Test Setup

Liquid3D's Test Setup
CPU	Intel Pentium 530 (3Ghz)
Mainboard	Abit AA8 (i925x, BIOS ver1.5)
Memory	* 2 x 512Mb Kingston HyperX 5400 DDR2
Other components	- ATI X800XT (Catalyst drivers 4.9) - OCZ PowerStream 420W (OCZ420ADJ) - Danger Den TDX kit for S775
Software Used	Windows XP SP2: - Sisoft Sandra Professional - CPUID and CPU-Z - Aida32 - PiFast - Futumark’s 3DMark2001SE / 03 / 05

First off I ran the system at Default FSB speed, which the Abit enhanced Award Phoenix BIOS increased the FSB to 204MHz, from the usual 200MHz. Motherboard manufacturers will employ this BIOS tweak to give their motherboard a slight advantage over the competition in benchmarks. None-the-less, I reduced the frequency to 200FSB.

Earlier I mentioned DDR2 implementation of dividers and this is because of its increased speed over that of current CPU FSB speeds. We’re testing PC2-5400 which effectively runs at a frequency of 667Mhz, in order to run synchronously or at 1:1 ratio, the FSB formula would have look like this; 15x 333MHz = 4995MHz. Obviously this would be a difficult overclock for a 3.0GHz CPU to maintain. Thus the reason for the dividers we see on i915X and i925X chipset based motherboards.

What became quite frustrating was my inability to adjust either the FSB:DRAM ratio, or the PCIe bus speed in the Abit enhanced Phoenix/Award BIOS. Any attempt to lock the PCIe, or change the memory divider resulted in a failure to post. I was unable to take advantage of the 2:3 DDR600 setting; therefore the remainder of the review will employ a 3:4 ratio. This is what most End-users would run regardless. In so far as the PCIe bus, this was a bitter pill to swallow. An overclocked bus can damage a video card, and at $600, this ATI X800 XT for PCIe wasn't all that easy to find last month. To run the memory at its rated speed required an overclock of 260FSB, or 3900MHz. While my P4 530 CPU was up to the task, anything above 3900MHz wasn't completely stable. Problem is at this frequency the PCIe bus is running at 120MHz. I hope the Overclocker's out there will appreciate my willingness to risk damage to my hardware in overclocking my graphics card, to attain these higher frequencies.

First a screenshot identifying the memory at default bus speeds.



The divider is shown in the CPU-Z portion of the screen at 3:4. At 200MHz, the memory is running at 266MHz (DDR = 533Hz) bandwidth at this speed isn't much more then that of DDR, however; given the latencies, it's doing quite well. Prior to conversion to tables, let’s take a look at the highest speed attainable by the system 260FSB or 3.9GHz. This limitation had nothing to do with the memory, nor the CPU which ran quite cool.



Thanks to Danger Den’s high end TDX water cooling kit, temps remained quite low while overclocking this SL7KK Malay P4. The CPU easily booted (Posted) into 4.0GHz (267FSB) however it would crash once Windows tried to load. From discussions with more knowledgeable Overclockers, it's most likely the SATA controller holding back the system. Seems there may even be an issue with Maxtor SATA drives specifically. I’m looking into it and will report my findings. Problem with the PCIe boards is their lack of multiple IDE connectors’ as well. I digress, here's some temps:





Benchmarks

Onto our bandwidth tables for both SiSoftware Sandra and Aida32:




The bandwidth at 260FSB or the memory's rated speed is excellent. While DDR does come very close, we must remember we’re attaining this performance level at lower voltage, lower operating temp, and faster speed which will synchronize with future CPU FSB speeds. It's difficult to see the immediate benefits for DDR2 from a consumer perspective, however; in the long run it's less costly to manufacture, and easily reproducible.

Onto the 3D performance marks. I ran Futuremark 3DMark2001, 3DMark03, and 3DMark05, all on the ATI X800 XT 256MB, running at its default clock speeds.



While it's difficult to judge memory performance in 3D marks, they do provide an example of how the memory boosts the overall system performance. It’s especially visible in the 3DMark2001SE benchmark which is more of a global benchmark and less GPU dependant like the 03/05 versions.


Finally were going to examine Hexus PiFast which is not as impressive as I've seen on other systems. This is most likely indicative of the higher latencies innate to DDR2. Hopefully these latencies will tighten up a bit, but more importantly as speeds increase PiFast times will reduce.




As you can see there quite a reduction, almost 20-seconds from 3.0GHz to 3.9GHz clock speed. This is indicative of the core architecture of 90-nanometer Prescott, and although Intel has renamed the Intel P4 530, it's still a 3.0E with prima facie (increased FCPGA count) changes. Essentially, Socket-775 gives the .09 P4 its much needed increase in current, which was causing Socket-478 motherboards PWM (motherboard CPU power circuitry) temps to run so high as to damage many motherboards. In my article Prescott Effects on motherboard power circuits PWM temps ran as high as 72°C while overclocking a 3.0E on an Abit AI7. While some motherboards such as the Asus P4C800E-Deluxe have more robust PWM circuitry which doesn't overheat as badly, motherboards designed for the P4 Northwoods are not adequately equipped to power the 90-nanometer Prescott.

Socket-775 ameliorates this problem while introducing Intel's i915x and i925x chipsets. Intel has also allowed board makers to equip i875x and i865x chipsets with Socket-775 support for those wanting to take advantage of the power circuitry designed for the .09-micron CPU. While it's true that the P4 for Socket-775 has changed somewhat, one must remember its core architecture remains unchanged, the pipeline depth still requires high frequencies, and low cache miss rates for performance to match that of Athlon64 with its on-die MCH.

Between slightly higher latencies, and pipeline depth these contribute to the sluggish PiFast scores above.