Overclocking the Athlon XP3200+

LIKE ITS PENTIUM 44 3.2 GHz competitor, the 2.2 GHz AthlonXP 3200+ "Barton" is supposed to be the last in its line, to be followed by a brand new replacement (in this case, Athlon 64 of course). This CPU is the peak of the three-year old 32-bit Athlon platform, the first serious competition to threaten (and often take away) the Intel performance lead - at least until the maddening "Northwood" Pentium 4 ultraquick frequency ramp-up over the past year.

XP 3200+ has everything maxed out: the L2 on-chip cache is at 512 KB, the FSB reaches 400 MHz throughput (200 MHz DDR), and the matching Nvidia Nforce2 chipset provides a dual-channel DDR400 memory subsystem to provide full FSB bandwidth to the CPU even if AGP 8X, PCI and peripherals are all continuously accessing the memory as well.

I was keen to know how much more frequency headroom is there beyond 2.2 GHz on this last Barton, using reasonable overclocking gear like a very good heatsink-fan combo, and a flexible, overclocker-friendly mainboard.

So, I assembled an interesting configuration using XP3200+, with CoolerMaster Aero 7 high-performance copper heatsink and turbine-like fan on top, 2x256 MB OCZ PC3200 Platinum EL DDR DIMMs in dual-channel setup at 2-3-2-6 latency settings, MSI GeForceFX 5600 128 MB 3-Dcard, all that sitting on the Asus A7N8X board. And all of that inside a modern, Korean-made 3R r101 metallic midi-tower casing with dual large 120 mm fans, temperature reader front-panel LED, and sliding storage bay door. (See pictures at foot of this article).

Since the world still revolves around "popular" Windows platform (well, cow dung is popular too - billions of flies love it), we had no choice but to run some Windoze-based tests to obtain comparative results - to simplify things, we stuck with Sandra2003 and PCMark2002 tests under WinXP Pro SP1.

Besides the default settings, we also ran one setting where the CPU FSB multiplier was changed to 12x to reach the 2.4 GHz actual operating frequency (someting like PR3500+ rating), one where the FSB became 202x2 MHz to reach 2.424 GHz CPU clock, and another set where the CPU FSB multiplier stayed at 11x, but the FSB and memory clocks went up to 210x2 MHz and 220x2 MHz for 2.31 and 2.42 GHz CPU speeds respectively. This was we could see how the various results sway depending on the varying CPU and FSB-memory setting combinations.

While the voltages could be adjusted as needed, one rule stayed - the fixed 2-3-2-6 memory latency. I believe that allowing much higher latencies like 3-4-4-8, for instance, just to get the higher memory clock, may not result in real net performance benefit.

Here are the numbers!

As you can see, PCMark was more sensitive to the CPU instability once we went above 2.4 GHz. In the 12x202 MHz case, PCMark2002 consistently crashed at CPU Test 4 (audio conversion), while in the 11x220 MHz case, it (again consistently) reported the MFC lib crash before the Test 1 (JPEG decode) would complete. I suspect CPU to be the problem here, as memory was basically not overclocked at all in the 12x202 MHz case. Sandra completed fine in all cases anyway. I also tried the 10x240 MHz setting, but the system didn't even boot with it.

CoolerMaster's Aero 7 heatsink-fan looks more like a jet engine on a copper pedestal than a PC fan. While its 4,500 rpm rated speed (we actually measured close to 4,700 rpm on test system) is nothing special, the horizontal air intake may help bring in more air, so to speak. The adjustable fan speed button (either via front panel or PCI slot opening) may save you a bit of power if leaving the PC idle or not too active.

The OCZ PC3200 Platinum EL DDR memory behaved pretty well in the test: while we couldn't get the perfect 2-2-2-6 setting, the 2-3-2-6 setting worked well and stable even at 220x2 (440 MHz DDR) clock, and I suspect it could go even higher provided the CPU & board allowed it.

While having only two fans, 3R r101 casing ventilation was truly mind-blowing (no pun intended) due to the sheer size of the fans - the hard-disk bay and the cable mess around the mainboard were not much of an obstacle to large fan diameter and resulting wind. When I inserted their sensor for the CPU thermometer LED display, the results were a bit different than what BIOS reported: for instance, in the default case where BIOS reported 43 C, the display showed 41 C, but in the 2424 MHz case, where BIOS reported 51 C, the display showed 55 C. But that's OK, I guess.

In summary, in a fine-tuned system with top-notch heatsink, well-ventilated case and an airconditioned room (or Alaskan/Icelandic/Siberian/Tibetan weather, pick your choice), you should be able to get between 8% and 10% extra oomph from your Athlon XP3200+, CPU frequency-wise. And that probably is the limit of the Barton platform anyway - at least we know AMD could probably do a 2.4 GHz "Barton" XP3400+ just in case Athlon64 still needs some work done... µ

THE R101

THE COOLERMASTER

OCZ MEMORY