Core i7 Xeon 5500 for landing
Code tweaks ahead
Nehalem Day After Monday's roll-out of Intel's long-awaited Xeon 5500 family, The Reg sat down with a pair of Intel honchos at the company's Santa Clara, California, headquarters and asked, "What took you so long?"
After all, Intel's Core i7 high-end desktop processors, based on the same new Nehalem microarchitecture, went on sale over four months ago.
We also had a few questions about how straightforward it would be for developers to implement what Intel general manager Pat Gelsinger referred to in his introductory presentation as the Xeon 5500's "wicked-cool new technologies."
But first, that four-plus month delay.
According to Doug Davis, VP of Intel's Digital Enterprise Group and GM of the Embedded & Communications Group, the delay was driven neither by terror of a collapsing market nor due to a hang-up during technology development, but instead was the result of an in-depth validation process.
"There are a lot of things about server and embedded products that go through a pretty rigorous validation process," Davis said. If you qualify as one of Intel's validation customers, "you get early access to all the information, you get early access to the product."
But with great access comes great responsibility. "You have to commit to put it on a board and run it through its paces and give us detailed feedback," said Davis. "There's a collaborative relationship that happens."
And, according to Davis, Intel listens and responds. "It's a cycle. We get some feedback, we make the product better. Do we tweak the BIOS? Do we tweak something in the silicon? We're going to refine it to get it to meet our landing-zone requirements before we'll launch the product."
That's all well and good, but didn't the Core i7 deserve the same fine-tuning?
According to Steve Thorne, the Xeon 5500's product line manager, not really. "Core i7 is only really validated for client operating systems whereas the Xeon 5500 is validated against all the server and enterprise operating systems."
Which makes sense, after a fashion. Server environments are more punishing, the variety of operating systems are more varied, and the virtualization environments are far more demanding.
As Davis put it, "There are a lot of things about server and embedded products that go through a pretty rigorous validation process. We're looking for different things: there's virtualization, there's hyperthreading, but it's also in relation to the OSes, the applications, the physical parameters that these products are being put into."
Which explains why Apple got its Xeon-equipped Mac Pro onto the market four weeks before Monday's official announcement. As Thorne put it, "They're specific to workstation/client applications, so they're a little unique." That's Apple all over, for good or for ill: "a little unique."
But Apple fanbois shouldn't feel too unique. According to Gelsinger, "hundreds of thousands" of Xeon 5500s have been released into the wild to testers, developers, and system integrators.
And those worthies must by now have learned a few bits of news that we learned today from Thorne.
For one, no code-tweaking is needed to take advantage of the Xeon 5500's unfortunately-named Turbo Boost capability, which shuts down cores when they're not being used and cranks up the clock speed of those that remain at work.
Thorne told us that Turbo Boost is all done in silicon. "The way it's accessed is through a standard ACPI [advanced configuration and power interface] call in the OS. So when the OS issues a P-zero-state call, the processor will automatically take over and run [the remaining cores] as fast as it can."
But not all the benefits of the Xeon 5500s are so easy to enjoy. Although Thorne told us that "There are many, many software companies that are just taking their binaries and just running them on Nehalem and they're seeing tremendous improvements," he admitted that to squeeze all the juice out of the Xeon 5500, code-tweaking is required.
Namely, Thorne recommended that developers use Intel's multithreading-optimization tools (or, he hastened to add, third-party ones as well) to tweak their code for the new Xeons. Referring to Gelsinger's comments about the Xeon 5500 doubling the 5400's performance, he said "To really get the double, you've got to optimize your code."
And thus he revealed another iteration of the increasingly oppressive two-threads-per-core, four-cores-per-processor conundrum: Now that we have all this processing power, how are we going to feed the beast? And now that hardware has outpaced software, there's no simple answer in sight.
Finally, in all the hoopla about the dual-socket Xeon 5500, the single-socket Xeon 3500, also officially announced today, received short shrift. Even product line manager Thorne referred to it as "essentially a rebranded Core i7" with added ECC memory support and enterprise-level validation.
And with that dismissal of the Xeon 3500 line, we realized that we've come a long way from the 1995 introduction of the Pentium Pro - which Gelsinger rightly hailed as the processor that began the high-volume server revolution - to today, when a 3.2GHz quad-core processor with 8MB of L2 cache and a 6.4GT/sec interconnect such as the Xeon W3570 is but a footnote. ®