(4) What is the difference between the mobile Athlon 64s and the desktop revision Athlon 64s

(a) Both CPUs feature a multiplier half lock. In other words, like the desktop chip, the mobile chips' multiplier is locked at a maximum number but unlocked for all lower multipliers down to 4x.

(b) The mobile processors are all, except for the 3400+ DTR, newer CG stepping CPUs. Most desktop CPUs remain the older C0 stepping. The advantages to the CG stepping is that it runs much cooler than the C0 revision chips. Max TDP for the line of C0 revision cores is 89W compared with a max TDP of 81.5W for the CG steppings. The CG stepping of the Athlon 64 also overclocks better than current desktop C0 revision chip. The CG stepping CPUs also feature an improved memory controller.

(c) Because the mobile Athon 64 CPUs are not desktop processors even though they can physically fit in the desktop socket, the BIOS of the motherboard you want to use must recognize the microcode. Otherwise the system will either fail to boot (most common problem) or the CPU will lock down to 800mhz (4x mutliplier). One user has claimed to be able to use clockgen to change the multiplier; I am unable to confirm this.

(d) The mobile CPUs lack the Integrated Heatspreader (IHS) common to the desktop processors. This allows better cooling of the CPUs as heatsink contact is directly with the CPU core itself. But as a result of this, the CPUs are "shorter" than their desktop counterparts and heatsink contact is a very real concern. Most backplates, whether included with the motherboard or the ones AMD includes with their retail chips will not work properly - the screw poles aren't deep enough. Only one heatsink setup works properly without modification with the mobile CPUs, the Thermalright SLK-948U Heatsink.


(5) So, I want to go mobile, but which CPU should I choose?

Here are some factors to consider:


(a) The difference between 1MB of L2 cache and 512k of L2 cache at the exact same speed is approximately 5%. What that means is that you would need for example a 2.1Ghz 512k L2 cache Athlon 64 to equal the performance of a 2.0Ghz 1MB L2 cache Athlon 64. So, obviously since all CG cores overclock to the same range, 2.4-2.5Ghz on aircooling, one should prefer a CPU with a full 1MB of L2 cache.

(b) Get a CG core. Simple. They overclock better and run cooler

(c) Given the choice, the obvious CPUs to select are the 3200+ DTR or any of the 1.4v mobile CPUs. These all feature a full 1MB of L2 cache for maximum performance. Please note that the lower the speed the lower the mutliplier of the chip and the higher the HTT/FSB needed to get a succesfully overclock into the 2.4-2.5Ghz range.

(d) Wait, but aren't the 1.4v mobile Athlon 64s much cooler running than the DTR processors? Actually, no they're not. Remember that when AMD rates thermal output they only give the max output for the single top of the food chain CPU. Thus, for the desktop and mobile C0 stepping, the highest is a 3400+ at 89W. For the CG stepping DTRs the highest is a 3400+ at 81.5W. Now here's where you need to pay attention. The highest CG stepping 1.4V mobile CPU is the 3200+ and NOT the 3400+ so the TDP listed of 62W cannot be compared to the other CPUs. In fact, if you compare it to the DTR CPUs and do a little guestimation (don't know if there's a linear relationship between clockspeed and heat) you'll see that these are basically CG DTR CPUs running at 1.4V These are just rebranded DTR CPUs.

(e) But still, shouldn't the 1.4v mobiles overclock higher than the CG DTR? Not at all. Given the choice there's no reason to not pick up the 1.4v mobiles but all CG stepping CPUs hit a wall at ~2.5Ghz with aircooling, whether they default at 1.5v, 1.4v, or even 1.2v


(6) Do I need to run 1:1 memory to achieve peak performance on the Athlon 64?

Not really, given the Athlon 64 architecture and the integrated memory controller, there is no statistically significant advantage to running memory at a 200 setting versus a 166 setting at, for example, around DDR400 speeds. Even at the 200 setting the memory in an Athlon 64 system is running async. The memory speed is arrived at by using a divisor off the Athlon 64 CPU speed (i.e. a /12 divisor.) Also, unlike the Pentium 4, the Athlon 64 is not bandwidth starved; thus, the difference in memory speeds between DDR400 and DDR 500 is at most 1-2%. In other words, don't go crazy and throw away perfectly good PC3200 RAM thinking you'll see a significant boost using DDR500 for example; you won't, and your wallet will hurt.