Nehalem 101 part1 - 1366 and X58

[LinusTech]

I am going to be doing a video version of "Nehalem Overclocking 101" for NCIX Tech Tips so this thread has been incredibly valuable.

Does anyone know how 12GB vs. 6GB of memory affects overclocking?

[tiro_uspsss]

this write-up is awesome!

[saaya]

I am going to be doing a video version of "Nehalem Overclocking 101" for NCIX Tech Tips so this thread has been incredibly valuable.

Does anyone know how 12GB vs. 6GB of memory affects overclocking?

your welcome
some things might be wrong though, im having a hard time either dealing with un or misinformed engineers and intel reps here or for some reason the language problem causes me to understand the exact opposite of whats true

it seems 920s and 940s will have mem and uncore multis unlocked now, which doesnt make sense though, why would intel send every mainboard maker cpus that are locked, and then sell retail chips that are unlocked... the other way around it makes sense, but this is really weird...
maybe it was a last minute decision to unlocked the mem multis on the retail chips...

regarding memory configs, well it depends on the amount of chips and the density, plus amount of sticks, plus what chips you actually use etc.
in general id say:
1gb stick samsung =/> 1gb stick micron
2gb stick samsung > 2gb stick micron
3x1gb =/> 3x2gb > 6x1gb >/= 6x2gb > 3x4gb

thanks tiro_uspsss

[saaya]

Does high Vdimm kill Core i7?
Some say yes, some say no...
for those of you who remember amd going IMC some years back, you might remember that the 90nm shrink brought some problems with it... high vdimm could kill or degrade the integrated memory controller. well how did we work around that back then? vcore had to be increased as well to keep the vcore vdimm ratio more or less the same and things were fine. Later when amd moved to K10, the same thing happened again with some imcs dieing at 1.9v ddr2 vdimm if vcore and other related voltages were kept low.
well for amds imc implementation the memory controller was actually powered by vcore, so thats most likely why the vcore vdimm ratio had to be maintained. for nehalem or core i7 the memory controller is powered by vtt...

Intel recommends a max vdimm of 1.65v, which is curiously 1.5x vtt...
Several people reported that running higher than default vtt plus higher than 1.65v vdimm works just fine. how come?
From what i know about manufacturing processes, you have to pick the target voltage you want to work with at some point, and then decide what transistor design to use. some transistors can take high voltages but switch slow and are rather beefy, others are small and can switch much faster but will degrade with higher voltages. which is exactly what people reported with vdimm damaged core i7 cpus. one way to work around this and stress the transistors less is by not grounding them to ground but to some other voltage.

I dont know why or how, but vdimm is definately related to vtt, and the fact that intel recommends a max vdimm of 1.5x vtt is not a coincidence if you ask me... so sticking to this 1.5x vtt rule is what we should do if we want to use high vdimm.

[Fr3ak]

What voltage do you refer to as VTT? Voltage of the integrated memory controller labeled as QPI/Vdimm on some boards?

[saaya]

only asus calls it that and its pretty confusing...
i think they call it that cause it helps clocking up qpi and memory...
but its not the voltage of the memory controller, its the entire uncore supply voltage.
it powers the memory controller AND the L3 cache and i think also a part of the qpi controller since more vtt helps to clock to higher qpi speeds.

[saaya]

most 965 chips dont seem to hit even 5G on ln2...
So far a 940 should be good enough to max out the current chips under ln2 (23x200=4600Mhz)

the gains from Ln2 are rather small for most chips, only around 300 mhz, and lower than -20 temps dont seem to gain much at all.

Most chips dont like higher voltages than around 1.5v and barely scale above that.

[Fr3ak]

Thanks for the explanation. Now a couple things are more clear to me =)

The 2 965 I tried seem to be way worse than some chips here at XS. I can run 215 MHz BCLK on a P6T with a CPU multi of 12, but cannot boot at 4 GHz. No way to hit 4.5 GHz on air with those CPUs. But there are so many new things for pverclocking with Nehalem, so it might need some fine tuning and a lot of trial and error to figure out which combinations work the best.

But nontheless, seeing what Nehalem CPUs can do on just aircooling at a relativly low voltage is very nice. I guess we will even see better results once it hits retail.

[saaya]

hmm whats the max you can hit then?
did you try high multi 133 bclock?
what vcore and vtt are you running?

[Fr3ak]

I can hit 385x MHz, but only 20 mins Prime stable before the PC resets. I tried up to 1.4V Vcore and VTT + 0.2 V max. Trying to find 24/7 settings here. BCLK below 133 MHz seems to be very buggy. 100 x anything does not work at all, 120 MHz works, 110 does not, etc

[G H Z]

Nice thread saaya

I dont know why or how, but vdimm is definately related to vtt, and the fact that intel recommends a max vdimm of 1.5x vtt is not a coincidence if you ask me... so sticking to this 1.5x vtt rule is what we should do if we want to use high vdimm.

That might be a hard rule to follow

[bmg]

Nice thread saaya

That might be a hard rule to follow

I'd be pretty sure it's not a ratio thing, but more like a max difference between voltages. I'm guessing it's something more like Vdimm shouldn't be more than .5v-.6v greater than Vtt, so Vtt should be raised along with Vdimm. I imagine this will happen anyway, as Vtt will normally have to be increased to run higher memory speeds. It might also apply to Vcore, but again most of us will be increasing Vcore a bit anyway.

[G.Foyle]

I can hit 385x MHz, but only 20 mins Prime stable before the PC resets. I tried up to 1.4V Vcore and VTT + 0.2 V max.

It takes much more voltage to get Nehalems to 4 GHz than with Penryns. I could hit 4GHz only with 1.45V on the 965 and 1.475 on the 920. Seems like 920 can be maxed on air, at around 4100 MHz and 1.5 V.

[saaya]

Nice thread saaya
That might be a hard rule to follow

thanks mark!

I'd be pretty sure it's not a ratio thing, but more like a max difference between voltages. I'm guessing it's something more like Vdimm shouldn't be more than .5v-.6v greater than Vtt, so Vtt should be raised along with Vdimm.

but then youd need around 1.8v vtt for 2.3v vdimm, and i know of some people who supposedly ran 2.3v fine with much less than 1.8v vtt...
and youd need 1.5v vtt to run 2v vdimm, and again i know people who run 2v vdimm with less than 1.5v vtt...
as a matter of fact asus boards automatically set vdimm to 2v and vtt to 1.3v as soon as you set the 12x memory divider. and they do this without even showing that they bump up vdimm and vtt, it shows as "auto".
Im pretty sure asus isnt stupid and sets those voltages by default if they are not safe and will kill the cpus after some weeks... im pretty sure they tested this and found that 2v vdimm is stable with 1.3v vtt...

if its an offset then im sure its more than .55v...
asus seems to be pretty sure that 2v vdimm 1.3v vtt is safe
several people reported 1.8v vdimm 1.1v vtt is safe
that points to an offset of rather .7v, or max .7v to be more precise...
and in that case it doesnt make a big difference if we are dealing with an offset or ratio thing between vdimm and vtt at all, unless we look at vdimm voltages way above 2.25v we get almost the same vtt to vdimm ratio for high vdimm...



i guess time will tell

It takes much more voltage to get Nehalems to 4 GHz than with Penryns. I could hit 4GHz only with 1.45V on the 965 and 1.475 on the 920. Seems like 920 can be maxed on air, at around 4100 MHz and 1.5 V.

yeah but olli tried 1.5v right? and he couldnt even get 3.8ghz stable... def sounds like either a bad chip or something else holding you back olli. fire off a pm to kinc and ask him for some settings, maybe he can help

I imagine this will happen anyway, as Vtt will normally have to be increased to run higher memory speeds. It might also apply to Vcore, but again most of us will be increasing Vcore a bit anyway.

i dont think its related to vcore... everbody came to the conclusion that vtt needs to be reaised for high vdimm to work ok, and many are automatically raising vtt if you set high vdimm or high mem ratios, nobody is touching vcore for that it seems... and vcore and the imc are on completely different power planes, there is no connection whatsoever from vdimm to the actual cpu cores, so i dont think vcore plays into this... but we will see

EDIT: i really dont think vcore has to do with this... just think about it, with dynamic vcore thanks to EIST intel would be in a LOT of trouble if low vcore would cause problems with high vdimm... cause in idle the cpu vcore will drop to 0.x and vdimm remains the same...

[Movieman]

I can hit 385x MHz, but only 20 mins Prime stable before the PC resets. I tried up to 1.4V Vcore and VTT + 0.2 V max. Trying to find 24/7 settings here. BCLK below 133 MHz seems to be very buggy. 100 x anything does not work at all, 120 MHz works, 110 does not, etc

Max I could get with the Intel board was 4226 on air(9.812s)
vcore : 1.4825
vdimm:1.62
31x136
SP1M stable at those settings but not much else.

max 24/7 100% load speed of 3733
vcore:1.35
vdimm:1.52
28x133
TRUE w/87cfm fan
Load temps at 60C
Very stable at that speed, had it at 100% load for the last 10 days.

[kiwi]

920:
max 4.5Ghz
~4.5Ghz 3d stable with all 4 cores on air

huh?

[Movieman]

huh?

yea, I want to see that also..

and room temps..maybe his air had a "chill" to it..mine was 72F..
Maybe a bit more..There was a 8 core clover right behind it at 100% load and they toss some heat..

[Fr3ak]

In fact I tested up to 1.7V (by accident, more or less). I have sen 4.2 GHz on air, but being completely stable is a different story.

I am doing a mainboard OC roundup, so I am interested in voltages, max BCLK and 2D/3D performance. No need to mess around with high CPU clocks for now.

[saaya]

those numbers are mostly based on that people told me, not my own results...
yeah i should edit that 4.5g on air thing

60C with true on 3.8 with 1.3v dave?
daym!

[Fr3ak]

Temps are pretty good for me too. Around 60-65C under load with 1.35V set in BIOS using a Noctua NH-C12P (in case Core Temp displays the right temperatures).

[dinos22]

great thread man

i emailed Richard over at bit-tech about his claim to keep uncore within 0.5v of vdimm which will keep the CPU safe

Richard confirmed on email that Francois from Intel told him this about uncore & vdimm relationship during his interview

maybe something you should add in first post

here is what Richard wrote about it

http://www.bit-tech.net/hardware/200...-core-i7-920/3

QPI/DRAM Voltage - 1.35V: This is poorly worded by Asus - it should read uncore or QPI/memory controller voltage so not to confuse it with the actual memory voltage. Increasing this is also necessary as it helps overclock the base frequency as the uncore area overclocks increase in relation to the CPU core overclocks. This voltage is tied to actual DRAM voltage - the two are directly connected on the motherboard. You'll need to increase this to keep the CPU safe.

While Asus and Intel (rightly) scare everyone (read: uneducated) into thinking that 1.65V on the DRAM voltage should be the absolute limit before you reach for the fire-blanket, all that's really needed it to obey this: keep the CPU uncore voltage within 0.5V difference of the DRAM voltage and there's no problem. Over this potential difference and you’ll greatly increase the chance of CPU death, but it certainly won't happen instantly in a big ball of fail fire if you make a mistake.

DRAM Voltage - 1.66V: This is the closest to the 1.65V the Corsair Dominator DIMMs wanted and it's within the 0.5V Uncore difference.

[Fr3ak]

Wasn't it kind of the same back in the AMD days? I don't remember what chipset it was, but A64 if I am not mistaken.

I take it that "uncore voltage" means VTT or QPI voltage? Pretty messed up naming =/ Intel should have published a "Nehalem naming and functions for dummys"-article. Seeing mainboard manufacturers using different wording, I am pretty sure, I am not the only one being a little confused here

But it seems Richard is wrong about one thing:
"BCLK Frequency - 200(MHz): BCLK = Base clock. This affects the QPI and uncore (L3 cache, northbridge) frequency too - some motherboards like the Intel Smackover allow the ratios to be adjusted, but the Asus does not. Here the ratio is 18x for the QPI and 16x for the uncore (northbridge/L3 cache). The uncore frequency must also always be below the QPI, we’re told."

Either he was using a different BIOS or he was testing the Intel Smack Over with a 965, because with anything non-XE, QPI and uncore cannot be changed with any board I have tested so far.

Edit: Might be a BIOS issue. The beta BIOS of the X58-Extreme also alows QPI and uncore clock changes on a 920.

[dinos22]

Edit: Might be a BIOS issue. The beta BIOS of the X58-Extreme also alows QPI and uncore clock changes on a 920.

say what

that IS interesting

[Oliver]

say what

that IS interesting

indeed

[Fr3ak]

Changing QPI and uncore multiplier on a 920 works on the Smack Over and Gigabyte EX58-Extreme =)

On the P6T and Rampage 2 Extreme changing those values is not possible onm non-XE CPUs.

Edit: Uncore multiplier cannot be changed with the Smack Over, only QPI multiplier.

Being able to just change the QPI multiplier might be beneficial for perfoamance, but not for overclocking as you set it to the lowest multiplier anyway to get the max BCLK out of the board.