Core i7/X58 Overclocking Thread

Quote:

---

Originally Posted by 003

I am having a problem with this. You are saying that 4 / (9/8) = 1.777. But that is wrong. It would have to be 2 / (9/8) for the resulting quotient of 1.777. 4 / (9/8) is 3.555.

Am I just missing something here or is there an error in your math? Either way please let me know because I am very confused.

---

You are right...my bad. I know what I was thinking, but I wrote it down wrong.

In my testing I showed that Uncore higher than 4.0GHz was unstable, and particularly unstable if it passed QPI maximum of 4.0GHz. Therefore, I assumed an upper Uncore limit of 4.0GHz. Since Uncore is 2xDRAM then max DRAM is 2.0GHz but going by the safe Uncore:QPI relationship that would make max Uncore 3.555GHz which is 1.777GHz memory.

I think with lower overclocks you can run memory higher and still be stable as the overall stress on the IMC and other components is much less.

Quote:

---

Originally Posted by revenant

Coolit Freezone Elite. :) nice self contained unit, don't need to worry about ambient temps bringing your base temp up.. I dunno, some people don't like them but I think it would be a good compliment to this hot quad.

---

Meh im not a fan of tec cooling, for the same money i can build a 600W water cooling loop, where i can cool everything in my system. ;)

Quote:

---

Originally Posted by Xello

Use Prime, man!

---

Ok, what settings? Sorry, this is my first time OCing in like six years!

Quote:

---

Originally Posted by Xello

Doesn't the 920 top out @ 20x multi? I doubt i could get my 23x perma-turbo multi on a 920 :shrug:

Anyone else getting really random vdroop? Setting 1.44v in bios gives me 1.424v in cpu-z under load, setting 1.46v in bios gives the same. Setting 1.475v in bios gives 1.456v real @ idle, 1.44v under load. Still that means i've nailed 185x23 =4.25ghz 10hr prime stable with 1.44v, which is not bad i guess. Now going for 4.3ghz with 23x187.

---

Actually I have no vdroop. The vcore under load is slightly higher, not lower, in my case. Goes from 1.34375 to roughly 1.35. No turbo on.

Quote:

---

Originally Posted by kompulsive

Does anyone know why running four instances of Orthos is only putting my loads at 100%, 100%, 50%, and 50%?

---

Quote:

---

Originally Posted by kompulsive

Ok, what settings? Sorry, this is my first time OCing in like six years!

---

You should run 8 instances of Orthos I guess, since there are 8 logical cores.
Anyway I think the best way to stress these systems is with LinX. You can run two instances of LinX, each on half the available memory, and change the affinity to core 0,2,4,5 for one instance and core 1,3,5,7 for the other. This will evenly distribute the load and really put your core i7 to the test. Temps will be about 5°C higher than with Prime95, and you'll see if your system's stable in about 20 minutes (10 runs of LinX) instead of 20 hours.

So can I assume overclocking the ram isn't even neccessary with Core i7 due to the low FSB needed? (or am I wrong about this because the memory multipliers are different?)

I remember back in the Conroe days to fully OC a E6300 required someone to have a nice set of Micron D9s pushing a FSB of 500+?

So am I safe to assume any old set of DDR3 ram will work for maximum OC?

Or does Memory bandwidth actually affect system performance with Core i7? On Conroe as I remember memory bandwith wasn't really the bottleneck (it was almost purely CPU performance that mattered)

Quote:

---

Originally Posted by Xello

Doesn't the 920 top out @ 20x multi? I doubt i could get my 23x perma-turbo multi on a 920 :shrug:

Anyone else getting really random vdroop? Setting 1.44v in bios gives me 1.424v in cpu-z under load, setting 1.46v in bios gives the same. Setting 1.475v in bios gives 1.456v real @ idle, 1.44v under load. Still that means i've nailed 185x23 =4.25ghz 10hr prime stable with 1.44v, which is not bad i guess. Now going for 4.3ghz with 23x187.

---

Your likely getting same vdroop at those similar settings, you are just seeing the limits of having the bios capable of setting volts in .00625v steps, but sensors read in steps of .016v. So when you increase volts .006, your idle likely increases .006, your load increases .006 but the sensor may read the same (depending on where volts actually are) until you increase vcore enough to get past midway point of next sensor value and thus rounds up to the next highest setting (even though you wont actually be there either).

If you plot out a bunch of OC points and record bios, cpuz idle and cpuz load, you can make an excel sheet in few minutes with bios value, cpuz idle sensor reading ie, vdrop (+/- .08v sensor rounding error), cpuz load sensor reading, ie vdroop (+/-.08v sensor rounding error), then easily plot out actual values for idle and load that cpuz is really reading, by knowing it increases in steps of .00625 and finding the rounding error points. (this is just what sensor is actually reading, does not mean same as multimeter)

I dont know if I explained that very well, actually takes just few minutes with excel, if you keep a tab of bios, cpuz idle and load values. Just helps me with seeing how OC really scales, so I can get to other stable points quickly.

Quote:

---

Originally Posted by kompulsive

Ok, what settings? Sorry, this is my first time OCing in like six years!

---

Aha, well i'm no pro but i can tell you i've found prime to be 100% reliable :) And i hear it's one of the best for quad cores.

You don't really need any settings for prime, it will run the appropriate number of threads automatically. Just choose 'blend' test and let it run for 8-10hr minimum :up: From my experience with IBT on my QX9650 i don't use it, my 24hr prime stable setting was not stable according to IBT. Yet it was crash-free for the entire time i ran my overclock :)

Quote:

---

Originally Posted by Cooperdale

Actually I have no vdroop. The vcore under load is slightly higher, not lower, in my case. Goes from 1.34375 to roughly 1.35. No turbo on.

---

Yeah i think it's this board, it's getting ridiculous, i up the voltage from 1.475v to 1.48v and i'm still getting 1.465v idle, 1.44v load. :(

Quote:

---

Originally Posted by Aryton Senna

Or does Memory bandwidth actually affect system performance with Core i7? On Conroe as I remember memory bandwith wasn't really the bottleneck (it was almost purely CPU performance that mattered)

---

From what i can figure out tri-channel is going to give you more bandwidth than you will ever need. On X48 i was running 2x2gb of corsair 1600mhz 9-9-9-24 with something like 10,000 read score in everest. Even that was more than i needed. With X58 i'm on 3x2gb corsair 1600mhz 9-9-9-24 and the read has went up to 16-17k without upping clocks / tuning timings :eek:

Depends what you use the system for i guess, for games / general use tri-channel is probably overkill but on the server side, just check out that article on Anandtech from yesterday :)

there is no difference between ddr3-1200 and ddr3-1600 in real world scenarios, IE Gaming, superpi 1m, wprime 32m + 1024m pifast

Quote:

---

Originally Posted by sonofander

I concur...

I went and returned my 965 and exchanged it for a 920.
My max stable oc on the 965 was 4.3 and guess what my max stable oc on the 920 is... Give up?

4.3 :D

Guess for me the 965 was just a waste of money since I can do the exact same thing with my 920. With the exact same volts by the way.
All I know is 4.3 for $320 is WAY better than 4.3 for $1150. Seriously

And my ram, qpi and uncore speeds are almost identical to what they were on my 965 4.3 overclock.
So lost 200 pts on everest read, write and copy but I put $700 back in my pocket...
no brainer, huh?

Anyway, it's funny how context is so important. I was really dissapointed in a 4.3 clock on the 965 but I'm thrilled with the same clock on the 920

---

And you put in context too that you can't seem to use memory @ a frequency much then 1600, because after that you'll have to sacrifice CPU OC in order to get more mem speed. And because with high qpi speed and uncore speed 920 don't seem to get stable, you most likely won't hit ddr2000.
But for what you said, you are indeed correct.

Quote:

---

Originally Posted by Hornet331

Meh im not a fan of tec cooling, for the same money i can build a 600W water cooling loop, where i can cool everything in my system. ;)

---

yeah, true... you're limited to the cpu only with the freezone. and it's not cheap.

Quote:

---

Originally Posted by D3mon_Hunt3r

And you put in context too that you can't seem to use memory @ a frequency much then 1600, because after that you'll have to sacrifice CPU OC in order to get more mem speed. And because with high qpi speed and uncore speed 920 don't seem to get stable, you most likely won't hit ddr2000.
But for what you said, you are indeed correct.

---

I cant imagine having a problem running 2000mhz ram (although its pointless) 200bclk x10 ram multi gets you there. Am I missing something?:yepp:

Quote:

---

Originally Posted by Bobbylite

I cant imagine having a problem running 2000mhz ram (although its pointless) 200bclk x10 ram multi gets you there. Am I missing something?:yepp:

---

This is an exerpt from Dejanh's post.

Quote:

---

1. Why is it that I cannot run my 2000MHz memory in my new Core i7 system?

To understand this note the maximum QPI link speed we mentioned earlier. Using the simple ratio we defined, we can see that with the maximum QPI link speed of 4.0GHz we can have a theoretical maximum Uncore of 4.0GHz and a maximum memory speed of 2.0GHz. These however are only theoretical maximums and are by no means guaranteed. For starters, QPI and Uncore cannot be run at the same speed as any clock oscillations can result in a crash if the ratio of QPI to Uncore ever falls below 1:1. Therefore, Uncore must always be below QPI to avoid this, and preferably below the 8:9 Uncore to QPI ratio to guarantee stability. Consequently, if we cannot achieve 4.0GHz Uncore, this means that we absolutely cannot achieve 2.0GHz memory and we can therefore not guarantee any stability for 2.0GHz memory. Technically, the maximum DRAM speed for a stable system is equal to ((Max. Uncore) / (Safe QPI:Uncore Ratio)) / 2 or (4.0GHz / 9:8) / 2 = 1.777GHz. In between 1.777GHz DRAM and 2.0GHz DRAM you are likely to experience some level of instability over the long term. Past 2.0GHz you are just plain lucky.

---

I dont think that should be a problem. Couple guys have uncore higher than qpi

Quote:

---

Originally Posted by Bobbylite

I dont think that should be a problem. Couple guys have uncore higher than qpi

---

user dependent ;)

congrats on the 100th post brutha.

Even if I couldn't run ddr3 2000, would it matter? All tests have shown minimal real world perforamance gain with memory ocing on i7.

FWIW I wasn't trying to run 2000 even when I had the 965. ~1700 @ 8-8-8 is more than good enough for me

Well, I can say that so far I've been fairly bios-dependent as to whether I can run 200bclk or not. F4k seems to work well and passes 8-12 hours of Prime95 while F4n doesn't even post. F4l was fine but the bios corrupted on my so I tried F4k.

I'm running the following multis:
Bclk - 200
CPU - 16
RAM - 8
Uncore - 16
QPI - 18 (38 in bios)

This is stable, while when I put the Uncore up to 18 it won't post.

I still have yet to tighten my RAM timings up to their rated spec, but that will come later.

So there is some merit to dejanh's theory in my case. Whether this is true for all boards/bios remains to be seen.

Quote:

---

Originally Posted by Bobbylite

I dont think that should be a problem. Couple guys have uncore higher than qpi

---

I just identified in my testing that running Uncore higher than QPI is not stable over long term. Also, running Uncore so high (4GHz+) is also not stable. Puts too much stress on components.

Some of these things change of course when you are sub-zero and again, I do not have access to every mobo and processors for cherry picking so there is always a chance I missed something. Seems to be a good functional guideline for a lot of users out there though :)

I wish I didn't have to, but I pay for my own testing equipment so there is only so much I can afford :p:

Quote:

---

Originally Posted by dejanh

I just identified in my testing that running Uncore higher than QPI is not stable over long term. Also, running Uncore so high (4GHz+) is also not stable. Puts too much stress on components.

Some of these things change of course when you are sub-zero.

---

I took your advice and I was finally able to run stable 200 x 21.

One question. that whole 8:9 ratio, which in actuality is .888333 instead of .888999 does that also need to apply when Bclk is under 200?

Reason is that I have to run mem at 6x and uncore at 15x to get 200x21 stable.

If I lower Bclk to 199, then I can run mem at 8x and uncore at 16x.

I've run prime 95 for a while and it seems stable, though, I know I need to let it run for a longer period of time, but normally when one of my configs is unstable, it crashes within 5 minutes.

I'm not specificaly pointing to dejanh post but more specifically talking about issues with high Uncore using 920's.

Quote:

---

Originally Posted by Lukee

I took your advice and I was finally able to run stable 200 x 21.

One question. that whole 8:9 ratio, which in actuality is .888333 instead of .888999 does that also need to apply when Bclk is under 200?

Reason is that I have to run mem at 6x and uncore at 15x to get 200x21 stable.

If I lower Bclk to 199, then I can run mem at 8x and uncore at 16x.

I've run prime 95 for a while and it seems stable, though, I know I need to let it run for a longer period of time, but normally when one of my configs is unstable, it crashes within 5 minutes.

---

I'd go Linpack or LinX if I were you. 10 minutes and you know if you're stable or not. And it's much more thorough than Prime as far as temps go.

So after reading this I should'nt wait on 2000MHz RAM but just be satisfied with 1600MHz?

How can you calcutate how mutch you need to push your CPU in order to get it stable and run your mem @ 2000MHz?

I love the LinX load. It definately heats the processor up more than anything else, and crashes sooner too.

For those that prefer Prime95, why are you not just getting the latest version? It will find and run 8 instance in 1 window. http://www.mersenne.org/freesoft/ Step 3.

Here is what I have so far. I am sure that I can tweak a bit more out of both the processor and memory, but I am pretty comfortable with this. I also had troubles with x20 CPU multi, but then I was running the RAM slower also.

http://i364.photobucket.com/albums/o.../i7OC38Ghz.jpg

Quote:

---

Originally Posted by Lukee

I took your advice and I was finally able to run stable 200 x 21.

One question. that whole 8:9 ratio, which in actuality is .888333 instead of .888999 does that also need to apply when Bclk is under 200?

Reason is that I have to run mem at 6x and uncore at 15x to get 200x21 stable.

If I lower Bclk to 199, then I can run mem at 8x and uncore at 16x.

I've run prime 95 for a while and it seems stable, though, I know I need to let it run for a longer period of time, but normally when one of my configs is unstable, it crashes within 5 minutes.

---

1. That is 21MHz less (case in point, I can run 23x179MHz all day long at 100% load on all 8 threads but no matter what I do I cannot make 23x180MHz stable on air cooling).
2. 199MHz BCLK with DRAM at 8x and uncore at 16x is still in 8:9 ratio to your QPI which would be 199x18 at that point in time (3184MHz Uncore / 3582MHz QPI == 8:9). The ratio is still maintained.

One thing to note is that these chips seem to have very fine stress points and certain combinations appear to be very demanding on the system. Particularly, when using BCLK of 200MHz+ the components appear to be really stressed so I found that staying below those "threshold" ratios usually results in a much more stable system. When BCLK is below 200MHz I found I have a lot more flexibility with DRAM, Uncore, and QPI. I do always try to at the very least maintain the ratio I specified, even when below 200MHz BCLK.

Strange thing happened today.

I was running BOINC and F@H GPU to do some further testing and I came home to find my rig shut down.

I checked my RealTemp log and nothing strange there. Temps never went over 80 and the system had previously passed Prime95 with these settings for 8+ hours. I guess it ran for about 3-4 hours then shut down. I'm going to try it with just BOINC to see if the same thing happens. If not then I'll try the F@H GPU.

I'm running the following multis:
Bclk - 200
CPU - 16
RAM - 8
Uncore - 16
QPI - 18 (36 in bios)

Everything else is auto.

CPU Volts are set at stock (1.26 in bios 1.2 idle 1.18 Load)
QPI/PLL is 1.355 bios
CPU PLL is 1.84v in bios
VDDR is 1.64 in bios

Everything else is auto.

You might wanna make sure that vista didn't put your rig to sleep. It seems on my system since I upgraded to i7, vista has this nasty habit of changing my settings to go to sleep after 30 min. I orignally thought my pc was shutting down till I noticed that.

Something to check on...