30 Minute Interview

We’ve brought the power envelope down and power consumption has gone back to the Pentium days

Narendra Bhandari, Director Asia Pacific, SSG Global Developer Relations Division, Intel talks to Prashant L Rao about Intel’s resurgence over the past year, the implications of multi-core architectures, the niches for Itanium and Xeon as well as the company’s roadmap.

Narendra Bhandari

Intel’s made a remarkable comeback in the past year. Care to tell us about it?

We introduced the Core microarchitecture in mid 2006 starting with the desktop and then moving on to the server and finally mobile platforms. Intel is reaping the benefits of that change and process technology across product lines. The path taken is to keep a common microarchitecture across product lines. The ramifications for end users, CIOs, business planners is that performance goes up by 40 percent plus with multi-core Xeons while at the same time power consumption drops by 40 percent. We’ve brought the power envelope down and power consumption has gone back to the Pentium days. The upshot is that system guys can design thinner and lighter blades.

The perception that there is an issue was there for the last two years. That changed with the Xeon 5300 and 5100 that topped benchmarks across the board. From the customer perspective, we have an interesting power story. Density is up, cost is down.

Some form or the other of consolidation is going on. With virtualisation consolidation has taken on a different meaning. You have hybrid environments where a box runs a variety of OSs and applications (financial, warehousing, etc).

Dual-core’s on the desktop, quad on the server. What’s the impact of this?

From the desktop perspective having more cores lets you do two things. vPro enhances manageability at the platform level. You can manage 500 to 1,000 desktops without software agents. It involves firmware level communication with machines going beyond traditional methods. The other benefit is with regard to power consumption on the desktop which is going down. Even with a dual-core you can start looking at virtualisation letting you run a service OS and a production OS. The production OS is the part that runs your e-mail, business applications et al. You don’t get to see the service OS. It is used by IT infrastructure managers to control, patch and secure your PC. In the past this used to run as part of your environment and things would pop up in the middle of a presentation (patches etc). With a service OS the IT manager can do what he wants without you delaying him. When you are doing something important you don’t suffer because a patch is being updated.

Manageability with vPro will debut in notebooks during the next few months with Santa Rosa.

Even as the mix of notebooks becomes richer, IT managers do not need two sets of policies for desktop PCs and notebooks.

Windows Vista’s just out. Is it a good thing from Intel’s perspective?

With regard to Windows Vista, security and the interface are important and interesting from our perspective. We have been collaborating with Microsoft for years and take advantage of their architecture at the OS level. The OS is now fully aware of the fact that there are multiple compute resources available.

Both Intel and Microsoft seem upbeat about the use of flash-based storage in PCs and notebooks. Why is this?

The introduction of flash-based storage and use of SSDs for internal storage allows the OS to save to flash as intermediate storage so that you don’t have to wait for the HDD. The Classmate PC prototype had no hard drive at all.

Itanium and Xeon are two very contrasting stories. How does Intel see these two processors today?

Servers: Over the past two years we have realised that Itanium is playing in a very niche segment. The software ecosystem and customer adoption cycles are extremely long at the high end of the enterprise. Two years is a trial period for them. Take an advanced platform, for instance one that runs back-end databases and there are six levels of checks in memory and four levels of failure in the bus.

The volume leader is the Xeon 5100. We sold a million units in two months from the launch in mid-July 2006. Here too the spec cycles are long but not as long as in the Itanium space. We are bringing machines to our customers by giving them access to log into machines in our lab. They can run apps and test drive online. People bring in code, set up the test environment. Sometimes we have to evict customers as there are more people waiting. End-users and CIOs are interested in performance/watt and value out of the box. Beyond a point, technical jargon of two vs. four cores or single die is irrelevant. With quad core we were able to show pretty good volume. Server space apps have been threaded for many years now. It is easier in the server space to talk to people about threading. This lines up reasonably well with our Core strategy. If somebody comes and says that our quad core chip is 30 percent slower then it is because the user has fine-tuned his application to the metal for an older architecture and that’s working against them.

Intel’s roadmap seems quite aggressive. Comments?

65nm production is in full volume. Many companies have not yet produced a single 65 nm part. We have shipped at least a few million of those. I haven’t seen any from the competition. 45 nm parts are being prototyped and showcased. We’ll do testing and validation in the later part of the year. Our roadmap has us transitioning to 45 nm in 2008 with Nehalem and 32 nm by 2010 with Gesher. By 2010 we intend to deliver a 300 percent improvement in performance/watt. The transition window from one process to another is at its shortest ever.

More cores are the stated strategy. Dual and quad and beyond to many cores. How this scenario will unfold depends upon how customers choose to use it. A prototype 80 core unit has been demonstrated at IDF. It could end up being a mix and match of large and small cores to make the same sized package. The question remains if performance is better with a mix of large and small cores or with large or small ones. For instance, you could have one large unit for security and lots of small units for transaction engines. You could also have cores focused on certain tasks. With artificial barriers within cores you can take virtualisation to a new level. With a hypervisor you are still sharing certain aspects. You need a VMM to moderate shared resources. We expect to see multiple hypervisors and cores going forward.

At the process level as you go down the line from 65 to 45 to 32, switching state in transistors creates problems. Leakage results in power consumption going up. Reliability issues crop up. Dielectrics and substrates have to be carefully measured. By addressing these issues you are able to reduce power consumption further.

How do we get to think in parallel? Parallelism has been around for 25 years. At an architectural level start thinking of logical partitions. The logical unit of computing has to be sizeable enough to take advantage of cores.

Is 2007 going to be the year of WiMAX?

In May 2006 we talked at the corporate level about how we could get technology to billions through the World Ahead program. Connectivity is one of the pillars of this effort and WiMAX has to be the transport mechanism. Accessibility is another issue because you can’t expect a device in every home. Shared devices are the way ahead with pay as you go models. With regard to content the simplest usage model is at the education level. With regard to WiMAX: spectrum policy got framed last year. Products using this technology are being tested by more than 60 carriers worldwide.