from neoseeker
Overclocking on the Athlon 64 is more complicated than on other architectures because of the backplane design and the often misunderstood relationship between the LDT and HyperTransport bus. When overclocking, oft-repeated is that "turning down your HT link speed has little effect on performance", or that "running with a memory divider is for weenies". As was the question with our Multimeter vs. Motherboard article, I had to ask myself "Has any of this ever been quantified and published?"
The answer, as a matter of fact, is "sort of". Nonetheless, I wanted to run my own set of tests in a controlled environment, testing the effects of every difference in settings, and publishing the results in one, unified article. Before setting out to run any kind of tests, I planned out exactly what I would test and how I would test it. I settled on the following:
Tests Performed
XviD Encoding With Auto Gordian Knot
SiSoft Sandra Arithmetic/Multimedia/Memory Bandwidth
LAME MP3 Encoding
Half-Life 2
Doom 3
Testing Methodology
1. Changing HTT link speed by adjusting the multiplier, keeping everything else constant
2. Changing the LDT (analogous to the front side bus for those who have dabbled in overclocking before) bus speed, using no memory divider (1:1), while keeping the total CPU clock constant by adjusting the CPU multiplier
3. Changing the LDT bus speed, keeping the CPU multiplier at 9x, while keeping the memory speed as close to 200 MHz as possible by adjusting the memory divider
4. Changing the LDT bus speed, keeping the CPU multiplier at 9x and the memory divider at 1:1 (regular overclocking)
More
Overclocking on the Athlon 64 is more complicated than on other architectures because of the backplane design and the often misunderstood relationship between the LDT and HyperTransport bus. When overclocking, oft-repeated is that "turning down your HT link speed has little effect on performance", or that "running with a memory divider is for weenies". As was the question with our Multimeter vs. Motherboard article, I had to ask myself "Has any of this ever been quantified and published?"
The answer, as a matter of fact, is "sort of". Nonetheless, I wanted to run my own set of tests in a controlled environment, testing the effects of every difference in settings, and publishing the results in one, unified article. Before setting out to run any kind of tests, I planned out exactly what I would test and how I would test it. I settled on the following:
Tests Performed
XviD Encoding With Auto Gordian Knot
SiSoft Sandra Arithmetic/Multimedia/Memory Bandwidth
LAME MP3 Encoding
Half-Life 2
Doom 3
Testing Methodology
1. Changing HTT link speed by adjusting the multiplier, keeping everything else constant
2. Changing the LDT (analogous to the front side bus for those who have dabbled in overclocking before) bus speed, using no memory divider (1:1), while keeping the total CPU clock constant by adjusting the CPU multiplier
3. Changing the LDT bus speed, keeping the CPU multiplier at 9x, while keeping the memory speed as close to 200 MHz as possible by adjusting the memory divider
4. Changing the LDT bus speed, keeping the CPU multiplier at 9x and the memory divider at 1:1 (regular overclocking)
More
