ASUS P8Z77-V Pro and a high voltage bump

[Kastor Krieg]

My previous ASROCK P67 Extreme4 B3 did not survive a thunderstorm and the death of a router (lightning hits telephone cables, cables carry to router, router fries, my computer is pinned as the first LAN line, my mobo fries via in-built LAN card). Thankfully no other hardware casualties

Now, I'm having a bit of a hard time doing a proper OC on my new board (latest BIOS installed). I can get what I had before, easily (4.6Ghz), but only through the OS interface and still, I'm concerned about a few things.

One, I expected to squeeze a bit more out of the CPU with this board (more powerful section, better parts, generally a better OC board and chipset). Theoretically, the TurboV thingie takes me even to 5Ghz in the Extreme settings, but the voltages it wants reach almost 1.5v and temps follow. Nu-uh. The "fast" version goes to a meager 4.3Ghz. Since I play BF3 and ARMA2, I need the most CPU power I can get. I'd be happy with a fully stable, non-ridiculous voltages, @4.8Ghz, really.

However, I apparently can't find the proper options in the BIOS. The CPU multiplier is there, but it's described as a maximum for the autotweaking, not a "proper" multiplier. I can up the BLCK manually, but the rest seems to handle itself in auto mode only? When I OC through the OS, apparently I can make the CPU to just go BLCK 100, x4.6, but despite LLC being turned off, the 1.325 voltage goes as high as 1.45v, while with Medium LLC, a 1.3v setting becomes 1.4v under load (!).

I don't know why the voltages get out of hand like that. HALP?

 

[BonzaiDuck]

My previous ASROCK P67 Extreme4 B3 did not survive a thunderstorm and the death of a router (lightning hits telephone cables, cables carry to router, router fries, my computer is pinned as the first LAN line, my mobo fries via in-built LAN card). Thankfully no other hardware casualties

Now, I'm having a bit of a hard time doing a proper OC on my new board (latest BIOS installed). I can get what I had before, easily (4.6Ghz), but only through the OS interface and still, I'm concerned about a few things.

One, I expected to squeeze a bit more out of the CPU with this board (more powerful section, better parts, generally a better OC board and chipset). Theoretically, the TurboV thingie takes me even to 5Ghz in the Extreme settings, but the voltages it wants reach almost 1.5v and temps follow. Nu-uh. The "fast" version goes to a meager 4.3Ghz. Since I play BF3 and ARMA2, I need the most CPU power I can get. I'd be happy with a fully stable, non-ridiculous voltages, @4.8Ghz, really.

However, I apparently can't find the proper options in the BIOS. The CPU multiplier is there, but it's described as a maximum for the autotweaking, not a "proper" multiplier. I can up the BLCK manually, but the rest seems to handle itself in auto mode only? When I OC through the OS, apparently I can make the CPU to just go BLCK 100, x4.6, but despite LLC being turned off, the 1.325 voltage goes as high as 1.45v, while with Medium LLC, a 1.3v setting becomes 1.4v under load (!).

I don't know why the voltages get out of hand like that. HALP?

It's hard for me to assess "what you know and don't know." I built my Z68 system exactly one year ago, and did the preliminary research for reviews and OC'ing guides (sometimes part of reviews) -- many of which referenced the P67 mobos (ASUS). There didn't seem to be too much difference between the UEFI-BIOS of P67 or Z68.

Hadn't seen too many "screenies" of Z77 BIOS settings, but those shouldn't be much different, either. I DID read some posts recently to suggest that the same three BIOS options are there: VCORE, OFFSET, and "Extra Voltage applied to Turbo."

Unless there's something I've missed, you only overclock these boards for their "Turbo" multiplier -- which you may have mentioned (or seem to have mentioned).

I briefly experimented with the three or so ways for "AI" or "auto" over-clocking. There's the motherboard TPU(?) switch, the turbo-overclocking option in BIOS, and the AI Suite software. Supposedly on the ASUS boards, you can fiddle with all three, but the last approach used is the one which is relevant without doing anything further on the others.

I used the auto-over-clocking feature within BIOS (forget what it's called -- "Turbo-V" or something . . . ) That gave me a prelim baseline settings reference, and of course -- the voltage was excessive (I think it was pushing 1.44V at 4.5Ghz).

I recorded the BIOS reported voltages at stock (default and "default-turbo"), and collected data on how the offset changes as you go higher. This enabled me to hone in on settings for the highest speeds so I could keep the peak voltage at lowest levels while reducing the number of trial "BSODs" as much as possible.

The auto-over-clocking features will also increase the bCLK -- as I recall, it was pushed up to 103. You can keep it and use it -- or reset to 100 and pursue manual over-clocking at that point. Most people here suggest limiting the OC approach to the multiplier and voltage settings.

 

[Kastor Krieg]

Well, I successfully OC'd my previous i3-540 to 4.5 Ghz and this current i5-2500k to 4.6 Ghz fully stable (on AsRock P67 Extreme4). The BIOS was way more straightforward there, without the automatization getting in the way that much.

I went with the AI Suite route for now. LLC is at high, with base voltage @ 1.245v. With BLCK 103 and x45, I'm @ 4635 Mhz, but the voltages spike up to ~1.35v under load. On the previous board, with strictly manual settings, I could have it benchable, but not fully stable @ 4.8 Ghz, 1.368v (from a ~1.35v preset!). The spike here is unbelievable.

 

[BonzaiDuck]

Well, I successfully OC'd my previous i3-540 to 4.5 Ghz and this current i5-2500k to 4.6 Ghz fully stable (on AsRock P67 Extreme4). The BIOS was way more straightforward there, without the automatization getting in the way that much.

I went with the AI Suite route for now. LLC is at high, with base voltage @ 1.245v. With BLCK 103 and x45, I'm @ 4635 Mhz, but the voltages spike up to ~1.35v under load. On the previous board, with strictly manual settings, I could have it benchable, but not fully stable @ 4.8 Ghz, 1.368v (from a ~1.35v preset!). The spike here is unbelievable.

Yeah -- difference in organizing BIOS screens between ASUS and ASRock, and ASrock supposedly a subsidiary of ASUS . . .

There's always been a slight confusion in threads and posts on the voltage topic. There's "stock base-clock" idle voltage (for i7-2600K 3.4Ghz, for instance). Then, there's the full 100% load voltage, and an "idle" voltage when the speed is still at full multiplier setting. And then there's the EIST voltage.

If your "1.35v under load" is the voltage with processor fully loaded, I'd say it's a tad too high. If it's the "idle" full-speed voltage, then it's "OK." But your LLC setting likely affects my perception here -- I'm using the second-lowest LLC setting on my system. It loads (fully) with a drooped voltage somewhere around 1.32v (this is an i7-2600K). When "un-loading," the voltage will spike briefly to 1.36+v. This latter is probably close to what happens when the processor is only at 30% load . . . .

 

[Kastor Krieg]

Here are the screenshots, mate. I'm still trying to figure out what's what on #3, I know most of the stuff, but not all. And it's all different than AsRock, with some of those options inaccessible from BIOS, which confuses me even further.

Unloaded:

Under full load with Prime:

Power settings:

 

[Kenmitch]

Don't use software to overclock just do it in the UEFI.

If your MB has the EPU and TPU switches I think Asus calls them make sure they are in the off position.

You could watch the following video of his overclock with your motherboard. It should get you started in the correct direction. Of course voltage requirements may be diff and I'd try using offset vcore instead.

http://www.youtube.com/watch?v=wnH0luwrcfE

If you need any pointers then just ask.

 

[Kastor Krieg]

Hi, thanks!

I have a problem using UEFI, with how counterintuitive it is. In the software I can set a proper multiplier, not a "maximum Turbo multiplier" (since I don't use Turbo, right?), I can select levels for the power settings visible above...

I'm going to watch the video and see if I can use the knowledge from there. Thanks again.

 

[Kenmitch]

I'm pretty sure a caveman can do it

It's not much diff than the old days....Just less needs to be changed.

 

[Kastor Krieg]

Yeah, but then there's me going "OK, this goes here... This gets tweaked to this... Off, off, on, off... All this on auto. This I can't tweak at all, strange... OH WHERE THE BLOODY HELL ARE YOU GOING WITH THAT VOLTAGE, KIND SIR?!"

 

[BonzaiDuck]

Yeah, but then there's me going "OK, this goes here... This gets tweaked to this... Off, off, on, off... All this on auto. This I can't tweak at all, strange... OH WHERE THE BLOODY HELL ARE YOU GOING WITH THAT VOLTAGE, KIND SIR?!"

There have to be several variations to what I suggest here.

First, here's a "review" of the -V Pro board with the screenies of the BIOS. [I can't see where much has changed from the P8Z68-V Pro . . ]

http://www.anandtech.com/show/5793/...ith-ivy-bridge-asrock-asus-gigabyte-and-msi/5

My memory is rusty on some of the details, although I've got about 20 pages of notes I took last year before I settled on some four different OC settings -- finally leaving one of them as the operative "24/7" combination.

You're best to find several reviews or guides and sort through them . . . take some notes.

Unless something has changed, I was informed by ASUS that they don't provide an ability to change anything other than the "Turbo" multiplier -- which means that (without EIST enabled) the processor always idles at its stock multiplier and speed . . . for the i5-2500K it would be 33 and 3.3 Ghz. I think the Turbo multiplier setting is on the "Advanced" page.

Before messing with anything else, you can select the "OC Tuner" on the AI Tweaker page, and the board will go through its own paces (rebooting, etc.) to find a stable OC setting. This will result in voltages higher than normal, and likely run up the bCLK from 100 to 103. But useful to find a preliminary stable setting -- just for reference.

You can then re-enter the BIOS, set the bCLK to 100 again, make note of some voltages as reported on these screens, and do some tweaking.

The RAM voltage might as well be fixed at the RAM's spec, and you can fix the RAM timings to the spec -- probably slightly tighter than the "XMP" values. Leave VCCSA on "auto." CPU PLL might be left on "auto" or fixed to the value shown in the center column (the reported value). Same with the PCH voltage. Ultimately, one of these must be within 0.5V of the RAM voltage -- my memory is vague, so . . . find some online guides.

Set the VCORE to "Offset mode." The "offset sign" can be either "+" or "-" -- but you need to get a foothold on a stable setting either way.

Also -- I'm pretty sure that "CPU Power Management" submenu contains an "Extra voltage applied to Turbo" or something worded similarly -- which is set initially to "auto." You can gauge what value this "Auto" is providing by watching a voltage monitor while stress-testing and doing simple arithmetic with previously-measured VCORE. Then, the only things you'll want to tweak are the offset value and this third voltage. Keep in mind that many veterans here had overlooked this "Extra . . . turbo" voltage setting -- probably leaving it on "Auto." But you can minimize the stable load voltage by tweaking it to find its minimum, just as with the offset.

After you get through the land-mine of BSODs as you refine the settings -- after you get it running with small-FFT Prime95 or 10+ iterations of Intel Burn Test -- most of your trouble should be over.

Someone else here can give you more refined advice about this, and as I said -- you should be able to find a lot of "guide" material online. There's an ASUS "Republic of Gamers" website which should have some of it, and of course, the many forums and exchanges between members often provides data and insight. . . .

 

[cmdrdredd]

Hi, thanks!

I have a problem using UEFI, with how counterintuitive it is. In the software I can set a proper multiplier, not a "maximum Turbo multiplier" (since I don't use Turbo, right?), I can select levels for the power settings visible above...

I'm going to watch the video and see if I can use the knowledge from there. Thanks again.

Once you start overclocking you are setting the max turbo right there. The CPU will idle down (assuming you have C1e and Speed Step on) and speed up when necessary.

I believe your BIOS should be pretty similar to mine. Here's a thread with some screenshots of my settings. Might point you in the right direction.

http://forums.anandtech.com/showpost.php?p=33642670&postcount=28

 

[BonzaiDuck]

There you go! Commander Dread gives some more insight. Only difference is the Maximus V BIOS, but close enough.

I'm inclined to advise using a lower LLC setting; some will encourage you to use a higher one. There are ramifications to the higher settings -- just to be aware of. You should be able to get a good over-clock setting so that the voltage at full load is maybe 1.32v and the unloaded or partially loaded "turbo" voltage isn't much more than 1.35v.

Or . . . . you can take your chances and push it higher. But the "auto-overclock" features and software are more likely going to push it higher than necessary, no matter what the clock-speed achieved.

 

[cmdrdredd]

I will note that my board has a .056 vdroop with LLC high and .107 droop with LLC off. So either way I use that to find my stable voltage.

Example 1: I need 1.33v to get 4.6Ghz stable So I can do +.13 and LLC high which gives me 1.256 + .13 = 1.386 - .056 droop = 1.33v final (as reported by CPU-z).
Or I can turn LLC off and it goes like this. 1.256 + .19 = 1.446 - .107 = 1.339v final

Example 2: I need 1.27v for 4.5Ghz stable. So I can do .07 and LLC high. 1.256 + .07 = 1.326 - .056 = 1.27v final.
LLC off is like this. 1.256 + .13 = 1.386 - .107 = 1.279v final

1.256 is the stock voltage at stock max turbo frequency.

 

[BonzaiDuck]

I will note that my board has a .056 vdroop with LLC high and .107 droop with LLC off. So either way I use that to find my stable voltage.

Example 1: I need 1.33v to get 4.6Ghz stable So I can do +.13 and LLC high which gives me 1.256 + .13 = 1.386 - .056 droop = 1.33v final (as reported by CPU-z).
Or I can turn LLC off and it goes like this. 1.256 + .19 = 1.446 - .107 = 1.339v final

Example 2: I need 1.27v for 4.5Ghz stable. So I can do .07 and LLC high. 1.256 + .07 = 1.326 - .056 = 1.27v final.
LLC off is like this. 1.256 + .13 = 1.386 - .107 = 1.279v final

1.256 is the stock voltage at stock max turbo frequency.

If I'd figured on determining that last vDroop detail under the LLC extremes last year, my settings would be even better today.

I haven't even thought about Samsung RAM since I built a RAMBUS system in 2003 . . . which was a "dead end." Nice voltage for those, especially if running at the rated speed . . .

 

[cmdrdredd]

If I'd figured on determining that last vDroop detail under the LLC extremes last year, my settings would be even better today.

I haven't even thought about Samsung RAM since I built a RAMBUS system in 2003 . . . which was a "dead end." Nice voltage for those, especially if running at the rated speed . . .

It was a lot of trial and error for me. I ended up setting voltage manually to find what i needed then I figured out what offset gave me that voltage. When running Intel burn test the voltage is reported lower, presumably for the higher load so that's why my sign says 1.26v when under prime testing it is at 1.27v.

 

[BonzaiDuck]

It was a lot of trial and error for me. I ended up setting voltage manually to find what i needed then I figured out what offset gave me that voltage. When running Intel burn test the voltage is reported lower, presumably for the higher load so that's why my sign says 1.26v when under prime testing it is at 1.27v.

Yeah . . . . that was pretty much my observation for the stress-test softwares. . .

I went looking for Samsung DDR3 modules . . . only found DDR3-1600 and DDR3-1333. but the cus-reviews tout over-clocking them to 1866 or 2133. So I'm guessing that's what you did.

I'm pretty much "stuck" with my Z68 board and i7-2600K -- with a lot invested (time and trouble) in configuration. I found a thread in "Memory and Storage" to which you'd contributed, featuring some XBit Labs links and benchmark screenies.

Somewhere else, someone had stated that both IB and SB had caching features that (and I quote) "blunted the hoped-for gains of high-performance RAM" [or . . . nicely over-clocked RAM]. I would need to read through that thread more carefully. Somewhere else, it had been stated that any gains were shown only in synthetic benchmarks . . . which would mean your AIDA-64 screenies.

At this point, I have a spare 2x4GB kit of G.SKILL -GBRL 1600's; found myself looking at 2x8GB DDR3-1600 and DDR3-2133 kits. And the 2x4GB kit I have in the system seems plenty adequate. So I'm contemplating both quantity and speed improvements -- wondering if it's going to matter that much anyway. Those Samsung modules rated at 1.35V (loose timings notwithstanding) look like they have "potential." Of course -- money in my wallet has "great potential."

EDIT -- just as an afterthought to your remarks . . . It really pays to take good notes while going through the OC process. I did that, but wish I'd taken stock of even more detail . . .

 

[cmdrdredd]

Yeah my memory is this one http://www.newegg.com/Product/Produc...82E16820147096

I overclocked them. I have benched them at DDR3-2400 11-13-13-33 CR1 @ 1.575v

When I started overclocking this I didn't start with a blank slate. I looked at what other peopler were getting on the forums and I tried that voltage and frequency. Had to tweak it here or there for my particular setup, but generally it got me in the ballpark.

 

[BonzaiDuck]

Yeah my memory is this one http://www.newegg.com/Product/Produc...82E16820147096

I overclocked them. I have benched them at DDR3-2400 11-13-13-33 CR1 @ 1.575v

When I started overclocking this I didn't start with a blank slate. I looked at what other peopler were getting on the forums and I tried that voltage and frequency. Had to tweak it here or there for my particular setup, but generally it got me in the ballpark.

[My apologies to Kastor Krieg for deflecting from the main topic . . .]

I'm guessing -- with a 1.5V spec requirement for these DDR3-1600 G.SKILLs -- I might push them to 1866 or 2133 with looser timings and a voltage of -- say -- 1.60V. But for me, the question still needs resolution as to "whether" IB does more to fully utilize those speeds, or if there are caching features on both IB and SB which make the performance increases less than proportional to the 33% (or so) increase in nominal speed (for an increase from 1600 to 2133, for instance).

I'd always attempted in earlier builds (i.e., LGA-775 systems) to tighten the latencies even if it meant "less Megahertz." With DDR2, latency timings meant somewhat less than it did with DDR, and I'm guessing the same trend for DDR3 is applicable.

 

[BonzaiDuck]

OK. I see no responses, and I rudely deflected Kastor's thread, 'tho sure we gave him some info of value.

This is my latest "intelligence" about high-performance RAM on either Ivy Bridge or Sandy Bridge:

http://hexus.net/tech/reviews/ram/38613-gskill-tridentx-ddr3-2400-ivy-bridge-memory/

A summary quote from it:

“Genuinely high-end products are released with the lure of exceptional performance. Our benchmarks show an 8GB (2 x 4GB) TridentX pack of DDR3-2,400 memory to be barely any faster, if at all, than well-tuned DDR3-1,600 memory. The explanation for what seems like a performance anomaly is actually quite simple: Intel's recent chips integrate smart, efficient caches that blunt the hoped-for benefits of running premium RAM.”

Some basic facts about the review's testing:

-- Uses an Ivy Bridge CPU

-- Compares DDR3-1600 9-9-9-24-1T with the Trident DDR3-2400's with CMD=2T

So it would seem the test has a bias if you choose to run your RAMs at the standard or "auto" command-rate.

However . . . Since they are comparing these for the Ivy core, and since little difference is shown (only slivers of advantage for the Tridents), it would then seem a logical conclusion that there would certainly be little advantage for a Sandy Bridge. The reason this is important to me: I'm not yet ready to build another system, having built this i7-2600K SB configuration with a Z68 board and G.SKILL RAMs.

So . . . I'm gathering information toward multiple choices . . . .

Sorry to trouble . . . .But one more thing . . .

I'm inclined to think that speed alone is not what one might be after in this pursuit.

Instead, it could well be voltage. Using either my G.SKILLs or Dredd's Samsungs, we might confirm that either set of modules is within warranty or "safe" up to 1.65V. Mine come spec'd to 1.5V; Dredd's are 1.35V (with looser timings). And G.SKILL had informed me "OK to 1.65."

There's a possibility that Dredd can tighten his latencies at least to the G.SKILL's spec, and might have to increase the voltage to do so. He may or may not have more range in tightening his timings -- I wouldn't know. But the fact that I might tighten my own by pushing the VDIMM to 1.60V -- well . . .

 

[cmdrdredd]

a good read http://www.xbitlabs.com/articles/memory/display/ivy-bridge-ddr3.html

 

[BonzaiDuck]

a good read http://www.xbitlabs.com/articles/memory/display/ivy-bridge-ddr3.html

I keep forgetting that latencies don't matter much anymore -- that's the profound error in my train-of-thought in the earlier post.

You can comment if you think you have a better grasp of this. It appears that the article suggests that IB extends the benefits of higher performance RAM, which would otherwise be limited to 2133 speeds or below for SB.

I'm not going to build an IB this year, because my SB system is otherwise stellar. Except . . . . for the fact that I've done very little to max out RAM performance. At most, I'm running my G.SKILLs at their rated speed and minimum voltage. And I'm running them at a CMD-rate of 2 -- the default.

I also contemplate increasing the amount of RAM from 8 to 16GB. If I simply fill all four slots, I can't run them at CMD=1. I can also increase the speed and loosen the timings.

I'm just not going to get the performance potential of IB -- even if the DDR3-3200 modules don't yet exist.

Does that sound like a ball-park appraisal?

 

[cmdrdredd]

Yeah that's pretty close.

Honestly, the difference they find is very small anyway. I generally tell people to save a few bucks if they can help it. No sense buying 2400 memory and knowing that if you took it out and installed 1600 you wouldn't have an appreciable difference.

 

[BonzaiDuck]

Yeah that's pretty close.

Honestly, the difference they find is very small anyway. I generally tell people to save a few bucks if they can help it. No sense buying 2400 memory and knowing that if you took it out and installed 1600 you wouldn't have an appreciable difference.

Well, at Medicare-eligible age, I see my memory is failing (not meant as a pun . . but . . ).

I had visited this issue last fall -- maybe November, and here at the forums. I'd posed a question about these RipJaws DDR3-1600 modules I'm using now, per tuning with CMD=1 for the command-rate.

Certainly, you are familiar with our colleague IDontCare. He offered his own "deposition:" that he had two kits of these or 4x4GB which he'd managed to set at 1T, and he twisted up the frequency to DDR3-1866 and slightly looser timings. Claims he had run HCI Memtest and everything else -- no problem. AND -- he didn't have to turn up the voltage beyond the stock 1.50V setting.

So at that point, I picked up a spare set of the RipJaws for about $50. Then, got sidetracked, and they sat in the bubble-wrap package all this time. I had completely forgotten what IDontCare had told me that made it seem worthwhile to buy them at all.

I've got some questions I put to G.SKILL tech-support about their high-end DDR3-2133 RipJaws 16GB kit, and I'll expect their answer soon-- they've always been stellar about this.

But I may just stick with the sticks I have . . . save the money . . . and probably get the full Sandy Bridge potential with them . . . Half the time I contemplate something like this these days, I procrastinate, because I know I'll want to do thorough stress-testing -- which means I'll have to use a different computer for one day and watch TV somewhere else . . . Now, I'm bristling at a need to run eight instances if HCI Memtest. I'll probably let Prime95 do the job . . . Or maybe just turn off hyper-threading while I do it . . .

As for "appreciable differences" as you say -- That's my biggest worry. That's partly why we're called "enthusiasts," because too many times we enthusiastically buy wonderful s*** hoping for "stunning improvements," then we convince ourselves that "it's still better and probably worth it."

But I can see and feel the improvement with the tighter command rate. If I can bump up the frequency another 266 Mhz even with CAS=10 . . . that's about as good as it can get for the Sandy Bridge -- at least according to the article you linked earlier (and thank you very much). But also, I saw a review showing that these "GBRL" G.SKILLs set at DDR3-1866 9-9-9-24 1T are quite viable with a voltage increase. So . . . "we're gonna find out . . . "