Intel is well known as the company whose processor chips have largely driven the hardware side of the PC revolution, but the firm has rarely stood still. Indeed, with increasingly stiff competition coming from AMD and relative newcomers such as VIA Technologies, the chip maker has in recent years redoubled efforts to keep its processors ahead of the game.
Later this year, Intel plans to introduce its latest chip family, known by the codename Nehalem. This flexible design will enable the firm to target market segments from ultra-portable laptops right through to high-performance servers using the same basic architecture.
These chips will be followed in the 2009-2010 timeframe by a die-shrink of Nehalem to a 32nm design, as part of what Intel calls its tick-tock lifecycle.
But the firm sees that the broader IT market is also changing, with a greater emphasis on workforce mobility, while at the same time faster and more pervasive networks are changing the way IT services are delivered. Like author Nicholas Carr, it seems Intel believes that in future, more and more IT functions will be outsourced to online service providers.
Tom Kilroy, vice president and general manager of Intel’s Digital Enterprise Group, said that the company is now seeing stronger sales of its technology in small-to-medium firms, and said that one reason is because of managed services delivered via the internet.
“Our vPro technology was targeted right at large enterprises, but many adopters have also been service providers. They have been using it to deliver managed services to customers,” he said.
VPro, which is now in both desktop and laptop platforms from Intel, provides some management capabilities in the hardware, so that IT support staff can reach a PC, even if the operating system will not boot.
In future, the technology is set to be extended with other capabilities that will prove useful for enterprise customers, but which Intel will also add with an eye to service providers. One of these will be the ability for a user to remotely call for help if their laptop is causing difficulties, Kilroy said.
“You’ll be able to ping the helpdesk, and they can do a remote diagnosis. It’s about saving money through not having to do a truck-roll… vPro is lowering the total cost of ownership through better management,” he said.
Also due to be added to vPro’s feature set is an anti-theft capability, recently announced at Intel’s developer forum in Shanghai. Little has yet been disclosed by the company concerning this technology, but it is expected to give administrators the ability to render a missing or stolen laptop unusable.
“You’ll be able to silently and remotely disable a stolen PC by sending it a ‘kill pill’,” said Kilroy, but he declined to delve any deeper into the technology, which is set to be introduced by the end of this year.
Back on firmer enterprise territory, Intel is set to introduce a six-core Xeon chip for servers in the second half of this year. Codenamed Dunnington, the chip will be aimed at four-socket multi-processor systems.
Dunnington is likely to be the last of the Penryn family of processors to come to market before the first Nehalem chips appear, but Kilroy dismissed suggestions that six cores was the limit of what could be achieved with Intel’s current architecture. Instead, he said it was optimised to hit a sweet-spot in the market for price and performance.
“It’s about delivering more for a given silicon area, and six core is a nice price/performance proposition,” he explained.
Nevertheless, Intel expects Nehalem will scale up to deliver server chips with up to eight cores, while each core will also be capable of executing two code threads at the same time.
“I don’t currently see any plans for Nehalem beyond eight cores,” Kilroy said, adding that a four-socket Nehalem system would be capable of handling up to 64 threads simultaneously.
“It all boils down to what is the optimum level of performance we have to consider the thermals as well when building these multi-core chips,” he explained.
Intel has tried multi-threading before, with a feature called HyperThreading that it added to Pentium 4 chips. This was meant to boost performance by allowing a second code thread to make use of processor resources that would otherwise be sitting idle. However, few applications at the time were written to use more than a single thread, and so the technology had a mixed reception, with some users even reporting a drop in performance for some applications.
“What we did with HyperThreading was raise the awareness that you could use more than one thread,” said Kilroy. “Software developers are now ready to take full advantage - you will see more uptake on Nehalem than on the P4,” he said.
Nehalem will also mark the transition to Intel’s new QuickPath Interconnect (QPI) for linking processor chips to each other and to the rest of the computer system. This provides about 25GB/s bandwidth per link, and along with a dedicated pool of memory controlled directly by each chip, removes the bottleneck of all data having to travel along a shared system bus, as it does in current Intel chips. The technology is similar to the HyperTransport used in AMD’s processors.
Kilroy said that moving to QPI will bring Intel’s chips greater flexibility for handling high-bandwidth applications.
“Today, we have leading performance without an integrated memory controller thanks to the large cache [on Xeon chips], but we knew we would soon reach a point of diminishing returns. Now [with Nehalem] we have large cache sizes as well as an integrated memory controller,” he said.
Such high-bandwidth applications include virtual workloads, a feature that Intel sees as key, not least because of the importance of this technology in datacentres. However, Kilroy said the technology should not be equated solely with server consolidation.
“Virtualisation has been around since the mainframe era, and allows you to have failover and disaster recovery as well as the optimum way to support scalable workloads,” he said.
Kilroy pointed to Intel’s FlexMigration technology as a key part of its virtualisation support. This enables live virtual machines to be moved around a server farm, regardless of the slight differences between different generations of Intel server processors.
“If you bought Woodcrest servers in 2006, and you add to them with servers this year, you will be able to migrate live virtual partitions between them, allowing firms to pool across different systems,” he explained.
QPI will also feature in Tukwila, the next version of Intel’s Itanium family of processors, also due by the end of 2008. This is expected to deliver a significant increase in performance over Montecito, just as that version was significantly faster than earlier chips.
Kilroy said there was still a place for the Itanium family, despite its not living up to Intel’s early high expectations. In particular, it is widely used in mission-critical applications, where a high level of reliability is called for.
“We’ve seen steady growth in revenue across the mainframe replacement category,” he said, pointing out that many of the top global stock exchanges use Itanium-based systems.
