More Athlon 64 FX-53 and Overclocking
Athlon 64 FX-53
There are a number of reasons that the Athlon 64 FX is such a large chip. Bigger caches are one; the FX-53 sports 64KB of both L1 data and instruction caches, in addition to a full 1MB of L2 memory onboard. Moreover, the processor incorporates its own memory controller with support for everything from PC1600 to PC3200 registered DDR memory modules. And because itís a 128-bit controller, it requires that modules be installed in pairs for maximum performance.
One direct result of the integrated controller is that the Athlon 64 FX realizes much more real-world throughput than the Pentium 4, which also employs a dual-channel memory bus through Intelís 875P chipset. The Sandra 2004 bandwidth test demonstrates this advantage.
The Athlon 64 execution pipeline is laid out much differently than that of Intelís new Prescott core and its extremely deep, 31-stage setup. The FXís integer pipeline is 12 stages and the floating-point pipeline is 17, making the chip less flexible in achieving higher frequencies, but much more efficient. Thus, it is very responsive to clock speed increases, such as the 200MHz bump seen moving from Athlon 64 FX-51 to FX-53. Again, the benchmarks should indicate the new 2.4GHz chipís advantage.
It should also be noted that, for its size and composition, the Athlon 64 FX-53 has much better thermal characteristics than Intelís Prescott. At 2.4GHz, AMD rates the chipís maximum thermal power at 89W, whereas Intelís thermal design power (not necessarily the maximum thermal power) sits at 103W for its 3.4GHz Prescott. On the other hand, Intelís 3.4GHz Northwood sports a much lower 89W TDP.
Because the Athlon 64 FX sports an unlocked multipler, AMDís Damon Muzny claims that it cannot support the CoolíníQuiet feature offered by standard Athlon 64 chips. Thatís of little consequence, though. Power users interested in a high-end part like the Athlon 64 FX generally arenít interested in clock throttling and thermal conservation, anyway.
When we received our Athlon 64 FX-51, we were told that it was an engineering sample, equal to a production part sans the unlocked multiplier. Thus, our production Athlon 64 FX-53 is the first sample weíve tested without that pesky multiplier locked in place. Now, it should be noted that proper support for multiplier manipulation isnít universal yet. We had problems setting different speeds with our ASUS SK8V, but our SK8N worked without incident. In fact, a recent BIOS revision adds half-steps, enabling 100MHz jumps.
Unfortunately, the FX-53 didnít prove to be an eager overclocker, at least with the stock cooling solution in place. Using a combination of voltages up to 1.7V, we were able to hit 2.6GHz and run a couple of benchmarks, but stability was fleeting, at best. The more stable setting was an even 2.5GHz, easily stable throughout testing.
In 3DMark03, the overclock yielded half a percent of additional performance in the 3D test and nearly three percent in the processor test. It bumped the PCMark04 test up an extra three percent, and knocked the Comanche 4 demo score up three percent at 800x600. Unreal Tournament 2004 benefited a percent and a half at 800x600 and Quake III jumped up a meager percent at the same resolution. The extra 100MHz didnít make an appreciable difference in any of the high-res gaming tests, unfortunately.