Home > Intel Communities > Open Port IT Community > Intel Premier IT Professional Zone > Blog > Tags > cpu

IPIP Community Blog

3 Posts tagged with the cpu tag
0

I'm going to be speaking this Thursday at the Intel Jones Farm site in Hillsboro on the technology roadmap. Here's some of the topics we're going to cover:

 

  • Intel processing and manufacturing technology

  • CPU architecture

  • The real value that the first two items bring to IT managers and end users

  • Server roadmap

  • Client roadmap

  • New trends that will be impacting IT over the next 5 years

 

You can find the details on the right hand margin of the web page where you found this blog, or search on "IPIP". If you aren't in Portland you'll see there are another set of events across the country over the next few months. I hope to see you at one!

0 Comments Permalink
3

If you attended the Intel Developer Forum in San Francisco on September 22-24, you may have stumbled upon Booth 425 in the Eco-Technology Community. The theme of the booth was Energy Efficient Data Center Power Management with Windows Server* 2008 Hyper-V* and Intel* Dynamic Power Technologies. The demo was a joint project between Microsoft and Intel to showcase the integration of power management features supported by Windows Server 2008 and Intel-based platforms in a virtualized environment.

 

Logically, development work for the demo consisted in integrating four main technology ingredients: server hardware based on the Bensley platform provisioned with Harpertown CPUs, firmware running on the baseboard of the managed nodes known as Node Manager, virtualized instances of the Microsoft Windows* Server 2008 operating system running on top of the same OS with the Hyper-V virtualization role enabled and a management console application built by Intel, Data Center Manager (DCM).

 

The relationship between the technology components is shown in the figure below:

 

 

I would like to share my personal experience in putting together this demo, an angle not always obvious when looking at the finished product. Personal means people. Let me introduce you to some of the really nice people who participated in the project. This is only a very small portion of the team, so I will apologize up front to team participants not represented here for lack of space.

 

Here is Susmita Nayak , based in California, who officiated as the project manager greeting some of the booth visitors:

 

 

... and here is Haim Cohen, a software engineer based inthe Israel Design Center and yours truly, a technical architect and chief gopher, part of the Oregon team.

 

 

 

The next picture captures an overall view of the booth. It is a pre-conference picture when the setup was almost finished. You may notice the empty boxes, cabling not yet hidden and the rack of servers discreetly tucked on the side. Don't be swayed by rack's apparent small size. The whole rig weighs about 400 pounds (200 kilos). It was shipped to the conference site prefabricated, in one piece. Racking the servers would take about half a day of lifting, bolting, rewiring and sanity testing, which we decided not to do at the conference site.

 

 

 

 

The demo consisted in four 5U SC5400 managed nodes with a S5000PSL (Star Lake) baseboards running Windows Server 2008 with Hyper-V and hardware virtualization support turned on. The management console server consisted of a S5000PAL baseboard on a 2U chassis running Microsoft Windows Server* 2003. The logic block diagram of the rig is shown below. The configuration of the nodes was similar; node 1 has been expanded for detail.

 

 

 

 

In addition to the main CPU, the S5000PSL baseboard carries a baseboard management controller (BMC). The BMC is an embedded computer. Node Manager is firmware that runs on the BMC. Data Center Manager actually communicates with the BMC to carry its functions using the server's Ethernet interface. The BMC supports a TCP/IP stack and carries an IP address separate from the CPU. This is necessary to support bare metal management capabilities. The server platform has an instrumented power supply providing real time readouts of the server power consumption. The information is carried through an out-of-band (OOB) network in the baseboard.

 

 

Finally, here is the console display of Data Center Manager:

 

 

 

 

Data Center Manager supports the notion of logical groups. In the picture above the the four servers were placed in two groups, namely Group 1 with one servers and Group 2 with three servers. The graph shows the power consumption over time of server named "Win11". On the left side, the graph starts with the server idling. The workload used in this demo is SPECpower. There are four instances of Microsoft Windows* Server in each physical machine, also running Microsoft Windows* Server 2008. The graph shows an idling power of about 160 watts. SPECpower was scripted to go through a calibration period of a few minutes and then settling at about 50% CPU utilization. Power consumption is proportional to the workload. Hence we see power peaking at 247 watts for a few minutes and settling at 219 watts thereafter. For a more detailed walkthrough of the demo, please take a look at Dialing in your Datacenter - using Intel Dynamic Power Datacenter Manager.

 

 

There were quite a few challenges in integrating the various technology components. Windows Server 2008 however, was a standout; its behavior was rock solid throughout in spite of being a recently introduced product. I never experienced hangs with the Hyper-V manager and the system was always good at saying what it was doing. These positive behaviors contributed to the general sense of robustness. A system configured with four virtual machines requires about 120 GB of hard drive space and 8 GB of main memory.

 

 

I found Windows Server 2008 very easy to install. Support for the newer platform features was right out of the box. On the other hand I had to tweak the BIOS SATA controller settings into legacy mode before the installation of Windows Server for Windows Server 2003 could proceed. The administrative functions for Hyper-V such as replicating virtual machines were easy to carry out, with Hyper-V Manager taking care of fixing the MAC addresses and SIDs in the clones.

 

 

 

 

3 Comments Permalink
0

Welcome to the new Intel Premier IT Professional zone! I hope that everyone who has visited the old IPIP site for content on technology and IT best practices will take advantage of the new Open Port site to engage us in a more interactive discussion on how we can make you more successful.

 

For my first few blogs I’m going to stick close to the script I normally use when speaking at IPIP events – HOWEVER, you’ll have to attend an event in person to take advantage of the live demos and the lively Q&A sessions, not to mention the animated frog video and the free Starbucks cards (admit it, you really want to get an IPIP event now, don’t you )

 

The theme of this blog is “Technology is providing tremendous benefits to IT”. I’m going to describe a number of key technology innovations and what they mean to you as an IT manager.

 

1) Multi-core CPUs

 

I was surprised (and a little dismayed) at a recent data center engineering workshop where I discovered that there were still a lot of engineers who perceived that Intel CPUs were still getting hotter and hotter over time. This hasn’t been the case for several years, and it’s because of multi-core technology. Since CPU power goes up with the square of the clock frequency it’s clear that we can’t just continue to run clocks faster and faster. Here’s a simple set of numbers to understand how multi-core works: If you increase the clock speed on a core by 20% you increase its power consumption by 73%. If you reduce the clock by 13% you reduce by power by 49% (but you still get 87% of the performance for most software). So let’s put two of these slower cores together – they consume the same power as the original core but provide 73% greater performance at that same power level. So you can see that multi-core is a key component of energy efficient CPUs.

 

2) 45nm High-K metal gate technology

 

The biggest recent advance in semiconductor technology is 45nm High-K metal gate technology. There is a ton of information that you can find on the intel.com website, but let’s hit the most important points. At these tiny topologies the insulation layer on a transistor is down to a width of 4 or 5 ATOMS, so there is a much greater opportunity for energy to escape through that insulator. The new materials we are using provide a much more effective insulator, and the result is that our new CPUs can run at higher frequencies (and therefore higher performance) with no increase in power consumption.

 

3) Microarchitecture

 

The microarchitecture is the foundation of how the CPU performs computing operations. Intel Core™ microarchitecture is optimized for multi-core CPUs and provides higher performance in the following ways:

  • More instructions per clock cycle (increase from 3 to 4)

  • Smart cache to reduce latency to frequently used data

  • Smart memory access with built-in intelligence to speculatively load data for instructions that are about to execute

  • New instructions that double the speed of execution of a wide range of video, imaging, financial, engineering and scientific applications

The microarchitecture also includes new instructions to increase energy efficiency. For data centers this means lower power and thermal burdens. For mobile users this means greater battery life and smaller form factors.

 

4) The “tick tock” model

 

Developing a new microarchitecture is a lot of hard work with a lot of extremely difficult technical challenges to overcome. So is a process shrink that reduces the distance between transistors and increases the density of the CPU package. To manage these innovations in a predictable and consistent cadence Intel follows the “tick tock” model, introducing a process shrink one year (the “tick”) and a new microarchitecture the next (the “tock”). In 2007 the Penryn CPU was introduced with a process shrink from 65nm to 45nm, as well as the high-k metal gate technology. In 2008 we will introduce the “Nehalem” microarchitecture with a host of new architectural improvements for performance and energy efficiency (see tomorrow’s blog for more details).

 

THE PUNCHLINE:

 

Put all those technology advances together and we have made tremendous strides in performance and energy efficiency. As we progressed from single core CPUs to the latest quad core CPUs performance has increased by 5.85 times with no increase in CPU power consumption.

 

Think of what that means for your data center where you may be worried about power and space. We did a paper calculation of a data center running 5.1 million business operations per second, which would have required 126 servers in 6 racks, requiring 240 square feet of space and 48kW of power in 2004 using the single core CPU of the day. Today with 45nm quad core CPUs the same work can be accomplished with 17 servers filling a fraction of a rack, and requiring only 40 square feet of space and 6kW of power. That’s an 83% reduction in floor space and 87% reduction in energy costs. At current power rates that’s a savings of $53K.

 

I hope I’ve been able to show how the cool technology we’re developing at Intel can translate into bottom line benefits for you as IT managers.

0 Comments Permalink