System setups and notes
Here's a quick rundown of the test system should you wish to compare benchmark results with your own.- Intel Pentium 4 3.00GHz HT S478 Northwood CP (800FSB)
- Intel Pentium 4 3.06GHz HT S478 Northwood CPU (533FSB)
- AMD Barton XP3000+ S462 CPU
- Intel 875BZ Canterwood motherboard (P03 BIOS)
- Asus P4C800 Deluxe
- EPoX 4PC3A+
- MSI 875P Neo FISR2
- MSI GNB MAX2 'Granite Bay'
- ABIT BH7 i845PE at 200FSB
- ABIT NF7-S v1.1 (16 BIOS)
Common components
- ATi Radeon 9800 Pro (380/340)
- 2 x 256MB Kingmax PC3200 run at 2.5-6-3-2 @ DDR-400 for i875BZ
- 2 x 256MB Corsair XMS3500C2 run at 2-6-2-2 at DDR-400 for BH7, NF7-S, and GNB-MAX2-L
- 41.5GB IBM 120GXP Hard Drive
- Liteon 16x DVD
- Samcheer 420w PSU
- Samsung 181T TFT monitor
- Cooler Master Fujiyama heatpipe cooler
Software
- Windows XP Professional Build 2600.xpclient.010817-1148
- DirectX9
- Intel 5.00.1009 chipset drivers
- NVIDIA nForce 2.03 drivers
- ATI CATALYST 3.2 drivers and control panel (6307s)
- Pifast v41 to 10m places
- Lame v3.91 MP3 encoding with Razor-Lame 1.15 front-end using U2's Pop album
- Hexus SETI benchmark
- 3DMark 2001SE
- UT2003 Demo (Build 2206)
- Comanche 4 benchmark
- Serious Sam 2 Demo
- Quake 3 v1.30 HQ
Notes and issues
We're attempting to compare like for like here. That, in theory, should be relatively easy as we have a number of motherboards based on a single chipset. It must be noted that certain manufacturers attempt to slant benchmarks in their boards' favour by artificially inflating the operational speed via exaggerated FSBs. The Asus P4C800D is reasonably guilty in this regard. Running our stock 3GHz / 200FSB P4 at 3030.27MHz, it eclipses the running speed of the EPoX and MSI boards by 9 and 16MHz respectively. EPoX sent us a new BIOS that added 14MHz over and above the initial BIOS. It seems as if there should be some kind of regulation over FSB speeds.
The Asus motherboard had a few idiosyncrasies. Running at what it perceived as stock speeds, i.e. 202FSB, it would lock-out the memory timings irrespective of what was entered into BIOS. Those timings with 2 x 256MB sticks of Corsair XMS3500 C2, though, were locked at 2-5-2-3 timings. Raising the FSB past 202 allowed one to actually define the various parameters correctly. Other than that there were no issues. The board booted successfully each and every time and it ran through the various benchmarks without a hiccup. The Asus board was the first in the labs. Through extensive consultation with Asus UK, a number of BIOSes were released that incrementally increased performance.
The MSI, too, showed a few peculiarities. It also set SPD timings regardless of what was entered. These timings were far more relaxed at 2.5-8-3-3. Further, manually setting lower latencies forced the board into dropping down to DDR-320 speeds at 200FSB. MSI assure us that this is a known issue with press boards that doesn't affect retail samples. In light of this, the board has to be run at the higher latency of 2.5-8-3-3 at 200FSB / DDR-400. The MSI's temperature reading seemed a little too high in comparison to the other 2 retail boards. Voltages seemed to fluctuate and spike that little bit more, too.
The EPoX board seemed to take a while to boot each and every time, but the excellent debug display informed you as to the board's status. No rebooting, memory, or stability problems to report here. It appears to apply whichever memory settings you choose from within BIOS, albeit with a slightly inflated FSB. CPU-Z v1.17 seemed to think so.
FSB Overclocking
A semi-unlocked 3.0GHz P4, from 12x - 15x, allowed us to push the boards' FSB right up in the hope of finding their limit. Running at 12x with the lowest RAM ratio, the boards were booted in 10FSB increments from 200FSB. We've heard that the Canterwoods only use the very finest silicon, so how far would they go ?
300FSB - Asus P4C800 Deluxe
277FSB - EPoX 4PC3A+
268FSB - MSI 875P Neo FISR2
We shouldn't infer that each and every board by a certain manufacturer will overclock equally far, but the opportunity of purchasing a cheaper P4 C, overclocking it heavily via the FSB and running memory at a lower ratio to the FSB seems like a successful recipe for ultimate performance.
All that with a large passive heatsink. We're pretty CPU limited here.
