[masotime]

I'm completely new to overclocking processors, and I've been trying to read up literature on the subject which, IMO is enormous....

Let's say I want to focus my scope on the A64 microprocessor. I would also like to state that I do not wish to be an extreme overclocker - i.e. overclocking to get best bang for buck, not world-class scores.

What I understand (vaguely) is that having extremely low timings would help to increase memory bandwidth (?) due to the minimization of inefficiences of the microprocessor waiting for the memory to respond with data.

Do high timings actually limit the maximum FSB overclock that can be achieved? Or do they negate the advantages obtained by having a higher FSB even though a higher overall CPU speed is achieved?

If not, does it really impact real-life performance of most normal applications? (not those synthetic benchmarks, e.g. real gaming performance, etc.)

I am asking this because I have read that it is not really worth it to purchase ultra-expensive RAMs just to overclock, and that you can achieve a relatively good overclock without the need for extremely-high-performance RAMs.

[lennardseah]

at most 10 to 15%. the extra money for the premium rams could be spent on a better graphics card or even more rams.

FSB and latencies dun go hand in hand. theres a tradeoff somewhere

Originally Posted by masotime

What I understand (vaguely) is that having extremely low timings would help to increase memory bandwidth (?) due to the minimization of inefficiences of the microprocessor waiting for the memory to respond with data.

That's not true. The low timings would reduce the time that the processor "waits" to get the data back and forth the memory sticks. It does not affect the bandwidth.

[masotime]

Originally Posted by lennardseah

at most 10 to 15%. the extra money for the premium rams could be spent on a better graphics card or even more rams.

Yes, that's what I heard/seen from a number of articles on this issue. And I have to agree with them, unless you're aiming for world records

FSB and latencies dun go hand in hand. theres a tradeoff somewhere

What I think I understand is that the latency calculations are based on clock cycles... so if we were to take for example 2 clock cycles @ 200 Mhz (5 ns per clock, total 5x2=10ns)... that's effectively 4 clock cycles @ 400 Mhz (2.5 ns per clock, total 2.5x4 =10ns) - if I'm right that would explain why it's harder to have low latency @ higher frequencies coz the real latency timing in (ns) would have to be much smaller to achieve the same "clock cycle" rating.

[masotime]

Originally Posted by Krado

That's not true. The low timings would reduce the time that the processor "waits" to get the data back and forth the memory sticks. It does not affect the bandwidth.

Sorry... guess I'm still not used to getting the terminology correct

Hmmm.. what I wanted to say was that it was because of the processor waiting that would cause the overall performance to be lowered i.e. that is the disadvantage of high latencies. Correct me if I'm wrong.

However, to what extent is the "waiting effect" going to be a significant effect on the processor's performance in various real world applications? e.g. Scenarios like

-> Adobe Photoshop operations on large files.
-> 3dsmax rendering
-> Gaming on e.g. Doom 3 / Half-life 2

:confused5

If I understand it correctly, it's possible to have a high FSB with high latencies i.e. the processor is still overclocked, just that it wastes time waiting for data from the data bus. Some simple reasoning might suggest that any operations involving massive data transfers from RAM or perhaps multiple random unpredictable accesses to RAM would increase the impact of latency?

[yantronic]

For office applications and Photoshop 1GB of CL3 is gonna beat 512MB of C2 for sure. There is a lot of things Photoshop dumps and retreives so RAM amount matters.

Gaming wise most of the framerates is GFX dependent. As such you won't find a noticeable difference between a CL2 and CL3 stick in a system playing Doom 3 or HL2. HL2 especially has a fine efficient engine that scales down based on your system specifications.

CL2.5 is prolly the best tradeoff.

If benchmarks matter, go CL2.

[MacClipper]

Expensive high end RAM is usually not worth the money for the bang it gives. Well, if you have money to throw and want to have a flashy rig sig...

OTOH, hit while the iron is hot and you may just get UTTs/BH5s, TCCDs at reasonable or even low prices if you know which RAM is which.

[tckrto]

depends on luck and mobo compatiblilty oso. like my msi neo2 plat cannot even clock the twinmos pc3500 bh-5 above 230mhz wif 3.4 Vdimm :sad4: