSmark isn't an extremely computation-intensive benchmark, you can expect even more dramatic differences when it comes to rendering, ray tracing, encryption or other forms of CPU-intensive workloads.

Intel's upcoming 45 nm Penryn has shown power requirements of only a few watts in idle, and decreased power requirements under load. This sounds promising, as it should result in an even lower power requirement to complete a given workload, as well as quicker completion time, which again helps to save power and reduce heat dissipation.

Patrick Schmid
3 Comments Comment from the forums
  • hkazemi
    The 'SYSmark Performance per Watt' can be misleading when comparing the dual and quad core processors, particularly when looking at other types of loads that make better use of the additional cores. In particular, look at x264 encoding, where a quad-core processor offers nearly double performance. See the graysky's articles on TechARP for actual charts and tables:
    http://www.techarp.com/showarticle.aspx?artno=442
    http://www.techarp.com/showarticle.aspx?artno=520

    For a load that will use all processors, I believe that testing will show better performance per watt for a quad core compared to a dual (or even single) core because there is little or no additional overhead from the motherboard or powersupply or hard drives when adding an additional core.

    It may be easier to understand this by comparing running 2 separate PCs with dual core CPUs in them vs. running a single PC with a quad core CPU. The quad core system won't need another motherboard, power supply, fans, drives, etc. so all those items are areas of power savings.

    From the published tests above, the 2 extra cores consume 195-132=63 watts at max load and 94-77=17 watts at idle. That is in contrast to adding a 2nd system identical to the E6850 testbox which used 132 watts at max and 77 at idle.

    A simplistic scaling using a spreadsheet, and assuming that performance doubles going from dual to quad cores (reasonable for x264 encodes), shows that:

    state cores watts perf/watt perf perf/core perf ratio
    Idle 2 77 1.46 112.42 56.21 1
    Idle 4 94 2.391914894 224.84 56.21 1.638
    Avg 2 90 1.46 131.4 65.7 1
    Avg 4 112 2.346428571 262.8 65.7 1.607
    Max 2 132 1.46 192.72 96.36 1
    Max 4 195 1.976615385 385.44 96.36 1.354

    In other words, using the system power numbers given, the quad core can be 35% to 64% more efficient than using a dual core, if given an appropriate load. A look at the Sysmark benchmark scores will show you that it did not scale up very much going from dual to quad.
    Reply
  • archp2008
    You said, "You cannot upgrade an existing Pentium D or Pentium 4 system with a Core 2 Duo processor, so you will have to purchase a new motherboard supporting the new processors, as well as DDR2 or DDR3 memory." My wife's P4d is on a Fox 45cmx mobo. The E6850 is listed on the support list. I was thinking of upgrading to the E6850 dual core.
    Reply
  • you sure about pentium D 3.00 GHz having TDP of 215 Watt
    Reply