Core i7/X58 Overclocking Thread

[T_Flight]

Who said the Tjmax was 100C?

[rge]

Realtemp is reading the MSR containing tjmax value on core i7's, so no more guessing tjmax since the value is software readable, tjmax on that one is 100C, distance to tjmax plus temp in C. 100 on all ones I have seen so far.

On another note, Intel white paper quote on running your cpu up to TCC max temp:
"With a properly designed and characterized thermal solution, it is anticipated that the TCC would only be activated for very short periods of time when running the most power intensive applications.... An under-designed thermal solution that is not able to prevent excessive activation of the TCC in the anticipated ambient environment may cause a noticeable performance loss, and in some cases may result in a TC that exceeds the specified maximum temperature and may affect the long-term reliability of the processor."

[kompulsive]

Here's my OC so far. I'm on air. I can't stay stable in Prime at 195FSB so I'm trying to figure out where to go from here to get the last 200MHz so I can hit 4GHz.

[fornowagain]

Personally I'd not want to run in the 90's for long, mine just bsod near a 100C on the stock cooler. But permanent damage if you stay under the TCC, throttling is not even active yet?

http://download.intel.com/design/pro...hts/320834.pdf

The Adaptive Thermal Monitor feature provides an enhanced method for controlling the
processor temperature when the processor silicon exceeds the Thermal Control Circuit
(TCC) activation temperature. Adaptive Thermal Monitor uses TCC activation to reduce
processor power via a combination of methods. The first method (Frequency/VID
control, similar to Thermal Monitor 2 (TM2) in previous generation processors) involves
the processor reducing its operating frequency (via the core ratio multiplier) and input
voltage (via the VID signals). This combination of lower frequency and VID results in a
reduction of the processor power consumption. The second method (clock modulation,
known as Thermal Monitor 1 (TM1) in previous generation processors) reduces power
consumption by modulating (starting and stopping) the internal processor core clocks.
The processor intelligently selects the appropriate TCC method to use on a dynamic
basis. BIOS is not required to select a specific method (as with previous-generation
processors supporting TM1 or TM2). The temperature at which Adaptive Thermal
Monitor activates the Thermal Control Circuit is factory calibrated and is not user
configurable. Snooping and interrupt processing are performed in the normal manner
while the TCC is active.

When the TCC activation temperature is reached, the processor will initiate TM2 in
attempt to reduce its temperature. If TM2 is unable to reduce the processor
temperature then TM1 will be also be activated. TM1 and TM2 will work together (clocks
will be modulated at the lowest frequency ratio) to reduce power dissipation and
temperature.

If TM1 and TM2 have both been active for greater than 20 ms and the processor
temperature has not dropped below the TCC activation point, then the Critical
Temperature Flag in the IA32_THERM_STATUS MSR will be set. This flag is an indicator
of a catastrophic thermal solution failure and that the processor cannot reduce its
temperature. Unless immediate action is taken to resolve the failure, the processor will
probably reach the Thermtrip temperature (see Section 6.2.3 Thermtrip Signal) within
a short time. To prevent possible permanent silicon damage, Intel recommends
removing power from the processor within ½ second of the Critical Temperature Flag
being set

Regardless of whether or not Adaptive Thermal Monitor is enabled, in the event of a
catastrophic cooling failure, the processor will automatically shut down when the silicon
has reached an elevated temperature (refer to the THERMTRIP# definition in
Table 5-1). THERMTRIP# activation is independent of processor activity. The
temperature at which THERMTRIP# asserts is not user configurable and is not software
visible.

TjMax can be read from each Core i7 processor but when Core Temp 0.99.3 was originally released, I don't know if Intel had released that information yet. If you want to double check you can try using RealTemp 2.83 which I know reads that new information. Don't copy the new .exe into your old directory or it will use your previous TjMax values from your INI file.

From early testing it seems that the Core i7 has more real world temperature head room available. The previous Core 2 Duo based Quads tended to lose Prime stability when well overclocked at about 70C. I've already seen screen shots of Core i7 running Prime stable at 80C and 90C like Pt1t is running in his screen shot above.

If your Core 2 Duo was running Prime stable, temperature was never an issue and it appears for Core i7, it is even less of an issue. By all means, the cooler the better but if Intel didn't think these could run reliably at 100C then they would lower the TjMax and start thermal throttling 10C or 20C sooner to prevent this.

My only concern from the screen shots so far is excessive voltage. We won't know how much is too much until more retail processors start getting used and the "my processor degraded" reports start coming in.

Edit: Thanks to coolaler for showing us what Core i7 thermal throttling looks like.



http://forum.coolaler.com/showpost.p...8&postcount=12

RealTemp will log any thermal throttling episodes and it looks like core0 at 98C and core1 at 100C both hit the throttle. You definitely want to keep it below this level or else your multi will start to cycle down to 12.0 which will kill performance. Core2 and core3 both hit 93C which is still OK. No throttling has occurred on these two cores.

Good news is that it didn't crash.

[ReverendMaynard]

Off the picture of his temps. Go back and read his reply. Yep it's pretty speciffic when he's showing RealTemp at 93C.

Also, the BIOS is default set to 90C. It shuts down at that temp. Again, that's pretty speciffic info that has most definitely been released. It's right in the BIOS.

Never looked at his temps...couldn't care. I'm asking you a simple question; where are you getting the info on ANY temperature related degradation for Intel i7's? Intel didn't tell you, so who did? Is it an educated guess? If you tell someone that their 93C is causing degradation then you must know when it happens. How do you know it's not 97C?

The bios shut off limit has not a bloody thing to do with degradation Every hear of the term "buffer"?

[Classic Satch]

I finally found my 24/7 configuration. Thirteen hours of Prime95 is good enough for me to determine stability. Now to try to see how far this thing can actually go. However, it doesn't like the 20x multiplier and the QPI threshold is holding it back. Come on Asus, let me drop that darn QPI multiplier already!

[kompulsive]

I finally found my 24/7 configuration. Thirteen hours of Prime95 is good enough for me to determine stability. Now to try to see how far this thing can actually go. However, it doesn't like the 20x multiplier and the QPI threshold is holding it back. Come on Asus, let me drop that darn QPI multiplier already!

Lemme get the BIOS settings! Pleaseeeeee?

[Cooperdale]

Personally I'd not want to run in the 90's for long, mine just bsod near a 100C on the stock cooler. But permanent damage if you stay under the TCC, throttling is not even active yet?

http://download.intel.com/design/pro...hts/320834.pdf

I totally agree, I don't think we should panic so much about temps. I think staying around 80 after hours on Prime95 is just plain safe. What's bad about it is, the cpu is consuming tons of energy. But come on, it's 20 below the Tjmax. Above 90 is bad, of course, but I don't think it's degrading, at least not for short periods of time.

Realtemp is reading the MSR containing tjmax value on core i7's, so no more guessing tjmax since the value is software readable, tjmax on that one is 100C, distance to tjmax plus temp in C. 100 on all ones I have seen so far.

On another note, Intel white paper quote on running your cpu up to TCC max temp:
"With a properly designed and characterized thermal solution, it is anticipated that the TCC would only be activated for very short periods of time when running the most power intensive applications.... An under-designed thermal solution that is not able to prevent excessive activation of the TCC in the anticipated ambient environment may cause a noticeable performance loss, and in some cases may result in a TC that exceeds the specified maximum temperature and may affect the long-term reliability of the processor."

Great. And it's great Tjmax is read via software now, didnt know it! Thanks!

I finally found my 24/7 configuration. Thirteen hours of Prime95 is good enough for me to determine stability. Now to try to see how far this thing can actually go. However, it doesn't like the 20x multiplier and the QPI threshold is holding it back. Come on Asus, let me drop that darn QPI multiplier already!

You're on water, right?

[kazukun]

http://blue.ap.teacup.com/kazukun/img/1229638228.jpg
i7 965　3837B209　WaterCooled
Core Speed：4200Mhz
BCLK：200MHz
Multiplier：21
QPI：3600MHz
QPIV：BIOS1.65V
Vcore：BIOS1.55V
LLC：ON
MEM:KHX16000D3T1K3/3GX
Vdimm：BIOS1.65V
Uncore：4000MHz
DRAMFrequency：2000MHz
DRAMTiming：9-9-9-27 1T
HT：ON
http://blue.ap.teacup.com/kazukun/

[ReverendMaynard]

http://blue.ap.teacup.com/kazukun/img/1229638228.jpg
i7 965　3837B209　WaterCooled
Core Speed：4200Mhz
BCLK：200MHz
Multiplier：21
QPI：3600MHz
QPIV：BIOS1.65V
Vcore：BIOS1.55V
LLC：ON
MEM:KHX16000D3T1K3/3GX
Vdimm：BIOS1.65V
Uncore：4000MHz
DRAMFrequency：2000MHz
DRAMTiming：9-9-9-27 1T
HT：ON

Nice rig you have there.

[unclewebb]

The value that RealTemp reads from Core i7 processors is called IA32_TEMPERATURE_TARGET and here's how Intel defines that:

Temperature Target.
the minimum temperature at which PROCHOT# will be asserted. The value is degree C.

Intel has said that TJMax is not a fixed value. It varies a little from one CPU to the next but they haven't confirmed how much variation is in that number. It's a number they aim for when calibrating but the actual TJMax may be higher.

My personal theory is that it can also vary from one core to the next in a CPU. That's very obvious in the 45nm Core 2 series but I think it's also true in Core i7. There's a reason why in the vast majority of Core i7 screen shots, core0 is the hottest and core3 is the coolest. If TJMax = 100C for core0 then it's likely to be closer to 105C for the average core3. If this was all just random chance then you wouldn't see the exact same pattern with almost everyone's reported temperatures.

Thermal shut down of Core 2 doesn't happen until about 125C and it's likely that Core i7 is the same. I can't imagine that Intel would set such high numbers for TJMax and thermal shut down if they thought people were going to destroy their processors. They wouldn't offer a 3 year warranty and they'd obviously lower these values if they didn't believe in the high temperature durability of their CPUs.

[Xello]

Well said... I think these things are a lot more durable than some people give them credit for, i'm sure when penryn launched Anandtech had at least a full page detailing how much more resilient chips are becoming to death (that's even in light of their high VTT warning which is what gave a few people quite a scare, i wish they'd think twice before publishing stuff like that)

With the temps being reported by most people as acceptable and so much higher than any penryns were run at, i'd say i7 is even stronger.

Of course, in a few months if we're all shelling out for new chips i'll need a hand wiping all this egg off me face.

[Splave]

Im happy sub mid 80's 24/7, depending on the voltage I will need to get me there

[Classic Satch]

You're on water, right?

Yep. I'm a bit lazy and bought a Koolance Exos-2 LX but I did pair it up with a new Apogee GTZ water block. Considering that my CPU is the only thing in the loop, it's working pretty well so far.

Lemme get the BIOS settings! Pleaseeeeee?

I'll post back when I get off of work.

[theoryzero]

And in all honesty, I feel a bit cheated too on the i7 940. I mean come on, every single i7 920 out there is performing as good if not better than both the i7 940 and the i7 965 XE.

ATM, I feel like I donated $350 to Intel.

Been there, felt the same way after buying the E6700. E6600 were just as good for about $300 USD less.

[Hornet331]

god Ci7 is a boinc monster...

I can do ~33% more seti client WUs at the same clock then my QX9650 and still have 4 threads to spare for WCG.

I just love that thing.

[Xello]

Been there, felt the same way after buying the E6700. E6600 were just as good for about $300 USD less.

Hmm yeah, alot of Q9450 / Q9550 would go as high as most QX9650 without extreme cooling, this is nothing new.

Damnit, i think i've toasted my raid 0 array

I have a balloon popup from intel matrix storage console 'a drive in a RAID 0 volume is failing. Try to backup data immediately'.

I've never had this before, is the disk actually unusable now or can i format and remake the volume from scratch and it'll be like new??

What made it happen was running 165bclk with 24x QPI and 24x uncore (i was trying to find max stable of both) which were 3960mhz each. This is pretty ridiculous as from what i know they're supposed to top out at 4ghz, but i never expected them to just fry one of my disks like that.

[Hornet331]

lol you have not fried your disk... you just killed your raid0, happens quite often when you oc.

Thats why i usually use either raid 10 or 5.
If you use suicide raid (aka raid 0) be prepared to have a backup ready.

[dejanh]

Been there, felt the same way after buying the E6700. E6600 were just as good for about $300 USD less.

Yeah well, I guess maybe it was karma for giving up on AMD after 12 years with them All in all, I was too cheap to pay for bragging rights of a 965 XE and I stupidly assumed that same thing which applies in AMD world applied in Intel world so I figured the 920 would be low-binned and a 940 would be high-binned (boy did this backfire). This is a lesson for all times, at least when using Intel. Either go low, or go top of the line just for the bragging rights. It's all the same thing anyway, just packaged differently.

That and I broke one of my cardinal rules...never adopt a new tech in the beginning...but I just could not resist the i7...

[Xello]

Yeah i do, this is a gaming pc so i only have music, media, etc. But it's all backed up on a storage disk, along with the installation files for all my programs, so restorations are never hard.

Isn't it weird to happen just from testing 4ghz QPI though? Maybe it was the uncore... This blows because i wanted to start at 4ghz and work my way down until i found my max stable clocks there (based on high multi and lower bclk), i guess i can't do that

I'm on the stock intel raid 0 option anyway, i reckon i'll try to figure out how to use the included hardware raid, J micron i think it is.

That and I broke one of my cardinal rules...never adopt a new tech in the beginning...but I just could not resist the i7...

I'd still be on my blackops if it wasn't for the disaster that was ATI Catalyst

:edit: well the deed is done, up and running on the hardware raid, quite a bit simpler actually than the software option. Oh, and it cuts out that 10-15 second period during post when it's detecting stuff / waking up... Jumps straight from splash screen to DMI table now, sweet!

Going to run HDtune i think just to compare.

[safan80]

http://blue.ap.teacup.com/kazukun/img/1229638228.jpg
i7 965　3837B209　WaterCooled
Core Speed：4200Mhz
BCLK：200MHz
Multiplier：21
QPI：3600MHz
QPIV：BIOS1.65V
Vcore：BIOS1.55V
LLC：ON
MEM:KHX16000D3T1K3/3GX
Vdimm：BIOS1.65V
Uncore：4000MHz
DRAMFrequency：2000MHz
DRAMTiming：9-9-9-27 1T
HT：ON

very nice work you got the mem up to 2000Mhz! any chance you can list all you voltage settings?

[speckled10]

Ok looks like I've finally cracked it for a stable (15mins prime) setup & good temps ..

Here are the settings:

Multi: 200x20
SpeedStep/Turbo = OFF
DRAM = 1600Mhz 8-8-8-24
Uncore = 3208Mhz
QPI = 7218MT/s
CPU Volts = 1.4V
LLC = Enabled
C-State = On/Auto

Rest is ASUS Default

4Ghz 1600Mhz Prime 8K HT On


4Ghz 1600Mhz Prime 1024K HT On


4Ghz 1600Mhz Cinebench HT On


4Ghz 1600Mhz Prime HT Off


I refitted the Noctua today with horizontal mount (fans blowing up out top of case) + smooth cover of paste over whole CPU + thin invisible layer rubbed into HSF side ... brought my temps down 10-15C at load and sub-30's at Idle as you can see ..

Finally happy with the build !!

[kazukun]

very nice work you got the mem up to 2000Mhz! any chance you can list all you voltage settings?

Bios voltage settings

[SteveRo]

Trying to get 4ghz prime stable with HT off, with lowest vcore and with qpi, cpu pll and dram v all within intel max spec -

[rge]

Trying to get 4ghz prime stable with HT off, with lowest vcore and with qpi, cpu pll and dram v all within intel max spec -

Nice job...have you tried lowering the multi with higher Bclk for same mhz to see if requires less cpu vcore, just curious as it seems to work with i920 and i940 but have not seen any i965's owners state such yet. My excel sheet of my OC's, big difference at 4.1ghz stock multi of 22 versus dropping the multi to 21 at same mhz, or enabling turbo with 23 multi which is another way to lower vcore. All in bold are 12+hrs prime stable (most had linpack run at same time as prime for first 1-2 hrs to try and crash OC early vs late if unstable)...did not record all failures, just ones lasted few hrs. All are 36 (18) QPI, 18 UC for lower bclk runs, 17 UC for higher bclk runs, 1.28-1.33 QPI/Vtt, and 8 mem divider...but I am using 3x1gb so dont need as high qpi/vtt as those with more mem.