Overclock.net › Forums › Case Mods & Cases › Builds & Case Mods › Build Logs › Three IvyBridge I7 3770 / I7 3770 K builds, from 4.3 GHz up to 5 GHz: Asus P8Z77-V LK, Asus Sabertooth Z77, I7 3770, I7 3770-K, Asus GTX 670 SLI
New Posts  All Forums:Forum Nav:

Three IvyBridge I7 3770 / I7 3770 K builds, from 4.3 GHz up to 5 GHz: Asus P8Z77-V LK, Asus Sabertooth Z77, I7 3770, I7 3770-K, Asus GTX 670 SLI

post #1 of 2
Thread Starter 
Three IvyBridge I7 3770 / I7 3770 K builds, from 4.3 GHz up to 5 GHz:
Asus P8Z77-V LK, Asus Sabertooth Z77, I7 3770, I7 3770-K, Asus GTX 670 SLI

Greetings !

Over the last 12 months, I put together some five new machines for both private fun and professional stuff, replacing older machines ...of interest here may be the latest build of 3 Ivy Bridge machines as they went through various levels of overclocking.

Two of the latest three are I7 3770 non-Ks, and one is a I7 3770K. All of them have been clocked up to various degrees (up to and beyond 5GHz in the case of the 'K') and I thought I share some of the settings and results, apart from a question or two I have...

Part 1: 2x I7 - 3770 (non-K) on Asus P8Z77- VLK - up to 4.560 GHz

Part 2: I7 3770K in Asus P8Z77-V LK to 4.833 GHz

Part 3: I7 3770K in an Asus Sabertooth Z77 at 4.5 GHz to 5 GHz

Part 4: more pics, screenshots and benchmarks

...I used to do a lot of overclocking ‘back in the day’, but for the last few years, not so much...things sure have changed ! BCLK and straight CPU voltage tuning is one thing, but I had no experience with speed-step and PLL load line currents and calibrations when I started this set of builds in late September 2012. Overclock.net and a few other sites really helped out with their info (thanks !)

A special thanks thumb.gif to the superb guide http://www.overclock.net/t/1247413/ivy-bridge-overclocking-guide-with-ln2-guide-at-the-end

The two non-K machines are now part of a Virtual Machine pool (thus the non-K version as the I7 3770 K does not support virtulization) that are used as a ‘LocalHost’ testing and configuring a bigger commercial setup that runs on Xeons etc.

All three machines of the latest build have 32 GB of G.Skill Ram (between DDR3 1600 and DDR3 2400), and run either Windows Server 2008 / 64 and/or Windows 7 / 64 pro or ultimate (some are dual-booted). The Ram size does impact a bit on CPU voltage requirements.

All machines (including Sabertooth Z77 / 3770K) run Microsoft ISS and various versions and instances of SQL server from Sql 2005 to 2012, including SQL 2008 Enterprise with large databases and some custom code. About 10 gigs of Ram per machine are reserved for various instances of SQL. I mentioned it because even when fully overclocked and stress-testing, these things are running and become part of the testing regime, ie. addressing ‘LocalHost’ to serve up database pages while Prime95 etc are running.

Starting Configurations for this build of three IvyBrige machines:

Their basic configuration per each of two machines:

Asus P8Z77 V-LK motherboards
Intel i7 3770 (3.4 / 3.9 GHz, 77 Watts, Costa Rica)
Antec 302 Case, minimum of 6x 120 mm silent fans (Noctua and In Win and Antec)
32 GB of G.Skill Ripjaws DDR3-1600 (PC3 12800) 1.5V
either 3 or 4 x Western Digital Caviar Black 1 TB (and also one with 2 TB drives)
Asus 24x DRW 24 B1ST
CoolerMaster EX2 625W PSU

Shared bits

Samsung LED 27 1080p and Viewsonic LCD 22 1080p
Asus RT-N65U Router (w 2x USB 3.0), Asus 8 port VIP GB switch, Linksys 8-port GB switch

Update 1 Configuration

As above, but with 7x 120 mm fans and 2x 80 mm fans per machine
Thermaltake Water 2.0 Extreme closed-loop system w/240 mm externally mounted radiator
32 GB of G.Skill TridentX DDR3 2400 MHz Ram (1.65v)
Asus GTX 670 Direct CU ii 2GB GDDR5 PCIe 3.0

Update 2 Configuration

As above, but with Intel I7 3770-K processor (3.5 GHz / 3.9 GHz, Costa Rica, 77 watts)

3rd discreet IvyBridge build (with Update 2 reverting back to original configuration for first two machines)

Asus Sabertooth Z77 motherboard
Intel i7 3770-K (3.5 GHz / 3.9 GHz, Costa Rica, 77 watts)
Antec 302 Case, 7x 120 mm silent fans (Noctua and In Win and Antec)
32 GB of G.Skill Trident X DDR3-2400 1.65v
2x (SLI) Asus GTX 670 Direct CU ii 2GB GDDR5 PCIe 3.0
3 or 4 x Western Digital Caviar Black 1 TB (one switched over to above)
Asus 24x DRW 24 B1ST
LG GH 24 NS95 DRW (not mounted yet)
Corsair TX850 V2 PSU
Logitech HD 1080 webcam

Part 1: 2x I7 - 3770(non-K) on Asus P8Z77- VLK - up to 4.560 GHz

Throughout all the builds when tuning them, I use the Bios as the base step, 'XMP', Turbo and 'multi-core enhancement' is turned on with the multipliers set to 41+, about 2 steps or more BELOW where I eventually want to go (using the Windows-based Asus AI Suite 2.01.01 to do v-core, PLL and Digi adjustments, in addition to Intel's excellent 'Extreme Tuning Utility' program though you may want to avoid using both at he same time, speaking from experience).

Both of these builds easily overclocked up to 4.2 GHz [now 4.3] with no 'non-stock voltage settings' and on modified Intel stock coolers but with Artic MX 2 thermal compound applied and some extra Noctua 120 mm fans. Once stress testing is complete, I switch the stable settings derived from Asus AI or Intel extreme tuning utility to the Bios

One of the aforementioned Noctua 120 mm fan in the side door blowing onto the motherboard and reaching a good part of the CPU cooler, which itself has another Noctua 120 mounted on top. Further, I have a whole drawer full of older and surprisingly quiet 80 mm fans, and I used two of those to blow onto the memory and CPU cooler from a side angle before I tried to go beyond 4.3 GHz via BCLK tuning. After adding these, 'up to' 4.3 GHz was possible, the 'up to' referring to just how many cores on a non-K 3770 are actually running at that speed

I have seen more two or more out of four physical cores at a 43 multiplier during stress testing, all 4 at 41, presumably related to Asus' multi-core enhancement. I mention this 'caveat' because in my builds, an earlier version of Asus AI suite only allowed up to a 41 multiplier for a locked 3770 non-K, but a Bios update later, 43x was possible, including what appeared for all cores. That is before BCLK tuning to gain more speed and also test the overall resiliency of components.

That said, I was now reaching temps during stress testing which were getting up there (around 88 C on the cores)...after all, this was the stock Intel air cooler, even with Artic MX2 thermal paste and extra fans assisting which helped a bit.

Of course, I could have just bought a bigger, better CPU cooler...but I knew that this system would soon be clocked back down for Virtual Machine duties....and also was planning to go for a 3770K as the third machine.

'max' voltage, temp limits and stress testing - some basic rules I try to follow:

With the multipliers maxed out, it is time to get busy on the BCLK frequencies, and also set some 'limits' beforehand, so that one does not go too crazy in the 'heat of the moment' (pardon the pun).

By that I mean 'how crazy' the voltages and temps would be allowed to get for a 24hr/7/365 'real usable' setting, ensuring that 'sleep state' and 'speed step' all remained fully functional etc. and absolutely 'everything worked' re software and hardware. 'Trouble-free every day usage' is most important to me, especially with the caveats I mentioned above re huge database programs w/huge databases running on the same systems.

For normal 'idling' and light web browsing, vCore readings have to be under 1.16v. Load vCore should not exceed 1.37v (lower is better), and 'spikes' (ie those recorded in CPUID Hardware Monitor) should stay below 1.385v.

Temperatures - the Achilles heel of Ivy Bridge CPUs, are obviously directly related to vCore, as well as your specific setup (ie. CPU cooler; water vs air etc). My 'personal' max allowable CPU temp limit per Asus monitor would be 70 C, and per physical core 85 C or less in Real Temp out of a TjMax of 105 C. The ambient temp is usually around 22 C for my tests.

Incidentally, for basic 'first tries' at a new higher speed setting, I recommend Cinebench 11.5. If Cinebench crashes on the CPU test, don't even bother with long Prime95 runs. I also use Aida64 extensively (benchmark and stress testing) when first testing a new speed / voltage setting.

But for me at least, the true test is Intel's 'BurnTest v2', re temps my systems reached in just over 2 minutes, it was more brutal than several hours on Prime95. Don't use the Intel BurnTest unless you have plenty of experience with tracking temps in your systems when overclocking.

For graphic tests, in addition to Cinebench, I use 3DMark11, 3D Vantage, Alien vs Predators benchmark, and Unigine / Heaven DX11 benchmark (some pics of results below). Also referring to GPU overclocking, 3DMark Vantage is usually the one which exposes too high a setting (CPU or GPU) first, perhaps along with Unigine / Heaven DX11 with all settings on 'max'.

I am glad to say that I broke my own rules just once or twice, and largely by accident....i.e. when having good safe voltage settings and temps and then kicking the multiplier up to 50 per the build description of the third machine (Sabertooth Z77 w/3770k at 5 GHz).

That got me into the overclock.net "5 GHz club" http://www.overclock.net/t/678487/5ghz-overclock-club at 1.384v, before reaching higher speeds http://valid.canardpc.com/show_oc.php?id=2647107 biggrin.gif . But back to the earlier iterations:

Pushing a 3770 non-K to 4.560 GHz

Trying different bus speeds showed that the G.Skill 32 GB DDR3 12800 would go up to about 1866Mhz (from its rated 1600) without a problem or extra voltage above 1.5V.

Then I really got bitten ‘by the OC bug’...adding a Thermaltake Water 2.0 Extreme’ kit and a 32 GB set of G.Skill Trident X DDR3 2400.
The new G.Skill memory worked great, though the Asus P8Z77 V-LK Bios would only allow for a 2200Mhz setting and balk at the 2400 setting...even after updating the Bios to the latest version...not a problem though as the 3770 non-K versions need to use bus over-clocking anyways for the final stretch due to their multiplier limitations, so I rather start BELOW the rated speed.

Intel(R) IPP JPEG encoder 7.0.1041 - Jul 19 2011;

The Thermaltake Water 2.0 Extreme kit was a different matter...its 240 mm extra-thick radiator barely fit physically in the Antec 302 case which I happen to love (especially at $59 a piece), but that would wreck the otherwise carefully designed airflow of the rest of the system. Besides, I rather not have a radiator inside the case but outside.

Intel(R) IPP JPEG encoder 7.0.1041 - Jul 19 2011;

Fortunately, the Antec 302 case has two holes with grommets for water-cooling hoses on the top rear anyway...I just cut out the space between the two holes and enlarged the resulting opening just a bit to fit the of this closed-loop water-block assembly through and also gave some extra plastic sleeves around the rubber hoses of the cooler for extra protection (pic below).

I mounted the radiator vertically, with the two included fans in 'pull' configuration at the back of the radiator...and the 120 mm Antec 'exhaust fan' from the case blows into the externally mounted radiator as well from the other side. Eagle-eyed folks will note that the pictures show a slightly twisted arrangement. For now, it has to be this way re. access to video connectors at the back.

Intel(R) IPP JPEG encoder 7.0.1041 - Jul 19 2011;

Once that ‘operation’ was done, I disabled CPU spread spectrum in the Bios, which I find is a 'must' with BCLK tuning. Clock speeds were now to 4.5 GHz (42x 1.071) which meant that I had to slow down the Trident X DDR 3 (native 2400 Mhz) further in the Bios before the BCLK boost of just over 7% ...temperatures were well under control now, and once I joined ‘Steam’ and 'Origin’ for a few flying and racing games, it was clear that the onboard iGPU, while quite good in its own right (IvyBridge have that new Intel iGPU HD 4000), was the limiting factor for fluid graphics w/max settings at 1080p.

...in went an Asus Direct CU ii GTX 670 (powerful and quiet even at load). Wow. Not only does it overclock nicely and allows for 'max setting' play with great frame rates, it also dropped the CPU voltage requirement, buying a bit more headroom. To wit: At 3.9 GHz 'standard Turbo' mode w/stock voltages and iGPU enabled, the CPU cores, when idling, will consume around 20 Watts...At 4.8 GHz, but with discreet video / GPU(s), it will consume around 12 Watts at idle.

Per pics below, I got the 3770 non-K on the Asus P8Z77 V-LK to a perfectly stable 4.560 GHz (43 x 106)., with vCore at 1.25v and EVERYTHING else stock (ie PLL voltage / calibration stock).

Clocking BCLK up usually has system-wide performance gains, and with the particular setup I was running then, 107.5 BCLK was the max before some random hangs would occur.

Part 2: I7 3770K in Asus P8Z77-V LK to 4.833 GHz

I ran the 3770 non-K @ the above overclock for just under two months. The 3770 non-K builds were pretty close to maxed out by now, but it was a great exercise for me to learn about Ivy Bridge voltages and cooling. The second 3770 non-K had to sooner or later join the first (that one still on air / modified stock CPU cooler now at 4.1 GHz) and other earlier new machines in the Virtual Machine build I am still expanding...

At the same time, I was quite amazed how fast a basic Ivy Bridge machine for around $1300 w/4 terabytes of hard storage (before taxes/ after rebates) was doing...the Asus P8Z77-V LK was a real surprise...right now for example I have seen it on special for about $125, and it does have some nice features ( 3PCI-E16 2PCI-E1 3PCI SATA3 SLI DVI HDMI DP USB3.0). It could use more Sata3 ports and may be a few more USB 3.0, yet I have plenty of USB 3 open, not least as the router I use also has dual USB 3.0s.

The P8Z77-V LK just has a 'basic' VRM phase design, yet it certainly never posed a problem in the builds I describe here. And that includes the next step to 4.7 GHz to 4.833 GHz...

Before deciding whether to get another Asus P8Z77-V LK or another type of board, I picked up a non-locked 'K' I7 3770 chip (like the other two 3770 non Ks, made in Costa Rica) and dropped it into the existing setup I just described...knowing how well-cooled, quiet and stable it had been with the other CPU as I had built that system up.

It took no more than 10 minutes to make the switch - and I immediacy noticed something during the first Bios 'post' with the new chip: The CPU temp was only 1 C above the motherboard temp (usually it is 2 C or 3 C), and the indicated voltage being used swung from the upper 0.7v to about 0.81v at its 3.5 GHz base setting on the first page of the Bios...now it was too early to draw some major conclusions, but nevertheless, this was encouraging ...

Boot-up into Windows was no problem, and Windows correctly recognized the new CPU and started to install some additional things (?) compared to the non-K that had been in there prior, then asking for a reboot. I 'caught' it during rebooting and got into the Bios and set the multiplier for 43, with multi core enhancement, Turbo and XMP on (still at 2200 MHz for the 32 GB Trident X 2400 MHz). Other than CPU spread spectrum still being 'off', all other settings were left at stock.

All stress test and benchmarking went without a hitch - if anything, I could have lowered vCore a bit more - but instead decided to do the only reasonable thing in that situation and clock up to 4.5 GHz...that too was uneventful during stress testing and benchmarking...and so was 4.6 GHz (all changes via multiplier only at that point, so 46x 100).

Then came 4.7 GHz, and now I needed to give it a bit more CPU vCore. I adjusted vCore from 1.250v to 1.260v, with PLL voltage and calibration still at 'stock'. That worked great, with temps still well under control, even w/Intel BurnTest v2 and prime95 long runs (cores at less than 82 C max).

Then I tried 4.8 Ghz via 48x 100...after bumping vCore to 1.280v, Intel BurnTest v2 and long Prime95 runs worked great with core temps still below 85 C max).

In the meantime, I had done some more reading at Overclock.net, Hardocp and also Anandtech and it became apparent that there is a trade-off between a slightly lower CPU vCore and increasing PLL load line calibration from its stock 'regular' to 'high', 'ultra high' or even 'extreme'.

I bumped PLL load line calibration to its 2nd setting (high) out of four and lowered vCore to 1.275...success ! The highest stable setting I ever tried with the Asus P8Z77-V LK with a 3770K and 32 GB of Ram was 4.833 GHz at 1.285 v per pic below, though with such high multipliers, BCLK tuning becomes a lot more tricky as it is much easier to overshoot a good setting due to less 'fine granular' control, also re memory speeds.

Part 3: I7 3770K in an Asus Sabertooth at 4.8 GHz to 5 GHz

Intel(R) IPP JPEG encoder 7.0.1041 - Jul 19 2011;

As the Asus P8Z77-V LK was 'needed elsewhere', and waiting for it was one of the 3770 non-Ks which already 'knew each other intimately' @ 4.56 GHz. But what to replace it with as the new home for the 3770 K ?

Also, I had never tried out an SLI setup, and since I have an older HD TV I will soon replace, I made the decision to look at two Asus Direct CU ii GTX 670s, requiring a strong board...since I spent at lot of time building up a very well-cooled system (I cannot stress the importance of overall temperature control enough after having built systems that ran commercially and continuously for almost ten years 24/7/365), I decided on the Asus Sabertooth Z77 drum.gif

The Sabertooth Z77 was conveniently available for $229 when I was looking, while the 2nd video card was still at the same $389 price I paid for the first one after rebate.

I opted for the Sabertooth as it has:

a.) a well-rated sturdy VRM assembly ESD guards, setup for Thunderbolt and 4x Sata3 6G

b.) a lot of attention to overall temp management, including two little 'assist fans' for the VRM and PCH assemblies and Thermal Radar

c.) an enhanced Asus AU suite with advanced setting for very fine voltage controls and such. Besides, it looks quite nice and certainly unique.

As to the Sabertooth Z77 / 3770 K / SLI build, I added a Corsair TX850 V2 PSU, and that build was mostly 'painless' - mostly, but not entriely... The closed-loop Thermaltake 2.0 Extreme water cooling arrangement obviously has very strong performance credentials per above, but it gets quite tricky re mounting the back 'plate' which is sort of a rubbery plastic that holds 4 soft-metal sleeves into which the water block screws mount from the front.

To begin with, it also uses several 'sticky pads' on the back and prying those loose without hurting the P8Z77 V LK board they were attached to was anything but easy, especially after multiple heat cycles...furthermore, I earlier had fortunately run across several forum entries that warned of the soft metal retainer sleeves (bolts?) stripping very easily, especially after a re-mount.

After some anxious moments, I got the Thermaltake closed-loop system off the old board and mounted securely onto the Sabertooth.

I originally booted up with only one GTX 670 so that I could uninstall the video drivers, VirtuMVP (great tech, but does not work with SLI) and also the existing Asus AI suite for the V-LK board.

To clock up, I used the same Bios approach on the Sabertooth as I did on the -V LK: multiplier to 43, spread spectrum off, XMP on, Gen3 PCI enabled and everything else stock (ie voltages).

I worked with vCore to 1.27v, and PLL voltage was kept at a stock 1.8 and PLL load line at the 2nd of our settings ('high') and tried 4.5 GHz, 4.6 GHz, 4.7 GHz and 4.8 Ghz at that voltage and per above testing regime, everything worked fine - at even lower temps than the -V LK. I also noticed (presumably because of the thermal armour of the Sabertooth and perhaps an extra fan compared to the prior setup described above) that the existing hard drives were all running 2 C to 3 C cooler).

Then I got a bit 'uppity' and tried 4.9 GHz and then 5 GHz (screen shots below and per link above)... and it seemed fine with that - other than CPU vCore voltages were now 'spiking' (per above) to 1.39 but that good news was that the actual load voltage stayed at 1.232v / 1.368v during the tests I ran, per pics, while idle would swing around 1.15v and sleep functions remained fully functional.

My latest settings resulted in the same 'load' voltages of between 1.232v and 1.360 (depending on the test), but 'spikes' are not above 1.385v now, unless I actually go beyond an even 5 GHz (50x 100) via extra BCLK. The system is stable and idles at about 1.16v and has no problems with 'sleep' states while temps even in max stress testing stays below 85 C on the cores.

I do believe that the fairly stable regular (and similar !) 'load' voltages at 4.8 GHz to 5 GHz while varying vCore little suggest that there is some sort of sweet spot I am in. In the end though, it seems to me that when I ask a system to overclock by 43%, it might just want up to 43% more 'juice' and I am not sure by what margin one can avoid that, even though 'heat' can clearly be controlled.

Other Current tests

With 32 GB of very well performing Ram, I have some spare capacity, even with the various SQLs.

Now, in a way it is all RamDrive anyways - the systems rarely uses the swap file because I tend to keep it running (thus idle an low vCore requirements) - only after a cold boot is there a lot of hard disk and some swapping activirty.

Still, I wanted to try out a RamDrive so I downloaded the trial of SuperSpeed's Ramdisk Pro - wow ! I created a 4 GB Ramdisk and dropped big folders with large vid files in...the speed is unbelievable and I have included some screenshots of drive benchmarks below.

Next steps and 'Dreaming in Technicolor' and more questions on liquid and Peltier cooling

I plan to convert the current Sabertooth Z77 / 3770K to a custom water-cooled setup, but one with a 'Peltier–Seebeck effect' twist:

I am not referring to a Peltier-style cooler between the water bloc and CPU or GPUs, but a Peltier cooler mounted on the outside to the radiator...I realize that I don't want to end up freezing the cooling medium (as Peltier coolers can dip to -25 or so), but there must be a way to slow the 'Peltier–Seebeck effect' down a bit (less juice?) and bring the overall cooling medium system temp down to may be 7 C or so...

...which also raises another question: I have seen a lot of folks use pretty amazing if not mesmerizing colours in custom water cooled systems, but is anyone actually using automotive antifreeze (which has multiple advantages, including lubrication of water pumps and also running below 0 C)?

Here are some pics and screen shots / benchmarks, stretching over several weeks and months and an 'iterative process' to zero in on the final setup:

Edited by Joa3d43 - 1/11/13 at 3:27pm
post #2 of 2
Thank you for this thread! I have the i7-3770k, but I use the Intel DZ77GA-70K motherboard and I'm running at 4.534GHz on the Intel liquid cooling setup and 32GB of Crucial Ballistix DDR3. I'm not a gamer, so I just run a GTX-650 and I've got the usual SSD, hard drives, etc... Thank you for the great information, I really appreciate the fact that you're sharing your knowledge with us!!
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Build Logs
Overclock.net › Forums › Case Mods & Cases › Builds & Case Mods › Build Logs › Three IvyBridge I7 3770 / I7 3770 K builds, from 4.3 GHz up to 5 GHz: Asus P8Z77-V LK, Asus Sabertooth Z77, I7 3770, I7 3770-K, Asus GTX 670 SLI