Originally posted by: DavidHull
IT's been well known that currently you can't go above 240 FSB with 1:1 memory ratio. Look, I'm not sure if you understand it, but the memory speed has VERY LITTLE effect on performance. For example, if you look at this Anandtech article where they compare various memory modules in different speeds, you'll see that the OCZ 3200 PL R2 scores 516.3 frames per seconds in Quake3 Arena when the system is run at 12x200, (CPU is at 2400 mhz, and RAM at 400 mhz DDR). At 267x9 ( CPU at 2400 mhz, RAM at 533 DDR), the score is just 525.8 fps - less than 2% improvement for 133 mhz increase in DDR clock rate (note that the CPU speed is the same in both cases). . .
What you fail to realize user1234 is that AT ran the memory at 1:1 at all times and only played with the CPU multipier and HTT speed. Just because a faster memory speed, with the CPU speed remaining constant, doesn't increase performance by a significant margin, it doesn't follow that you should drop the CPU/RAM ratio. You are taking a hit in performance versus running 1:1, even if your CPU is clocked 300 - 400 mhz higher. That is why many of us are still running at 240 mhz.
Memory has a gigantic effect on performance, but not by itself - it needs to work in unison with the CPU to return the best results.
Yes, amusing but completely untrue. What you call 1:1 or a different divider has absolutely no performance hit whatsoever, except the fact that changing the ratio changes the memory speed. I read all the details in many articles reviewing such experiments and in all of them the colcnousion is that there is no penalty for running not 1:1 on the Athlon 64 platform. Now I can understand where these halucinations are coming from, mostly because in the past data had to travel from the memory modules to the chipset and then again from the chipset to the cpu (using the fsb). But for A64 the fsb doesn't exist as an external pathway, it's internal to the cpu. That's why the A64 doesn't have the same chipset dividers of other platforms, like 5:6, 4:5, 3:4, 2:3 etc. The ratio, or dividers, for A64 are purely CPU clock to memory clock, so the memory clock is usually something like CPU/9, CPU/10, CPU/11, etc. It just so happens that in the so-called 1:1 case, the CPU-multiplier of the external clock is equal to the cpu-to-memory divider, but they are completely uncorrelated, so changing either one of these ratio doesn't affect the other. For example, if the CPU is running 10x200=2000, and the memory is running at CPU/10=200, it would be 1:1 in your book. Then if we increase the FSB to 250 and drop the CPU multiplier to 8, and leave the memory divider at CPU/10 (assuming the bios has such a setting, equiv to 4:5 divider), we'll get the CPU running at 8x250=2000, and memory at 2000/10=200, so it's the same aboslute speeds, as before but it's no longer 1:1, because the FSB:MEM is now 250:200. Would you say that there is a penalty for this setup which would make it slower than before ??? If you think so, you need a reality check.
If you don't believe it, just go read about it, or more simply just run some benchmark at 1:1 and non-1:1 setting and convince yourself that it doesn't matter. All that matter for A64 is the absolute CPU speed, and absolute memory speed, the FSB doesn't matter (as well as the derivative 1:1 concept which hinges on the notion of FSB, but FSB doesn't exist as an interconnect device in this platform, it's only a term used to refer to the external clock rate).
What you need to internalize is that there is no such thing as CPU and memory working in unison at some magical setting, because how does the CPU keeps working in unison with the memory when we change CPU multiplier alone ? obviously, this changes the internal cpu/mem divider so the memory keeps the same speed as the fsb, so now "working in unison" suddenly means a different divider? this kind of silly theories make no sense, it's easier to just look at the reality, which is simply that there is ALWAYS A SINGLE DIVIDER involved in memory accesses on the A64 platform.
Kapish ?