Nehalem Performance Preview

April 7, 2009 by

Pages: 1 2 3 4 5 6 7 8 9 10 11 12

System Configuration

Our Asus R12A 1U server came with two X5570 CPUs, running at 2.93GHz with 8MB L3 cache and a 95W TDP. This particular model can increase frequency by three bins (400MHz) if only one or two cores are active, and two bins if three or more cores are active. So a CPU with four active cores could easily be operating at 3.2GHz – especially with aggressive cooling to lower junction temperatures and hence leakage. Our Harpertown server is the same one tested previously with two 3.0GHz E5472 CPUs with a 1.6GT/s bus and 80W TDP.The two servers were configured as indicated in the chart below. Note that the Supermicro is a 2U system that supports up to eight drives, while the Asus is a slender 1U with only four drives at most – so this will account for some of the differences in power. The FB-DIMMs are a pretty big contributor as well – 8 FBD will consume around 40W of extra power.
Chart 1 – Nehalem and Harpertown System Configurations
Unless otherwise stated, performance for each benchmark is the average of three runs. Certain benchmarks were run more than three times (to verify that the observed behavior was correct), while other benchmarks were run fewer than three times (due to the run time of such tests). In any case, we will clearly state when we deviated from our standard practices.We have slightly changed up our selection of benchmarks from our last preview. Sungard Adaptiv Credit Risk and XMLMark were left out, but we have added SPECpower_ssj2008 (in research mode, rather than a fully compliant test). We’d also call attention to the change in the operating system to Windows Server 2008 Enterprise x64, the latest server OS from Microsoft. The next preview (presumably for Westmere or perhaps a LV Nehalem) will be even more interesting, since SP2 for Windows Server 2008 includes some changes to the Windows scheduler that improve thread placement and increase power efficiency and performance (resolving some issues that we ran into here).Our benchmarks include:
  • POV-Ray 3.7 beta 30
  • Valve VRAD
  • Fire Spread Probability model
  • Euler3D
  • Myrimatch
  • SPECjbb2005 v1.07 with Oracle Jrockit 6 P28.0
  • SPECpower_ssj2008 v1.01 with Oracle Jrockit 6 P28.0
  • SPECcpu2006
These benchmarks span a variety of workloads and special thanks go out to a number of individuals for helping us put together this benchmark collection. Henrik Stahl of Oracle, Stuart Brittain at Systems for Environmental Management, Scott Wasson of Tech Report (who helped out with Myrimatch and Euler3D) and Jeff Reilly and his team at Intel for the SPECcpu binaries. There are some other folks who have contributed workloads for Linux, which we will hopefully get to in a future installment.Every benchmark is accompanied by power measurements for the entire server at the wall. We are using a Watts Up Pro meter that was graciously provided to us by the manufacturer, Electronic Educational Devices. The Watts Up Pro meter samples a variety of information including power, voltage, current, and others every second, which can easily be logged with the right software. Smaller tests can be logged using the Watts Up software, but that is limited by the capacity of the power meter’s memory. The Watts Up Pro is simple to set up and we highly recommend it for any casual testing.

Pages: « Prev   1 2 3 4 5 6 7 8 9 10 11 12   Next »

Discuss (52 comments)