facebook rss twitter

Review: Intel Xeon 3.4GHz ['Nocona' core]

by Ryszard Sommefeldt on 18 August 2004, 00:00

Tags: Intel (NASDAQ:INTC)

Quick Link: HEXUS.net/qazw

Add to My Vault: x

Introduction

Intel recently launched their first processors supporting AMD's AMD64 instruction set architecture, their extension to the venerable x86 ISA. Bringing x86 into the world of native 64-bit processing, AMD64 first debuted with a physical CPU way back on the 22nd of April 2003 in New York, as AMD launched their Opteron series of server CPUs.

That ISA then extended onto a full desktop range with Athlon 64, a current CPU line that straddles two sockets and over a dozen discrete model variants. In the near eighteen months since the first Opteron CPUs launched, the market has waited with somewhat baited breath to see whether Intel would produce a compatible processor that implemented the ISA. When the concrete news came through that they were indeed going to, with their Nocona Xeon variant, it was simply a matter of playing the waiting game.

And with their recent introduction of the Pentium 4 'F', a desktop CPU with the Prescott core that has AMD64 support enabled, all we're really left waiting for is Microsoft and their AMD64 builds of popular operating systems, especially Windows XP.

Nocona

Nocona is more than just an AMD64-supporting Xeon. It also brings a couple more features to the Xeon range, that had been lacking before. The introduction of 1MB L2 cache as the lowest L2 cache size for the new range, and the front side bus speed bump to 800MHz, up from 533MHz on the older Xeons. What's more, keeping in step with Intel's desktop P4 platform, the Xeon range also gets a supporting set of DDR-II-capable core logic, that further implements PCI Express as a peripheral interconnect, including the PEG16X part of the standard for graphics.

SSE3 is now a feature on Xeon processors, the result of using the Prescott core in the Xeon range as the basis for Nocona. Finally, there's the obligatory speed bump. With previous Xeon CPUs topping out at 3.2GHz, new CPUs have arrived that are capable of 3.4GHz and 3.6GHz initially, with 3.6GHz+ future speed grade targets for Intel.

To sum up, the old SSE2-capable Xeon range on 533MHz front side bus with 512KB of L2 on the basic model, that stopped at 3.2GHz, now gets SSE3, 800MHz front side bus, up to 3.6GHz currently and 1MB of L2 cache memory.

Tumwater

As mentioned, the new Nocona Xeon gets a fair range of new supporting core logic, depending on whether you wish to implement the new Xeon in a server or workstation environment. The server core logic leaves off such items as AGP or PEG16X graphics interconnects, and focusses on things like ECC memory instead.

The main workstation-class chipset is Tumwater, or E7525 as it's formally known. Tumwater has support for DDR-II memory, 24 lanes of PCI Express and 800MHz Nocona Xeon CPUs. It supports the PXH I/O ASIC, that provides PCI-X segments bridged over four of the PCI Express lanes, leaving the MCH free to provide twenty lanes for graphics, split between a full PEG16X physical slot and a 4X slot with a PEG16X physical connector, should the mainboard provide it. It also supports ECC memory.

Intel, in their infinite wisdom, decided to leave their new ICH6-series of I/O southbridge ASICs out of the new core logic packages for Nocona, instead pairing the ICH5-series with everything, at least initially.

EM64T

EM64T, Extended Memory 64 Technology, is Intel's implementation of the AMD64 ISA in the Nocona Xeon and enabled Pentium 4 desktop processors, both of which are based on Prescott, Intel's newest 90nm process technology.

It's almost exactly comparable to AMD's implementation in their own processors, bar one or two differences, one major one happening to be the amount of physical memory the processor is able to address. AMD gave Opteron and all their other first-generation AMD64 implementations the ability to address 40 bits of physical memory (128GB on any connected memory controller, wherever that resides), putting that information and the CPUID data needed to report it to the operating system in their documentation for the new ISA.

With Intel seemingly engineering their EM64T implementation from published AMD documentation on AMD64, they've programmed the new EM64T CPUs to return a supporting PA size of 40 bits to the operating system, while the CPUs themselves are only capable of addressing the same 36 bits of physical memory that all Intel CPUs have for the past few years, via the Physical Address Extention mechanism. Whoops.

Apart from that however, there's little to distinguish between first-gen AMD64 implementations in Opteron and Athlon 64, and Intel's EM64T attempt, in terms of capabilities. The proof of that particular pudding is in the performance in 64-bit long mode. More on that later.

Putting it all together

So Intel's new Xeon platform, with the new CPU, new core logic and new ISA support in EM64T, is the biggest at-once shakeup of their mainstream server and workstation platform since it was born. Intel have dragged it kicking and screaming to the same basic specification levels as the latest Pentium 4s.

The remit of this article? To pit Nocona and Tumwater up against AMD's current finest in a workstation-oriented series of tests using a 32-bit operating system, to see if Intel have enhanced the Xeon platform's performance enough to stop Opteron's relentless march into the workstation space.

Workstation-class is an ugly thing to define. It means something different to anyone you ask, but in the most vague of general terms as far as I choose to define it, it means a basic system that usually implements two physical processors, plenty (over 1GB) of memory and is aimed at one of the professional levels of computing, such as digital content creation, CAD or CAM, 3D rendering or programming.

Boxes with a bit of oomph, over and above what's usually put together by the well-heeled enthusiast.

Why no 64-bit focus?

At the time of writing, core logic, networking and graphics drivers for the EM64T-supporting build of Windows Server 2003, on the Nocona-supporting mainboard used for testing, were none existant. While the box was happy booting the build, there was precious little to do with it, bar giggle at System Properties showing an Intel-based machine booting an operating system designed for specific AMD processors. If you can't network it, accelerate the graphics or provide basic support for the board features, it's like owning the CPUs without a mainboard to stick them in.

So while I managed to squeeze in a tiny bit of 64-bit analysis at the end, it's certainly not my focus until Microsoft and Intel collaborate to ship the software needed to test Nocona in full, using a Windows operating system. Linux was not an option at the time of writing, due to the amount of time required to generate non-server benchmark tests being far larger than the time spent with the hardware before it was plucked from my hands and returned to the vendor.

Enough waffling. You should know enough about Xeon for me to not bother going into much detail about its architecture, especially since Nocona is based on a desktop CPU core, so I'll just cover the CPU basics in upcoming pages. The rest I'll cover in detail. Onwards!