AMD made a huge splash with the introduction of their Socket 939 platform, which allowed dual-channel memory performance using standard DDR. Since that fateful announcement, AMD has continued the performance offensive, launching higher-speed Athlon 64 and Athlon 64 FX models on a regular basis. There were only two holes in that strategy, the most obvious being the need for lower-cost 939-pin processors, which would take away another of the Intel strongholds. Another key element was to migrate the Athlon 64 core from the 130nm (0.13-micron) Newcastle to the 90nm Winchester, thus joining Intel at the 90nm table.
AMD has since attained both of these goals, and has released their 90nm Winchester in a number of 939-pin iterations. The Athlon 64 3500+ matches its 130nm predecessor and runs at 2.2 GHz, while the Athlon 64 3200+ and 3000+ hit 2.0 and 1.8 GHz clocks speeds, respectively. These last two processors fill a very real market niche for AMD, as Socket 939 supplies dual-channel DDR for the masses and these processors hit the needed sub-$200 price point. The single-channel DDR controller on the Socket 754 processors is certainly powerful, but keep in mind that Intel holds the mindset by allowing even entry-level Celeron CPUs to enjoy dual-channel DDR performance.
In terms of outward physical design, there is no difference between the current 90nm and 130nm Athlon 64 Socket 939 models, and the internal architecture also remains mostly unchanged. The 90nm Athlon 64 3500+, 3200+ and 3000+ feature 128K L2 and 512K L2 cache levels, and include an integrated dual-channel DDR controller. The core voltage has been lowered slightly, with the new 90nm processors running at 1.4V, down from the 1.5-1.55V of their 130nm cousins.
The most important change concerns motherboard support, and a BIOS update will likely be needed in order to properly initialize these new processors and ensure proper voltage levels. Those looking for 90nm models also need to be careful with the Athlon 64 3500+, as it is available in both 90nm and 130nm variants, but the 939-pin Athlon 64 3200+ and 3000+ are 90nm-only parts.
The benefits of a die shrink from 130nm to 90nm can yield important benefits, both for consumers and AMD. The smaller die size allows for lower operating voltages, lower power consumption, and cooler operating temperatures, which are all pluses for PC buyers. For AMD, one obvious factor is the physical die size of the Winchester core, which has dropped from 144mm2 (square millimeters) to only 84mm2. This can result in significantly higher chip yields and lower-cost per chip, which can be integral when AMD is already battling the 90nm Intel Prescott-based processors.
Since AMD has introduced the new 90nm Winchester core at only 1.8 to 2.2 GHz clock speeds, core temperature is obviously one of the main benefits. The Athlon 64 FX line and Athlon 64 4000+ and 3800+ outstrip the 90nm entrants in overall performance, but the smaller core should provide lower temperatures and power consumption than a same-speed 130nm processor. Measuring core temperatures can be an exercise in frustration sometimes, as different motherboards can yield different results and it is impossible to run comparisons between two models.
In this case, we needed to settle on a single Socket 939 board and after testing a few K8T800 Pro and nForce3 Ultra boards, the MSI K8N Neo2 Platinum (nForce3 Ultra) proved to be the most consistent for temperature evaluations. The heatsink-fan for this was a high-end ThermalRight copper cooler with ShinEtsu compound. For this test, we took both 90nm and 130nm Athlon 64 3500+ processors, measured them at idle, and then ran a series of CPU-heavy benchmarks and took a load-type reading. The results are presented in the chart below.
At idle, the 90 nm Athlon 64 3500+ displays a significant advantage over its 130nm counterpart, to the tune of over 8 degrees Celsius cooler. This shows the inherent advantage of a smaller core, but it is just as important to determine core temperatures while under heavy stress. In this scenario, the gap shrinks considerably, and the 90nm advantage has been cut almost in half. This is not totally unexpected, as the 90nm core not only uses less power and runs cooler, but is also physically smaller with less real estate for cooling. This is an important consideration, and with only a 0.1V drop in core voltage, and an approximate 40% decrease in core dimensions, there is a definite trade-off in the AMD 90nm transition.
Overclocking is an important factor in evaluating any new CPU core, but as this does void your warranty and can still be risky, we didn't want to make it the focal point. In our basic testing, we maintained the 11X multiplier and managed to achieve a clock speed of approximately 2.55 GHz at a bus speed of 232 MHz. Core voltage was increased from 1.4V to 1.5V (and higher with no change), and although moving the multiplier down resulted in better performance (due to the higher bus) but we were still unable to surpass the 2.55 GHz speed by any noticeable amount.
These results are certainly good news for AMD enthusiasts, especially those looking at racking the less-expensive 90nm Athlon 64 3200+ and 3000+ to the limit. Given our results, it will only be a matter of time before higher-speed 90nm processors start appearing, potentially replacing the Athlon 64 3800+ and 4000+ 130nm models. All overclock testing was performed on the same K8N Neo2 Platinum (nForce3 Ultra) motherboard and ThermalRight cooler/ShinEtsu compound combo as we used in the core temperature section.
