The Nocona variant of the Xeon DP processor will have the Extended Memory 64-bit Technology that was embedded into the Prescott Pentium 4 core upon which Nocona is based turned on.
It is widely believed that Intel reversed engineered the x86-64 instruction set from the hybrid 32-bit/64-bit Opteron processors from Advanced Micro Devices, and is launching the Noconas in workstations and soon in servers to try to blunt the attack that AMD has made on Intel’s corporate computing turf.
Intel had promised back in February that it would ship the Nocona chips, which are built using a 90 nanometer chip process, before the end of the second quarter, and it is just squeaking by on that deadline as it gets the first chips and related Tumwater chipsets out the door for workstations on June 28. Back in February, Intel said that the Noconas would run at 3.2GHz, have 1MB of L3 cache, and will sport an 800MHz frontside bus. The Noconas might be running tight up against their launch window, but Intel has been able to crank the clock speeds from 2.8GHz through 3GHz, 3.2GHz, 3.4GHz, and on up to 3.6GHz in the workstation versions of the new Xeon DPs.
In addition to the 64-bit processing support (and the pipe going out to memory is only 64-bits wide, by the way), Nocona sports a lot of new features, including the SpeedStep power management technology that Intel developed for its Pentium M and Centrino mobile processors. SpeedStep varies the voltage and frequency of the processor, stepping it down as software is asking the processor to do less work, thereby eliminating some of the power consumption and heat exhaust that a regular Xeon chip has. This is particularly important in a chip like the Xeon, which has a lot of transistors and burns a lot of juice.
The Tumwater chipset for workstations, known as the E7525, supports 400MHz DDR2 memory as well, which consumes anywhere from 30% to 40% less energy than DDR1 memory running at 266MHz or 333MHz. According to Alan Priestley, strategic marketing manager for Intel’s enterprise solutions in EMEA, the Nocona chip can step down from 3.6GHz at 1.4 volts and dissipating 100 watts down to 2.8 GHz running at 1.2 volts and dissipating 70 watts. Other improvements in system design can reduce power consumption for a typical workstation, which eats anywhere from 300 watts to 400 watts, by around 28%, he says.
The Tumwater chipset and Nocona chips are also the first Intel machine to support PCI Express, which offers 8GB/sec of total bandwidth between the memory controller, graphics processor, and Nocona processor. For graphics, the PCI Express link will be comprised of the X16 16-lane bi-directional point-to-point bus, which will offer two times the bandwidth as AGP8X graphics subsystem in current Xeon workstations. The Tumwater chipset has an aggregate I/O bandwidth of 6.4GB/sec, and supports up to 16GB of main memory (two channels, four DIMMs per channel). Without using the 64-bit memory extensions or the X16 graphics, a top-of-the-line 3.6GHz Nocona workstation benchmarked by Intel has run workstation applications 30% faster than the current 32-bit Xeons running at 3.2GHz. This is roughly the same performance benefit that AMD has been saying that Opterons give over Xeons running 32-bit X86 applications.
Pricing for the Nocona chip and chipset were not available before today’s launch, but the word on the street is that the Tumwater chipset will cost around $100 in 1,000-unit quantities, and that prices for the Nocona chips will range from $209 for the 2.8GHz chip to $455 for the 3.2GHz version to $851 for the 3.6GHz version. As is always the case with an new Intel chip, that faster 3.6GHz part will be scarce as hen’s teeth as the company ramps up production. Full volume shipments will probably be in swing by the middle to the end of the third quarter.
As for the server version of the Nocona chip and its associated Lindenhurst chipset, mum’s the word. But the word on the street is that we can expect to see very similar configurations for servers sometime in the middle of summer.
