how to achieve matching core and uncore speeds on i7 965

[dfn]

I saw that you can run an i7 965 at 3.2GHz core speed while having the uncore speed also at 3.2GHz. I have been unable to get this result. I have an i7 965, 3x2GB DDR3 1600 RAM, and an Asus P6T X58.

If I load my RAM's XMP profile, and leave everything on auto, my core speed is a little above 3.4GHz, and my uncore is a little above 3.2GHz. If I load the XMP profile, and change all the settings manually, except the cpu ratio, again the core is above 3.4GHz, and the uncore is above 3.2GHz. Finally, if I load the XMP profile, and set everything manually (ratio=24, QPI, uclk, DRAM speed), I get a core speed of a little UNDER 3.2GHz (~3.18) and an uncore speed of roughly 3.2GHz. When the XMP profile is loaded, the DRAM voltage is 1.66, and I tried changing it down to 1.64, with no change in my results.

What am I doing wrong exactly? I'd like to get a core speed of 3.2 and an uncore speed of 3.2. When I leave the cpu ratio on Auto, the multiplier (as displayed in CPU-Z) fluctuates between 25 and 26. When I set the cpu ratio to 24, it pretty much stays at 24, except for occasionally going down to 23.5.

I'd appreciate any help. Either getting both to run at 3.2GHz, or being able to manually set the cpu ratio to 24 and have it not go below that.
TIA

 

[Idontcare]

Originally posted by: dfn
set everything manually (ratio=24, QPI, uclk, DRAM speed), I get a core speed of a little UNDER 3.2GHz (~3.18) and an uncore speed of roughly 3.2GHz.

What am I doing wrong exactly? I'd like to get a core speed of 3.2 and an uncore speed of 3.2.

The core's aren't hitting 3.2GHz exactly because the system is likely using spread-spectrum.

You aren't likely to ever get the uncore and the core frequencies to match exactly, the precision to do this is absent in your rig (absent because it adds cost to add this capability and it is not viewed as necessary for performance purposes).

Most people would consider a core clock of 3.18GHz and an uncore clock of 3.2GHz as "matching".

For the purposes of your objective, does it make a difference if the clockspeed differential is 0.6%? (3.18 vs 3.2)

 

[VirtualLarry]

What point am I missing here. If both clocks are derived from the BCLK, and a multiplier for both core and uncore, then if you set the multis the same, you should end up with the same freq for both components, no?

Regardless of the spread-spectrum BS. That just dithers the clock signals slightly, it does NOT alter their actual frequency.

 

[Idontcare]

Originally posted by: VirtualLarry
What point am I missing here. If both clocks are derived from the BCLK, and a multiplier for both core and uncore, then if you set the multis the same, you should end up with the same freq for both components, no?

Yep, you are absolutely right. No doubt the software being used to report the uncore frequency is not updating the calculated value with the same algorithm that is being used to update the reporting of the observed core frequency value.

When he sees 3.18GHz core, for his specific example in the OP, no doubt the uncore is also operating at 3.18GHz.

Originally posted by: VirtualLarry
Regardless of the spread-spectrum BS. That just dithers the clock signals slightly, it does NOT alter their actual frequency.

How is "dithers the clock signals slightly" not a change in "their actual frequency"?

Spread-spectrum does change the frequency, there's actually nothing else for spread-spectrum to change except the frequency.

Unless I missed the boat on that one? Can you elaborate further on your post if you are convinced my perception is in error? (I'd be most appreciative of having my perception of spread-spectrum corrected if it is in fact wrong)

 

[dfn]

First, thanks for the replies!

Originally posted by: Idontcare
For the purposes of your objective, does it make a difference if the clockspeed differential is 0.6%? (3.18 vs 3.2)

You are right, it really doesn't make a difference. However, my concern is that I see the processor behaving differently if it is at a ratio of 24. When I set it to 24, it never goes above that. When I leave it on Auto, it starts out at 25+ and goes *up*. In the event I leave it on 24, it will only go *down*. If it acted the same with a ratio of 24, and an Auto ratio, then I would not care.

So in summary, if multis are the same, and I end up with same speed for core and uncore, the core ratio goes down, but never up. When they are not the same, the core ratio goes up, never down (from 25).

 

[Idontcare]

When you say the core multi "goes up" are you referring to dynamic clocking?

(all i7's support the core clock multiplier increase by 2 ticks if the PCU determines the chip is operating within its thermal budget)

 

[aigomorla]

Originally posted by: Idontcare
When you say the core multi "goes up" are you referring to dynamic clocking?

(all i7's support the core clock multiplier increase by 2 ticks if the PCU determines the chip is operating within its thermal budget)

yeah but crapsus unless your on the R2E skipped turbo on.

why do you want your unclock to match your overclock?

 

[Idontcare]

Originally posted by: aigomorla

Originally posted by: Idontcare
When you say the core multi "goes up" are you referring to dynamic clocking?

(all i7's support the core clock multiplier increase by 2 ticks if the PCU determines the chip is operating within its thermal budget)

yeah but crapsus unless your on the R2E skipped turbo on.

why do you want your unclock to match your overclock?

Aigo you been drinking your LC water again?

 

[aigomorla]

Originally posted by: Idontcare

Originally posted by: aigomorla

Originally posted by: Idontcare
When you say the core multi "goes up" are you referring to dynamic clocking?

(all i7's support the core clock multiplier increase by 2 ticks if the PCU determines the chip is operating within its thermal budget)

yeah but crapsus unless your on the R2E skipped turbo on.

why do you want your unclock to match your overclock?

Aigo you been drinking your LC water again?

no... but in real world honestly why would you need your bandwith that high when most likely he wont use it?

It only adds a lot more stress to your board and system.

right now im at 2800ish in uniclock

What im wondering is why he wants a 3200 unclock, or NB Ram clock. Bleh whatever you want to call it.

OH okey he's running stock.. and he cant get stock values... Blah... yeah i must be drinking my coolant...

OP, when you overclock the chip, your not gonna care about matching unclock with your cpu clock. As you take your cpu higher to like 3.8 ranges, you'll see what i mean.

And please tell me you bought the 965 to overclock.
(your gonna make a lot of people sad if you say no).

 

[Idontcare]

Originally posted by: aigomorla

Originally posted by: Idontcare

Originally posted by: aigomorla

Originally posted by: Idontcare
When you say the core multi "goes up" are you referring to dynamic clocking?

(all i7's support the core clock multiplier increase by 2 ticks if the PCU determines the chip is operating within its thermal budget)

yeah but crapsus unless your on the R2E skipped turbo on.

why do you want your unclock to match your overclock?

Aigo you been drinking your LC water again?

no... but in real world honestly why would you need your bandwith that high when most likely he wont use it?

Yeah the OP didn't elaborate why they wanted uncore = core for clockspeed, but I've lusted after getting an i7 965 to do this very test as well purely for the academic reasons (really am curious to see how things scale as uncore scales, etc).

My coolant joke wasn't so much regarding whether you understood the OP (I never question that) but rather was me just joking with you over the fact the english/grammar in your post made the post non-understandable (to me at least). :laugh:

I mean c'mon, are you channeling Nemesis today or what? Seriously, help me understand what this sentence means: (or have I seriously fallen behind on the technical lingo you kids are using these days? )

Originally posted by: aigomorla
yeah but crapsus unless your on the R2E skipped turbo on.

 

[dfn]

Dynamic clocking - I think so. I am using latest version of CPU-Z to monitor the clock speed. That is where I observe the changes.

So if they support increase of 2 ticks, how come it *increases* when the multiplier is on Auto (CPU-Z shows fluctuations from 25 to 26.5), and not when I set it to 24 (where it shows fluctuations from 23.5 to 24)?
Higher temp at 24 than 25??

Perhaps "matching" isn't the right word to use - I think that misplaced the emphasis. I want my ram to run at 1600 Mhz. For that the "uncore" must be at 3.2GHz. At the same time, I do not wish to OC the core, so I want it running at 3.2GHz.
Now, when I did that manually (load XMP profile, change ratio to 24), it never went above 24, though it is my impression it should (in at least some cases), since if I leave it on Auto, it goes above 26 (even though Auto sets it to 25 to start).

Am I making more sense, or less sense?