Do the Sandy Bridge "K" CPUs Turn Stress-Test SW on its head?

[BonzaiDuck]

Here's something somebody might find of interest -- if you have a Z68 mobo and an SB "K" chip.

I won't belabor this too much: The ASUS Z68 BIOS is quite a wonder and enough to make you worry about "wild horses running away." But I've figured out enough to get it to OC with EIST enabled, "mild power saving," "Turbo-mode" so that its idle state is about 1650 Mhz and -- if you just load it up enough -- the "turbo" state soars up to about 4.33 Ghz. The voltage features on the mobo keep it stable without fixing the VCORE at a constant, so under "Auto," the idle VCORE may be around 0.896V, and at the loaded 4.33 Ghz shows to top out at around 1.288V.

I've just done some preliminary stress-testing with PRIME-95's latest version -- the most recent release. Now that Intel has restored Hyper-Threading "HT," PRIME automatically chooses to run -- not four threads for the 2600K processor, but eight.

Under Small-FFTs with eight PRIME "workers," about 79F to 80F room ambient, three cores show temperatures around 62/63C with one core flipping 66/67C. OF course the throttling temperature threshold for the SB i7 cores is about 72+C as I recall. The other day, for maybe an hour of the same test, room ambient 85C and no "power-saving" enabled, that one Core #1 (of 0,1,2,3) hits about 71C.

So I decided to try just a short IntelBurnTest run.

Suddenly, IntelBurnTest seems to be a milk-toast. 78F room-ambient, and the cores are showing temps between the high 40s Celsius and upper-mid-50s.

IBT is the new pussy-cat.

If you want to say I'm a "johnny-come-lately" to this new Sandy Bridge ball-game, I'll say I'm sorry I wasted your time. But between the 2600K chip and these new Z68 boards with some significant tech-innovations like ISRT, somebody should re-examine the stress-test programs to see if they still mean anything.

Also, the latest CPU-Z doesn't report the "Turbo" loaded speed, and only shows the default value. I don't even think it shows the low EIST value -- check again . . . In fact -- I don't even trust the VID reading -- the mobo-maker's AI-Suite is suddenly more trustworthy. Same with CoreTemp: I'm quite sure the temperatures are spot on, the sensors are working correctly. But it reports a VID too high.

With this kind of product from ASUS, it's just too slick and refined to imagine that the AI-Suite monitor is reporting the wrong values. That's a first: It used to be I wouldn't trust the ASUS monitor over the other programs . . .

Oh. I stand corrected. IBT actually pushed the temperature on the hot core #1 to about 61C while I was typing this . . .

 

[MTDEW]

Thermal throttling on SB is in the 90c core temp(Tjunction) range. (i think its actually 98c)
BTW, i personally would never even think of approaching 90c temp range!

You're thinking of the Tcase temp.
The I7-2600k max Tcase temp is 72.6c but that has nothing to do with the core temperature which is what most the software monitors.
Core temperatures and TCase temperatures are measured at two different spots on the CPU and are not comparable.

CPU-Z does correctly report my default,eist and turbo values correctly while testing on my ASRock Z68 Extreme4. (so not sure why yours doesnt)
Of course i also use other programs to verify, like the eXtreme tuner software that comes with my board, HWMonitor etc...etc..

 

[MrTransistorm]

If you want to see how hot your SB can get, try LinX with the latest Linpack. It got my 2600K about 5C hotter than Prime95 small FFT.

 

[BonzaiDuck]

Thermal throttling on SB is in the 90c core temp(Tjunction) range. (i think its actually 98c)
BTW, i personally would never even think of approaching 90c temp range!

No question . . .

You're thinking of the Tcase temp.
The I7-2600k max Tcase temp is 72.6c but that has nothing to do with the core temperature which is what most the software monitors.
Core temperatures and TCase temperatures are measured at two different spots on the CPU and are not comparable.

Yes. TCase. As of Conroe, the spread between TJ-Average and TCAse was about 10C. With Penryn/Wolf/Yorkie -- it looked like the spread had narrowed, but only for my own experiences with the Wolfies E8400 and E8600.

CPU-Z does correctly report my default,eist and turbo values correctly while testing on my ASRock Z68 Extreme4. (so not sure why yours doesnt)
Of course i also use other programs to verify, like the eXtreme tuner software that comes with my board, HWMonitor etc...etc..

My premature assumption from looking at the ASUS monitor: Since it reported one single temperature, it was using the TCASE sensor. In fact, I also observe that it is exactly equal to the highest TJunction temperature at any given stage of load-testing.

 

[Ben90]

With this kind of product from ASUS, it's just too slick and refined to imagine that the AI-Suite monitor is reporting the wrong values. That's a first: It used to be I wouldn't trust the ASUS monitor over the other programs . . .

I wouldn't trust anything that didn't tell me the distance to TJunction directly, or at least what it is set to.

 

[BonzaiDuck]

Thermal throttling on SB is in the 90c core temp(Tjunction) range. (i think its actually 98c)
BTW, i personally would never even think of approaching 90c temp range!

You're thinking of the Tcase temp.
The I7-2600k max Tcase temp is 72.6c but that has nothing to do with the core temperature which is what most the software monitors.
Core temperatures and TCase temperatures are measured at two different spots on the CPU and are not comparable.

CPU-Z does correctly report my default,eist and turbo values correctly while testing on my ASRock Z68 Extreme4. (so not sure why yours doesnt)
Of course i also use other programs to verify, like the eXtreme tuner software that comes with my board, HWMonitor etc...etc..

. . . As do we all, or try to do . . .

Assertions have been made that one must disable the features commonly thought troublesome in prior generations of hardware to OC'ing, like EIST, C1E etc. And that once done, the shareware monitors would work properly. All that I can say -- from observing the original stock BIOS setting results against these softwares -- is that it appears very convincing to me that the ASUS AI Suite monitoring is reporting accurately on those things that need to be monitored.

Also, in the past, you could see the mobo software would report the TCASE, which had always been the terms in which the spec was stated. So even if the TJunction or core temperatures are reported, its important to know the spread between TCASE and SUM(TJ1:TJn)/n where n=4 for the Sandy and 6 for the Phenom II X6 or a Gulftown. That is, I've never seen the throttling spec expressed in any other terms than TCASE, so if we only have monitoring for the cores or an average of the cores, you'd want to know how they would differ.

If we could assume, as with the old Conroe C2D's, that TJ_Average - TCase = 10C, then we could feel good if our core thermometers never broke 79 or 80C. But I'm not sure we can make that assumption three generations of CPU's later. I'll probably have to download HWMonitor for this machine, because I think it reports both TCASE and TJunction temps.

 

[BonzaiDuck]

I wouldn't trust anything that didn't tell me the distance to TJunction directly, or at least what it is set to.

. . And, yeah . . . that's why we're running RealTemp or CoreTemp, because they do. It isn't that we can't trust the ASUS monitor for "CPU," but that we'd like to know which sensor it reads . . . .

 

[BonzaiDuck]

OK . . . the Duckster is back, here . . .

Downleaded HW Monitor -- latest version .. . 1.8 or something.

Here -- the voltages under "CPU VCORE" match those of AI-Suite at EIST-idle and Turbo-load.

The machine is clocked to 4.34 Ghz. Nothing special, but a useful step.

The "INtel Core i7-2600K" temperatures are Core#0=60C; Core#1=67C; Core#2=61C; Core#3=60C. There is a "Package" temperature of 69C. All these reported as "Max" after an hour of PRIME95 small FFT.

So what is "Package?" On the one hand, you might think it were "TCASE." On the other hand, TCASE should be lower than any average of the other core values.

 

[Nemesis 1]

Here's something somebody might find of interest -- if you have a Z68 mobo and an SB "K" chip.

I won't belabor this too much: The ASUS Z68 BIOS is quite a wonder and enough to make you worry about "wild horses running away." But I've figured out enough to get it to OC with EIST enabled, "mild power saving," "Turbo-mode" so that its idle state is about 1650 Mhz and -- if you just load it up enough -- the "turbo" state soars up to about 4.33 Ghz. The voltage features on the mobo keep it stable without fixing the VCORE at a constant, so under "Auto," the idle VCORE may be around 0.896V, and at the loaded 4.33 Ghz shows to top out at around 1.288V.

I've just done some preliminary stress-testing with PRIME-95's latest version -- the most recent release. Now that Intel has restored Hyper-Threading "HT," PRIME automatically chooses to run -- not four threads for the 2600K processor, but eight.

Under Small-FFTs with eight PRIME "workers," about 79F to 80F room ambient, three cores show temperatures around 62/63C with one core flipping 66/67C. OF course the throttling temperature threshold for the SB i7 cores is about 72+C as I recall. The other day, for maybe an hour of the same test, room ambient 85C and no "power-saving" enabled, that one Core #1 (of 0,1,2,3) hits about 71C.

So I decided to try just a short IntelBurnTest run.

Suddenly, IntelBurnTest seems to be a milk-toast. 78F room-ambient, and the cores are showing temps between the high 40s Celsius and upper-mid-50s.

IBT is the new pussy-cat.

If you want to say I'm a "johnny-come-lately" to this new Sandy Bridge ball-game, I'll say I'm sorry I wasted your time. But between the 2600K chip and these new Z68 boards with some significant tech-innovations like ISRT, somebody should re-examine the stress-test programs to see if they still mean anything.

Also, the latest CPU-Z doesn't report the "Turbo" loaded speed, and only shows the default value. I don't even think it shows the low EIST value -- check again . . . In fact -- I don't even trust the VID reading -- the mobo-maker's AI-Suite is suddenly more trustworthy. Same with CoreTemp: I'm quite sure the temperatures are spot on, the sensors are working correctly. But it reports a VID too high.

With this kind of product from ASUS, it's just too slick and refined to imagine that the AI-Suite monitor is reporting the wrong values. That's a first: It used to be I wouldn't trust the ASUS monitor over the other programs . . .

Oh. I stand corrected. IBT actually pushed the temperature on the hot core #1 to about 61C while I was typing this . . .

I have had 2 Asus M/Bs Both died in less than 1 year . The M/B it replaced was the abit IC7-MAX. That had the NB fan go bad . SO i decided to give Asus a try . That board lasted about 3 months. The free replacement Lasted about 9 months . So I took the Passive NB heatsink off of the asus and installed on the Abit max. After Abit closed. I decide to give MSI a try and their boards were OK . Than A while back A friend gave some prerelease parts to help him debug. The one he gave my wife was an MSI z68 G80 B3 with pci-e III(cpu unknown) and so I bought the Z68 G65 for the browser . I can say flat out these 2 boards from MSI are the best and most stable M/Bs I have ever used. If they hold up Abit will be replaced in my heart as the beast of all MOTHER BOARDS.

 

[Dadofamunky]

Amazing how none of these mobo companies ever give us a break and actually document the BIOS settings. That's what happens when you outsource everything to China. That said, With SB you don't want to be impatient. 4.33 without playing with the multiplier settings is darn good. With me also, there is a single "hot" core that is significantly warmer than the other three... that seems to be the pattern with SB. At some point, I might upgrade to a 2600K, but I want to hold out to see if Intel has an 1155 Ivy Bridge processor.

 

[MTDEW]

Ok, its Sat morning and wife is still in bed so here comes my wall of text for the day.

Assertions have been made that one must disable the features commonly thought troublesome in prior generations of hardware to OC'ing, like EIST, C1E etc. And that once done, the shareware monitors would work properly. All that I can say -- from observing the original stock BIOS setting results against these softwares -- is that it appears very convincing to me that the ASUS AI Suite monitoring is reporting accurately on those things that need to be monitored.

Actually i only suggested to disable eist,c1e etc.. in that other thread since the OP was new and seemed to be a bit confused as to why his VID was changing "on the fly" and if he disabled those to run a fixed clockspeed, he could see his VID change along with whatever multipler he set, then he'd understand why and how it changes based on clockspeed.
He obviously wasn't associating his current VID with his current cpu speed and was expecting the VID to be a set fixed value.

Of course I dont have that stuff disabled myself, as i personally use those features.
And as i said before cpu-z works just fine with the ASRock z68 whether those settings are enabled/disabled, doesnt matter.


The only thing ive ever disabled while testing was HT (just out of pure curiosity) just to see what difference it made and it does make quite a difference in voltages and temps.

With HT on i need 1.245v for 4.6ghz and with HT off i only need 1.224v and the difference in temps is that with HT off i get about 10c lower load temps on all cores in Prime small FFTs.
And those differences were consistent in my testing all the way up to 5ghz.
The difference in temps seems is more related to less threads more-so than the voltage differences which completely makes sense.

Of course we buy a 2600k for a reason, so my HT stays on for my daily use and my "real" stability testing.

With this kind of product from ASUS, it's just too slick and refined to imagine that the AI-Suite monitor is reporting the wrong values. That's a first: It used to be I wouldn't trust the ASUS monitor over the other programs . . .

Yeah, im also inclined to believe your Asus software is reporting correctly since my ASRock software numbers are also always inline with realtemp,coretemp, cpu-z,HWmonitor etc ... also.
Which i agree it makes the software actually useful for monitoring and part of my permanent utilities i keep installed.

I still cannot get myself to use the software for overclock testing/tweaking though, i guess im still too old school and prefer to do all my OCing in the bios/UEFI. :biggrin:

Still not sure why CPU-Z reports my correct clockspeeds and voltages in realtime as i go from throttle to stressing on my ASRock z68 and yet it doesnt work on the Asus z68.
It even detects my memory speed / timings and mobo info all correctly also.
Anyway, be assured it does currently work correctly with the ASRock and given the Asus boards popularity, i'm sure the next version will work fine for Asus also.

Amazing how none of these mobo companies ever give us a break and actually document the BIOS settings.

No doubt.
And it would be great if they could be more consistent in what they group and label the settings as also.

Its been this way for so long, i guess we kinda got used to it.

And maybe even some of us have grown to actually like getting a new board and figuring it all out without even realizing it.
It beats watching tv in my free time and certainly "feels" more productive. :biggrin:

But yeah, at the very least it would be nice to have stickies at the top of the official forums documenting the bios settings for each new board as opposed to the current "The official WizBangZing mobo owners thread" that is always 200 pages long and so unorganized you've gotta read 199 pages to find that one piece of info you're looking for.
One dedicated to bios settings and a separate one for sharing/discussion would be nice.

 

[BonzaiDuck]

Ok, its Sat morning and wife is still in bed so here comes my wall of text for the day.
.

Two thumbs up for your "wall of text." All this stuff is very useful.

I have had 2 Asus M/Bs Both died in less than 1 year . The M/B it replaced was the abit IC7-MAX. That had the NB fan go bad . SO i decided to give Asus a try . That board lasted about 3 months. The free replacement Lasted about 9 months . So I took the Passive NB heatsink off of the asus and installed on the Abit max. After Abit closed. I decide to give MSI a try and their boards were OK . Than A while back A friend gave some prerelease parts to help him debug. The one he gave my wife was an MSI z68 G80 B3 with pci-e III(cpu unknown) and so I bought the Z68 G65 for the browser . I can say flat out these 2 boards from MSI are the best and most stable M/Bs I have ever used. If they hold up Abit will be replaced in my heart as the beast of all MOTHER BOARDS.

I've seen mobo manuals from ASUS, Gigabyte, and eVGA -- all very sparse about how those settings would work together. It may be for us to figure out. You can only see the syntactical, usage and diction errors coming through in the ASUS manual as "Chinese-English-as-a-second-language," which is not too daunting, even so. But all the descriptions and explanations have a tone of sales-hype about the feature.

"Technical writing" issues aside, the tables keep turning from one year to the next. For instance, there seems to be a very reliable myth about the luxury-sports-car Jaguar, replacing the old defunct adage about the meaning of "FORD" : "Fix or repair daily." I wouldn't know as to how Jaguar may have improved from when I first heard the rumors, but economist George Stigler has much to say of importance about "firms" and the long-run maximizing goal of "survivability" over "profitability." It's all about expanding customer-base and market-share.

With ASUS, I'd also experienced some disappointments per two motherboards in the past from 2003 to 2007. The 2003 disappointments far outweighed those of 2007, but there was more "contrast" for the more recent experience according to an eVGA mobo using a chipset only one generation newer.

But I see this stuff turn around from year to year. At one time in the "DDR" era, OCZ was in the forefront of some performance RAM kits. Now -- they're just running with the pack. Back in '03, Norton/Symantec's IS package was a terrible slug, and [sorry for the analog here . . . but . . ] it was like an abortion gone bad, leaving parts of itself all over your registry and hard-disk uterus after an uninstall. Too easy for me to continue clinging to my Kaspersky allegiance.

These new P8Z68 boards seem to be a new page turned in ASUS' business. Only time will tell, and I cannot answer in any finality. This was a real gamble for me. I'd usually wait six months and let others among us sort it out and watch the BIOS-revision history.

Also, I'm saying that this looks to be a "quality" improvement for ASUS, even if -- so I'll stick with the thought -- it's still "running with the pack" among Giga and AsRock. I just chose to plop my money down on this part of the green-felt roulette table, no intentions to buy one of the other boards for "comparison," so I have no basis for making any. . . We look to the avaiilable reviews containing the least amount of hype.

[You know, of course, that the mobo-makers are looking over our shoulders as we speak here . .. ]

I still cannot get myself to use the software for overclock testing/tweaking though, i guess im still too old school and prefer to do all my OCing in the bios/UEFI. :biggrin:

Well, it's interesting to look at behavior when you don't have a budget to do the testing like Bit-Tech or Tech-Report. Whatever we would salvage even if our board has disappointments, we're still faced with choices to sell it or accommodate to it.

And that goes for the software. Once you choose to avoid it when your configuration is fairly new, it's easy to continue with that decision. Certainly more likely for the newbie -- "curiosity killed the cat" -- but I lost my sense of caution momentarily and clicked the Extreme" button in AI Suite. It had been well-behaved the first time -- finding some setting at 4.5+ Ghz. Somehow, I clicked that button and the progress bars were racing through the stress-tests. Then -- BSOD, reboot, and the motherboard sorted it all out. I do NOT LIKE the BSOD experiences, always have a goal to avoid them entirely if possible. It's just comforting that the board "sorts it out on its own."

So I may go back and tiptoe back into that part of the software . . . . carefully . . . and later . . .

 

[BonzaiDuck]

I'm posting this after my last -- and apparently ending -- post to this thread.

The topic was raised that the CPU-Z -- possibly CoreTemp and RealTemp software -- wasn't in current enough revision to give good results for a Z68 chipset.

CPU-Z at the CPUID web-site has just been released (June 24, 2011) in revision 1.58.

The download I got -- causing "aberrant" readings on my Z68 Sandy Bridge was 1.55.

Even the 1.58 release states "PRELIMINARY revisons for Sandy Bridge/EP support." So they pretty much say it themselves.

Also, I mentioned to someone that there is an "ASUS ROG" version of CPU-Z, but couldn't find it on the ROG web-site. It is explicitly offered -- HYPED no less! -- on the CPUID web-site. But -- its last revision was January, 2011.

"Over and out, Sergeant Anandi!! 10-4!!"

EDIT: "Come in, Sergeant Anandi!! OVer!"

When you go to the version 1.18 download, there is a 1.18 "ROG" version also posted. I've run it. All the stuff per my i7-2600K, showing correct "Turbo" speed for the over-clock, the voltage, the memory latencies -- all resolved. RE-SOLVED!!