Ram speed to max or clock speed to max? Q6600

[Zak33]

Lo people.

On A64's clock speed was all. But I'm just putting a Q6600 into an older Gigabyte board, with some DDR2 800 ram, and I wondered if I needed to get the ram speed back to it's max, in performance terms, over and above the core clock speed of the cpu?

Because the ram to fsb multipliers are 2x, 2.5x, 2.66x, 3x etc and I'm missing a 2.33x or 2.25x.

This chip goes like mad... it boots into XP at 3.6 but is a tad unstable there.. and tbh, that's way too fast for the requirements, but that uses the 2x multiplier, at full 400/800 ram.

Going down to 2.5x drops the core speed to 2.88 at 400, and that's what the ram is... and while that's a lovely overclock, this thing runs sweet as a nut around 3gig to 3.2, so does it matter having the ram at somewhere around 677 to 700 speed?

Core Speed over ram?

[Zak33]

or should I drop the multi?

When I did that, it showed the correct speed in BIOS, but XP showed it back at 3.6.

[Tonka777]

Having the ram slightly slower will only affect the memory bandwidth a little, and you may be able to tighten the timings if its running under spec speed.

Could XP be working the same as Vista with regard to the multi and clock speed? Vista shows the core clock based on the CPU's default multi, not the actual. Try checking it in CPU-Z to confirm either way.

[Pob255]

On socket775 having the ram faster than the base fsb doesn't do much.

Get the fsb & OC to the point you're happy with it, then bring the memory devider off 1:1 untill you can get that as close to 400mhz as poss.
If it's under that's not a major worry, you may beable to tighten up the timings.
You may be able to push the ram over 400mhz all depends on the ram.

Watch the voltages, you may need to put a little more through the cpu to counter v-droop, depends on the motherboard.

[arthurleung]

On socket775 having the ram faster than the base fsb doesn't do much.

Get the fsb & OC to the point you're happy with it, then bring the memory devider off 1:1 untill you can get that as close to 400mhz as poss.
If it's under that's not a major worry, you may beable to tighten up the timings.
You may be able to push the ram over 400mhz all depends on the ram.

Watch the voltages, you may need to put a little more through the cpu to counter v-droop, depends on the motherboard.

I felt that the timing is everything. When I put 8GB of ram into my P5B last year the CPU was starved of bandwidth it effectively only run equivalent to 2.8Ghz (compared to my laptop) while the CPU is at 3.4Ghz. Bandwidth was a main problem there because there are 4 cores to feed. To keep my system stable I had to drop from 5-4-4-12 PL8 to 6-6-6-20 PL12 which is terrible.

If you can get the ram to run faster, do it. You're certainly fine with the Q6600 and any motherboard can easily push 400FSB. I've been stuck with a stupid 435FSB wall with P965/X38 boards which means I can't run my ram anywhere near its full potential.

[Zak33]

I felt that the timing is everything. When I put 8GB of ram into my P5B last year the CPU was starved of bandwidth it effectively only run equivalent to 2.8Ghz (compared to my laptop) while the CPU is at 3.4Ghz. Bandwidth was a main problem there because there are 4 cores to feed. To keep my system stable I had to drop from 5-4-4-12 PL8 to 6-6-6-20 PL12 which is terrible.

If you can get the ram to run faster, do it. You're certainly fine with the Q6600 and any motherboard can easily push 400FSB. I've been stuck with a stupid 435FSB wall with P965/X38 boards which means I can't run my ram anywhere near its full potential.

Those were the same thoughts as I had Arthur. With 4 cores, the clock speed is great, but if it can't access the ram fast enough, it's gonna be waiting anyway.

It's a 965 chipset, so I don't expect miracles, but it's good to know whether settling for 2.88 gig is fine, when I know the thing will run a lot faster but at the expense of ram bandwidth.

I won't push the ram past default, I like ram to be happy in it's job, though as the you good people mention, if it's under speed of it's spec, a tweak on the CAS etc would help.

It feels a shame the the memory multiplier is 2 or 2.5 because a 2.25 or 2.33 would be ace right now

[Zak33]

rightyo ... some playing and some use of Memtest....

Note: I was using an OLD version of Memtest, and these bandwidths are rubbish

Changing the mem multiplier and increasing FSB to bring the ram back to it's 800 speed
(3x266=800 etc)
3x Mem multiplier requires 266 FSB, = 15864 MB/s
2.66x Mem multi requires 300 FSB = 17644 MB/s
2.5x Mem multi requires 320 FSB = 18821 MB/s
and finally
2.0x Mem multi requires 400 FSB = 23526 MB/s

Now this might not be rocket science to the devoted C2D and Quad core overclockers, but I hope it will assist anyone else with an older C2D/Quad board to understand the vast increase in bandwidth available by lowering the mutliplier/divider (according to which was round you look at the data transfer from CPU to ram) and increasing the FSB.

However.. obviously a 400 FSB on a 9x multiplier Q6600 gives 3.6, and I'm happy to say it boots and runs Windows fine, but it's not rock solid.. so a drop of the multi to 8x brings in 3.2 gig quad core and so far..... so good.

[Pob255]

That's a little confusing as most motherboards/bios's run a ratio of the fsb then multiply by x2
ie if the fsb is 266 and the ratio is 1:1 then the memory runs at 533mhz

This is what I was getting at by running the memory higher than the fsb is not that important, to clarify that's running it higher than a 1:1 ratio before the x2 multiplier
(AFAIK due to the simutaionious read+write nature of ddr, pc2-6400 is only running at 400mhz but can be both read and written to at the same time so is rated at 800mhz)
Remember this is down to how the base fsb on a 775 motherboard works, mainly on the north bridge that is running the memory controller.
As your tests show nicely, it doesn't matter if the memory is running at high speeds if the controller is only feeding it to the cpu at a far slower rate.

In your first test, 3x Mem multiplier requires 266 FSB, = 15864 MB/s, while the memory is working at 400mhz the cpu is only talking to at 266mhz the extra speed means the memory is just sitting there waiting for the next command.
What you really need to do is tests at the same fsb with different ratios to see what effect running the memory faster than the FSB has.
Basic rule of overclocking or any logical analysis only change one factor at a time, while on the surface your numbers there look like you're only changing one thing the base fsb, that one factor is the base of the chipset as well as the cpu and the memory.
you'd need to also adjust the cpu multiplier each time so the cpu speed is also stactic, then your results will show the effect of increasing the Northbridge speed on memory bandwidth.

[Sair33]

you'd need to also adjust the cpu multiplier each time so the cpu speed is also stactic, then your results will show the effect of increasing the Northbridge speed on memory bandwidth.

why would cpu speed and cpu multiplier effect the memory bandwidth?

edit:Zak on Sair's PC... obviously

[Zak33]

back on my log in



that's what the screen looks like on this Gigabyte 965P-DS3 mobo

[Zak33]

What you really need to do is tests at the same fsb with different ratios to see what effect running the memory faster than the FSB has.

I'll do that now.. just cos you're lovely

edit: tests done with LATEST version of Memtest (you were right Arthurleung/Steve.. ty both)

266 FSB x 3 multi = 3798 MB/s
266 FSB x 2.66 muti = 3665 MB/s
266 FSB x 2.5 multi = 3555 MB/s
266 FSB x 2 multi = 3257 MB/s

[arthurleung]

10000MB/s+ is certainly wrong, you're looking at L1/L2 cache there NOT Memory bandwidth.
L1/L2 is only dependent on the clock speed.

Realistic value on X38/965 chipset will be between 4000-6000M depending on memory setting.

[Zak33]

10000MB/s+ is certainly wrong, you're looking at L1/L2 cache there NOT Memory bandwidth.
L1/L2 is only dependent on the clock speed.

Realistic value on X38/965 chipset will be between 4000-6000M depending on memory setting.

no... no I'm not.. I'm using memtest and thats what it says.. L1 cache is higher than that.. L2 says unknown

[Steve]

I'm not sure how reliable memtest is for gauging bandwidth, especially when it doesn't even know about all your caches.

[Zak33]

I'm not sure how reliable memtest is for gauging bandwidth, especially when it doesn't even know about all your caches.

roger.. but so long as I compare apples with apples, I think it's good to use.

So my version of memtest is prolly reading something different to his.. but so long as I see an increase on THIS rig... using same version... it's "relative" and comparable I think. It is consistent

[Zak33]

new version of Memtest shows it differently. Results edited into previous post

Thanks guys... need showing the road sometimes....