![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
|
|
![]() |
|
|||||||||||||
By Chris Angelini |
||||||||||||||
SMALL businesses are tired of server boxes with fans blowing furiously to keep temperatures down. Large businesses will give anything to reduce the power of their server farms. And yet the most popular CPUs are crammed into tiny 1U form factors where they consume more than 100W of power each all day and night long. As AMD and Intel have jockeyed for position in the server market, they've successively cranked the performance dial. That's the measure most IT professionals use to compare one platform to another, after all. Supercomputers and desktop workstations alike are evaluated on the basis of horsepower rather than efficiency. But because upping clock speeds and adding cache seem to be the most popular techniques for procuring better benchmark scores, power consumption numbers keep on creeping upward. The common neutralizer—process technology shrinks—simply can't keep up with demand for speed. The situation has come to a bit of a head lately, though. Intel realized its NetBurst micro-architecture probably wouldn't hit the 4 and 5 GHz marks forecasted previously. And AMD is still struggling to get 65 nm production-ready, limiting what it can do with today's 90 nm parts. In addition to the issues encountered by AMD and Intel, rising energy prices have given your SMB customers something to think about as they add and replace servers. Consequently, it's no longer fashionable to compare raw speed. The new phrase du jour is performance per watt. For two companies accustomed to competing on the basis of benchmark scores and price, game plans are most definitely being revised. AMD is a little nebulous on its strategy. But based on the company's three-year outlook, you can safely guess that quad-core processing will emerge in 2007 along with a shared L3 cache and extensions to the AMD64 instruction set. Advanced power and thermal management features are both slated for 2006. And while it isn't quite clear what those two terms mean technologically, AMD has already started talking about optimizing the power consumption of its multi-core designs. Intel is taking a different approach, abandoning NetBurst altogether in favor of a new architecture, known today as Core. The Core architecture features a shorter execution pipeline less dependant on super-high clock speeds for reasonable performance. Gone are the aspirations of 10 GHz by the end of the decade. Now Intel's focus is on maximizing the performance per watt of power consumed. Server Technology Goes Mobile
That the efficiency-oriented Core design actually incorporates strengths from the original mobile Pentium M design should come as no surprise since it has long been lauded for delivering ample speed on the road. The Pentium M did trail Intel's Pentium 4 and Xeon processors with regard to features, though, missing 64-bit extensions, Hyper-Threading, SSE3, and a fully modern complementary chipset. Naturally, Intel has had to put some work into updating the architecture with a few of its more desktop-oriented value-adds. The most recognizable incarnation of Core persists on the mobile side as Core Duo and Core Solo, flagships of Intel's Centrino initiative. Notably, 64-bit and Hyper-Threading support are both still absent, but plenty of media-, power-, and execution-oriented optimizations effectively close the gap between the old Pentium M and today's Core-series. Intel quickly realized that the same hardware used in its mobile platform would also work well elsewhere. A handful of motherboard vendors had already offered desktop products for the older Pentium M, touting low power, minimal heat, and admirable performance. Those same three attributes seemed like a natural fit for the server market, too.
Sossaman is Born Although Intel's recent, successful projects have all centered on platform initiatives, Sossaman is simply a code name for the Xeon LV processor. Think of the chip as a Core Duo with special concessions for enterprise operation—and a compromise or two. Perhaps most interesting is the fact that Xeon LV is only available in two variations, one running at 2 GHz and the other at 1.67 GHz. Both models are dual-core-only, unlike the mobile Core chips, which can be had in single-core trim. Further, the Xeon LV chips support dual-processor configurations, paving the way for some incredibly powerful server setups. There is a lot of value built into the Xeon LV. DBS (Demand-Based Switching), a feature previously constrained to the full-power Xeon line, makes its way into the Xeon LV, reducing power demands even further on an already heavily optimized CPU. Execute Disable Bit support, used to curb the proliferation of certain types of malicious software code, also comes as a standard feature. Multimedia enhancements to the chip's architecture include SSE3, which undoubtedly makes a huge difference on media servers tasked with encoding and decoding video. Lackluster media performance was one of the Pentium M's initial shortcomings. Unfortunately, the Xeon LV doesn't yet leverage 64-bit processing, a weakness it shares with the mobile Core Duo processor. Of course, the sting is much more poignant at the server level since there are plenty of 64-bit applications currently taking advantage of the technology. Microsoft has even announced that its upcoming version of Exchange will require a 64-bit processor, outdating the Xeon LV in any mail server role. Nevertheless, Intel does give the chip 36-bit PAE (Physical Address Extension) support, which at least enables platforms with up to 16GB of memory as opposed to 4GB. Xeon LV's fundamental speeds and feeds remain closely tied to Core Duo. The processor operates on a 667 MHz front-side bus. While some customers might point out that their other Xeon boxes proffer 800 MHz and higher bus speeds, it's important to note that the Xeon LV centers on an entirely different architecture less dependant on amped-up clock speeds, even with a pair of dual-core chips crunching four threads simultaneously. Performance is no doubt aided by the Xeon LV's 2MB Smart Cache. Shared between the two cores and configured to dynamically allocate space, Intel claims that the cache further minimizes front-side bus traffic. Superb performance and a copious list of features are exciting, sure. But more than anything expect your customers to want the Xeon LV for its power consumption numbers. Rated at just 31W, you can drop two chips onto a motherboard with all of the other server trimmings and still come in under the draw of just one of Intel's older Xeon chips on the NetBurst micro-architecture. Since many small businesses keep their servers in offices where employees work, those customers will appreciate Xeon LV's ultra-low noise output. Larger companies deploying several servers in dedicated space are more likely to look favorably on the drastic power savings enabled by the chip. A Platform in Place Many of Intel's processor unveilings also involve brand new platforms. With Xeon LV, resellers are able to lean on a familiar chipset, though they will need to purchase corresponding motherboards with the Socket 479 interface. VARs who've actively worked with Xeon processors in the past should already be familiar with the E7520 chipset, formerly known as Lindenhurst. Extremely flexible in bus speed support, memory compatibility, and peripheral connectivity the E7520 easily accommodates the bandwidth requirements of Xeon LV while still endowing the platform with plenty of modern functionality. Given the strong value proposition affixed to Xeon LV, most board vendors pair the E7520 memory controller hub up to an Intel 6300 ESB I/O controller, extending budget-friendly technologies such as SATA to the SMB space. Between the two, you have access to 66 MHz PCI-X slots, vanilla PCI, and as many as three x8 PCI Express connections. That should be plenty of expandability for any pedestal or rackmount server. Though the chipset does support regular DDR memory modules, Intel is pushing DDR2 quite heavily, so expect Xeon LV motherboards to leverage DDR2-400 almost exclusively. Fortunately, DDR2 features a smaller power footprint than DDR, allowing Xeon LV servers to run even more efficiently.
Tyan's Tiger i7520SD exemplifies the platform's utility, combining a pair of Socket 479 interfaces, eight DDR2 DIMM slots, PCI Express, PCI-X, and standard PCI connectivity. Populate the board with two Xeon LV chips and you're looking at a quad-core server draining half the power of last generation's Xeon powerhouses. Tyan builds even more value onto the board through its proprietary TARO SO-DIMM slot, which can be used to add hardware-accelerated SATA RAID or Ultra 320 SCSI. What About Bensely? The Bensley launch saw Intel selling its newest Xeon processor family, the 5000-series, along the 5000P/V server chipsets. At the same time, Intel announced the Xeon 5100-series, admitting it wouldn't be available for a month or more.
Although the model 5000 and 5100 Xeons would seem to be closely related, they center on completely different processor designs. The 5000-series chips represent the last remnants of NetBurst running at speeds of up to 3.73 GHz and sporting 4MB total L2 cache. That brute force approach is on its way out, though, as Intel deemphasizes its aging micro-architecture. The timing couldn't be better, either. Each CPU in a server built using 5000-series chips is rated at up to 130W. No wonder customers are concerning themselves with power consumption numbers nowadays.
The 5100-series will be much more attractive once it's widely available. The design is a Core derivative armed with Virtualization Technology, 64-bit extensions, a unified 4MB cache, and a front-side bus operating at up to 1,333 MHz. Perhaps more importantly, it'll consume much less power than the 5000 family before it. According to Intel, the very fastest versions will be rated for 80W while the lower bins should dip in at around 65W. Next to an older Xeon, the Core-based models should be much more power friendly. But at 80W, and even 65W, you're still talking about significantly more power, heat, cooling, and noise than the Xeon LV and its 31W rating.
In short, Bensley is the future of Intel's multi-core Xeon lineup. The Dempsey (NetBurst) design drops into the platform today, Woodcrest (the first Core-based model) will extend its life moving forward, and Clovertown (Intel's introductory quad-core processor) is expected to see the platform to its end. With rising complexity, though, expect higher Xeon power draws going forward. Sossaman, For the Win Xeon LV clearly addresses a different market than any of Intel's other server processors. Positioned as the ultimate in performance per watt, the chip gives your customers all of the features they'd get from today's Core Duo plus DBS and 36-bit PAE. The flexibility to operate in dual-processor configurations, effectively yielding four cores per server, is great for scalability in a small business environment, too. The lack of 64-bit extensions will discourage some from Xeon LV, especially those planning to run Microsoft's Exchange 12. But think about the number of SMBs currently running 32-bit software packages, such as Small Business Server 2003, in environments resistant to change. In short, some folks stand to benefit from a shift to 64-bit down the road, while others are more likely to enjoy a 32-bit processor capable of saving money on electricity today. Granted, it doesn't hurt that Xeon LV also puts out less heat and runs more quietly than any other server chip at your disposal. In the real world, sometimes the least publicized solutions deliver the best value. |
||||||||||||||
Copyright © 2007 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form. |
||||||||||||||