New Xeons: has Intel done enough?

Intel threw a slew of server chips at the market this week, as well as unveiling a host of OEMs whose products will turn the raw silicon into servers. But is it enough to see off AMD's Opteron challenge?


Intel threw a slew of chips at the market this week, as well as unveiling a host of OEMs whose products will turn the raw silicon into servers. But can they see off AMD's Opteron challenge?

Acolytes attending the launch of the dual-core 3.4GHz 64-bit Xeon 7100 series of processors -- codenamed Tulsa -- included HP, Dell and Unisys.

Tulsa is Intel's response to AMD's Opteron 800-series processors, challenging its dominance in the server processor market. The Texan upstart's share of the x86 processor market has reached almost 25 per cent from near-zero just a handful of quarters before. And Intel has admitted to haste in getting Tulsa out of the door. "It certainly added to our sense of urgency," said Tom Kilroy, vice president of Intel's digital enterprise group.

Hardly surprising when you consider that Dell opened its doors to AMD for the first time ever in order to meet customer demand for high-end x86-based servers. IBM too, though it's long been an AMD customer, intensified Intel's concern by launching four-way machines which, as Dell's seem likely to be, are based on the Opteron 800-series. It's worth noting though that, despite the hoopla over Dell's AMD move, real products won't appear until sometime towards the end of the year.

The importance of this choice to many enterprises and data centre managers can hardly be over-estimated. Data centres in particular are struggling to keep power consumption down. Their problem is not just rocketing power bills, but also because there are doubts both over the National Grid's ability cope with the power that data centres are sucking from it, and the internal infrastructure of many data centres to handle such power levels. Despite that, enterprises need more computing power, not just to feed growth in a growing economy, but also to underpin consolidation projects based on virtualisation, for which four-way servers are ideal for medium-large and large organisations.

So what of Intel's new products? Will they be good enough to assuage customer demand and see off the AMD challenge, both from the point of view of power and compute performance? At first blush, it would appear that some minus points still apply.

Looking at the positive first, according to Intel, the new chips deliver up to 2.8 times more performance per watt than the previous-generation Paxville, with performance gains from 38 per cent to 70 per cent, depending on workload. This is largely as a result of the new chips being based on 65nm technology rather than Paxville's 90nm. They boast 16MB of shared cache and Intel's Cache Safe technology, which allows the device to recover from a level 3 cache error. Intel argues that Tulsa's large shared cache means that its lack of AMD-style on-die memory controller is less of an issue, because the chip has to access main memory less frequently. Systems builders can chain up to 32 of them together too.

However, Tulsa is also based on the now-tainted NetBurst architecture, which is known to be power-hungry. Intel reckons it's got the edge in the performance/watt department and this generation does improve on Paxville, so widely recognised as hardly worth building systems around that many of Intel's core OEMs simply ignored it; part of its problem was profligate power consumption.

However, the newer Core architecture improves on it even further, and it's this that Intel has been busily trumpeting in every piece of media space it can lay its hands on.

Alas, Tulsa doesn't fit into that box. While the lowest-consumption Tulsa chip sucks up 95W -- on a par with AMD's Opterons -- the top end swallows far more. And that's before you add the power consumed by the memory controller and other frontside bus components to your electricity bill -- all consume significant amounts of power.

In contrast, AMD famously reckons that including much of this architecture on-chip is what helps it keep the power bill down -- that and the silicon-on-insulator processor manufacturing technology that it licenses from IBM.

That said, Intel's OEMs are enthusiastically adopting the new part. As Kilroy says, they're not cutting Intel out in favour of AMD, they're simply offering users a choice. And Intel reckons it's done enough until it can produce the next generation, codenamed Tigerton, in the second half of 2007.

These processors are slated to include a new high-speed interconnect, CSI, that allows each core to link directly to the chipset rather than share the front-side bus. It sounds not a million miles from AMD's now well-established HyperTransport, as found in the Opteron. Intel would never say so of course.

With a wide range of products and enough of a nod to the performance/watt issue, Intel may well have done enough to keep the dogs of war at bay. But Tigerton had better deliver because, in the meantime, AMD will be busy trying to chomp through Intel's breakfast, lunch and more.

Competition in the server processor market: it's about time.

"Recommended For You"

AMD Opteron processor update targets the cloud Beyond dual-core: 2007 desktop CPU road