Higher FSB = Higher Memory Latency

[Tropicocity]

Hi all, long-time viewer first-time poster here

Just wondering about this, looking for some input from experienced OCers, or indeed just general-knowledged people.

Recently went about experimenting with a higher fsb and lower multi on my cpu, as I'd heard it was "better".
Edit: I'm strictly asking this from a gaming point of view, as well as general PC operations

e6600
2Gb OCZ PC8500 Sli-Ready
Asus P5B Deluxe

Was running a 377 fsb with the default 9x multi on the e6600 for a 3.4ghz, with a 1:1 FSB:RAM Ratio, giving me effective DDR2-754 speed with 4-4-4-10 at 1.8v.

Decided to reboot comp and change it to a higher fsb with lower multi to see what I could end up with.

Now running a 425 fsb with an 8x multi for the same clock of 3.4ghz, 1:1 for DDR2-850 speed, still with 4-4-4-10 but at 1.9v.

The thing I'm looking at is Everest Ultimate's memory benchmark tests.

With my old setup I was getting around 7.8 Mb/sec read, 6.8Mb/sec write with a memory latency of ~57ns.

On the new setup, with clock speed, fsb:ram ratio and timings all kept the same, but now "faster" operating ram, I'm getting 7.6Mb/sec read, 7.75Mb/sec write, but I'm up to 68ns latency.

Now i'm fairly good with overclocking, I'd say I had a decent level of experience with it, my question here is whether or not what I've done is justified?

Is DDR2-850 really faster than DDR2-754 with the same timings, given that my read speed has dropped slightly and memory latency increased greatly?

Am i better off trying to achieve the highest FSB with a lower multi and nice high 1:1 ram speeds (voltage aside) or would i be better off reverting to my 9x multi and using modest fsb speeds with an effectively lower ram frequency but reduced memory latency? (10ns difference mind you)

Any help would be appreciated, thanks

[SoulsCollective]

Congratulations, you've encountered a strap change.

For more on this and the technical explanation, see Here or search XS, but basically the NB has internal latency as well as raw speed, much like RAM does. As you raise the FSB, the motherboard switches strap to maintain stability at several predetermined points, in much the same way as you need to loosen timings when raising RAM speed.

Play around a bit, see if you can find out when exactly the strap change occurs, and raise your FSB as high as you can without triggering it.

[Tropicocity]

Congratulations, you've encountered a strap change.

For more on this and the technical explanation, see Here or search XS, but basically the NB has internal latency as well as raw speed, much like RAM does. As you raise the FSB, the motherboard switches strap to maintain stability at several predetermined points, in much the same way as you need to loosen timings when raising RAM speed.

Play around a bit, see if you can find out when exactly the strap change occurs, and raise your FSB as high as you can without triggering it.

I've heard about the strap thing, apparently it changes strap at 380 or something, which was why i was avoiding the 380 fsb and opting for a 377 i think.

I read the link you gave me but i'm not particularly understanding any of it even after a couple of reads xD Possibly the lack of sleep.

Would you be able to offer any advice as to possible FSB i should be setting it at? I'm running a modest 3.4ghz overclock but I'm willing to go higher for this no problem. thanks

[Slay0r]

Strap changes at 399 on p5b-d (you can test this by booting at 398 and then going into the bios and setting 399 and rebooting. The board will shut down to allow strap changing and then boot up, thing that doesnt happen when you're going from say 350 to 351 or 450 to 451, those will change on the fly )

I'd say if you cant push 480-500 fsb, set a higher multi and stay underneath 400 for better latency and speed overall.

[saaya]

boot at a lower fsb and then increase fsb in windows.
it used to be possible to do this with clockgen, just create a shortcut to a clockgen script file in the startup folder and it would automatically change the fsb at load
i think cpu fsb can do this too, and it supports a lot of plls, so it might work. cpu fsb isnt freeware tho, i think it costs 15$ or so

[saxovtsmike]

I´ve tried the same with my setup too but I canged from 360x9 ram 4:5 (450mhz 4-4-4-12) to a corresponding setting with cpu multiplier at 8.
the result was really lame as i had the strap change and only 405 fsb and I had to set the ram divider to 1:1 as my supertalents do not like 500mhz even with 5-5-5-15 timings.
I was running everest tests and pi1M and all test except 1 spoke for the first setting.
As I´m more gamer than everything I´m havin the sys running as I had it b4 (360*9 ram 4:5)as it is primestable and a couple of lanparties made absolute no problem.
You could give the ram divider a try.. it did it for me..
=> One thing to add: Fix the NB Voltage to 1.45V as on auto it went tits on mine (=1.55) when I thightened up the timings and that resulted in THE p5b pcie x1 bug where you get only pcie x1 instead of pcie x16 for your gpu.

[PnoT]

What tRD (Performance Level) are you running at each FSB?

[BTK]

use memset to change TRD to test and then implement via BIOS

except only some TRD is possible at certain CAS and FSB otherwise it wont boot