Intel 'Nehalem' Xeons poised for March 31 launch
IT managers, start your checkbooks
Chip maker Intel has been dropping hints like water balloons from the rooftops that its "Nehalem" Xeon EP processors for two-socket servers will arriv before the end of the quarter. And it looks like the chips will finally fall where they may on March 31.
As it turns out, Intel and x64 server maker Sun Microsystems - Intel's new best buddy and a vendor that is dying to rev up its "Galaxy" x64 server business to compensate for declining UltraSparc-IV+ and Sparc64 server sales - are going to be hosting a big shindig showing off Sun's riff on Nehalem iron at the Hospital Club in London on the evening of March 30. Presuming that this event is being held ahead of the actual launch by Intel, which the invite implied, that leaves only March 31 as the launch day before Intel will have slipped into the second quarter of 2009.
In a normal economic situation, companies would be enthusiastically looking forward to new iron from Intel, and they might even be a bit grumpy that the Nehalem EP chips were not here in the fourth quarter of last year or at least by January of this year. But since about August of last year, the server racket has been anything but normal. It has seen a sharp contraction in revenues and shipments that we have not witnessed since the Y2K, ERP, and dot-com booms all went pop in 2001. The question now is whether the new lineup of two-socket servers can shake IT managers out of their gloomy moods enough to spend money on these machines.
With the quad-core Nehalem EP processors and their QuickPath Interconnect offering between three and four times the memory bandwidth of the current quad-core Xeon 5400 series processors and their antiquated front side bus architecture, it will be easy to make the technical case for an upgrade to the new chips, code-named "Gainestown" and paired with the "Tylersburg" chipset. The Nehalem EPs will be sold as the Xeon 5500 series. They were outed last week by Apple, which plunked the Nehalem 3500 (for single-socket machines) and 5500 (for dual-socket boxes) into its Mac lineup. It is not clear how much more raw oomph the Nehalems will have, but on some early benchmarks, system performance has increased by nearly 80 per cent.
There will probably be some grumbling about the DDR3 main memory used with the Nehalem EPs being hotter than DDR2 memory, but if the chips are priced correctly and the thermals work out to be about the same at a system level, then some customers will be ready to buy no matter how bad the economy looks. (More than two million servers did ship in the fourth quarter, after all, and that was not a great time to spend cash on anything). Toss in better support for virtualization and expanded main memory to support more virtual machines per physical server, and you might just have enough of a sales pitch as IT managers try to consolidate machines and reduce administration headaches and increase resiliency in their data centers by going virtual.
The two-socket x64 server is the workhorse of the data center, and with lots of memory and some tweaks to improve the way server virtualization hypervisors and their guest operating systems behave, you can bet that the same companies that might have been looking at larger four-socket machines, their larger main memories, and their much larger price tags are now going to consider a Nehalem EP box sporting larger memory capacities and a relatively smaller price tag a better fit for the compressed 2009 budget.
This kind of platform form-factor downshifting is nothing new. As x86 and x64 workloads keep growing, the underlying iron has kept pace with increasing clock speeds and then core counts on the CPU front and ever-expanding disk and memory capacity on the storage front so the two-socket box usually ends up being the best box for the buck except when we get stuck between processor transitions.
What this may mean is that companies that have tended to sell bigger x64 boxes - I am thinking here mostly about IBM, but also a little bit about Unisys - may have taken such big financial hits in recent quarters with their x64 lineups because their customers were waiting for Nehalem instead of buying more expensive and available machines using the "Dunnington" quad-core and six-core Xeon 7400 processors.
Other factors could have been in play at IBM in the fourth quarter, of course, but this is one reasonable explanation, particularly since Hewlett-Packard and Dell, which do not sell x64 machines with more than four sockets, were not hit as hard as IBM was in Q4. Sun, as it turns out, was one of the few vendors to grow x64 server shipments, but it's growing from a very small base and has a long way to go to become one of the volume players. At this point, even catching IBM seems difficult for Sun to do in this space. But a decent lineup of Nehalem EP rack and blade servers could help. Sun has shown no inclination to chase the small and medium business market with tower servers and is leaving all of those volumes to HP, Dell, and IBM and their respective channels. ®
So much for Tukwila, Right?
It looks like Nehalem EX will be four times as fast with roughly the same transistor count and available more than six months sooner so one has to wonder. HP should have a boiler room full of people right now porting HP-UX to Nehalem. According to Intel Tukwila will only be twice as fast as Montvale and that's only because it has twice the core count and Quickpath. The Itanic was probably a bad idea though it did have its day. When you need to have 24MB or more of cache and that many execution units to support a dual core CPU at a blazing 1.66 MHz you clearly have a problem. I had a DEC Alpha well over a decade ago with compilers that could do the same things that EPIC claims to do if you compiled and bound the whole program at once. Plus a chip that had aggressive branch prediction and instruction reordering. So one has to wonder what HP was thinking when the PA-RISC 8200 was arguably the first truly four way superscalar microprocessor. Why throw away half of your performance optimization for Taliban style ideology? If HP doesn't port HP-UX to Nehalem which will require some work due to the endian issues I don't see how it can survive because of the delays in delivery of Itanium upgrades. Montvale was originally touted to be a 3+ GHz chip and instead it ramped up an astounding 0.06 MHz. It's pretty obvious where Intel is putting their R+D money and it obviously isn't Itanic. At this point you could probably build a Nehalem emulator that would outperform Tukwila at one third the cost per core.
There are pre-release Dell quad socket Nehalem boxes in existence, some premium customers have them already. I am told April release IIRC.
when can I buy one
Anyone comment on how long it will be before dell etc get these into shipping servers?