Home / Opinions / EM64T Server Roundup



EM64T Server Roundup

When AMD announced its Opteron processor and 8000-series chipsets over year ago it seemed to be taking a huge gamble by offering the server market simultaneous 32 and 64bit processing. However, it looks to have paid off as in the understatement of the year, Intel announced its EM64T (extended memory 64 technology) confirming that AMD’s hunch was right all along. Battle looks set for the enterprise server market as AMD has had a full year head start over Intel and hasn’t been idle either. Most blue-chip vendors including IBM and HP now offer dual- and quad-Opteron servers finally making AMD a serious alterative to Intel.

In this round-up we take a look at three examples of Intel’s new processors and chipsets from Supermicro, HP and Dell but first we delve deeper into EM64T as it adds a number of significant features that AMD will be hard-pushed to match. Codenamed ‘Nocona’, the new Xeon processors are available in speeds ranging from 2.8GHz up to 3.6GHz and all support the new 800MHz front side bus (FSB). The pin package hasn’t changed but internally these chips use the 90nm Prescott core with additional 64bit instruction sets. The processors no longer use L3 cache but have had L2 cache increased to 1MB. There’s no indication whether L3 cache will be used in later versions but we understand that the L2 cache will be increased to 2MB in 2005. Cooling will need to be good as the larger L2 cache makes these chips run a lot hotter.

Three new chipsets are available with the workstation market looked after by the E7525 ‘Tumwater’. The server market gets the other two with the E7520 ‘Lindenhurst’ handling dual processing performance and volume server applications, while the E7320 ‘Lindenhurst-VS’ with its reduced features is aimed at low-cost dual-processing server platforms.

All three chipsets support both DDR and DDR2 memory but both types cannot be mixed on the same motherboard as the key slots are in different positions. At present prices are substantially higher for DDR2 with it costing around 45 per cent more. The current PC3200 DDR2 modules don’t offer any significant performance benefits but whereas DDR technology is end of life, using DDR2 modules will get you onto the future proofing ladder. As faster modules appear, demand will inevitably bring prices down with parity expected by mid-2005. DDR2 uses an FBGA (fine ball grid array) packaging which allows chip densities to be increased as the memory chips are much smaller. Cooling demands are also lower as DDR2 only requires 1.8v of power as opposed to DDR’s 2.5v.

PCI Express is Intel’s other weapon and each chipset has a different implementation. The E7525 provides an x16 interface aimed purely at graphics applications whereas the E7520 supports three x8 configurable PCI Express interfaces, two PCI-X segments, dual Gigabit Ethernet interfaces and the Dobson I/O processor. Networking is a key focus of Lindenhurst as the bandwidth available with PCI Express can support 10-Gigabit Ethernet and Infiniband. For the E7320 support is reduced to a single configurable x8 PCI Express interface split into two x4 PCI Express interfaces supporting dual Gigabit Ethernet, a pair of PCI-X segments and a single x4 PCI Express slot.

One point to remember with Intel is its processing roadmap will almost certainly converge on one solution – Itanium. As with AMD’s solution the new Xeon processors and chipsets are merely a first step in the inevitable move to 64bit computing. When it will finally happen is anyone’s guess, but happen it will and for the moment Intel looks a serious alternative to AMD for companies starting out on the migration path.

comments powered by Disqus