Core i5-750 system - moderate overclocking at 3.4GHz - cool and low power PC

[TyrAntilles]

[Because of the limit of 10 images on this site I can post only the links. I posted my setup on my blog too, with all images included: http://tyrantilles.blogspot.com/ I believe is a lot easier to read this way.]


Core i5-750 system - moderate overclocking at 3.4GHz - cool and low power PC (last update 2010.04.28 - I will update this page if I will change anything)

This guide is for the ones who like to bring i5-750 closer to it's "true potential" without any risk for their components on the long therm. I am writing this based only on my personal experience with my components. After a lot of searching trial and testing, I managed to SAFELY overclock my system to 3.4GHz, keeping very low voltages and temperatures. This is a system designed to run on hot ambient up to 40ºC.

Why 3.4GHz? Because:
 3.4GHz over 2.66GHz default is a decent overclock of roughly 30% (27.82%) and it is more then enough for all my needs. I will see no benefits to run a 4GHz rocket, my games would not run much more faster and my work won't be done essentially quicker.
 I want to be completely on the safe side, I don't like fried CPU's.
 Keeping low voltages means less heat and less stress on components that leads to longer life; I intend to keep my computer a very long time.
 Power consumption is a LOT lower with all power saving features enabled. At 4 GHz you need to disable them.

The noise level is very low from the three 12cm fans that came with Lian-Li case, and the CPU fan don’t even counts. The fan from video card is audible only in 3D applications and the loudest components of the system are the three HDD’s. Overall the system is fairly quiet and under the desk where I keep it is totally not disturbing.


- HARDWARE:
o MB Gigabyte P55-UD3
o CPU Intel Core i5-750
o CPU Cooler Scythe NINJA MINI rev B (SCMNJ-1100)
o Therm paste Arctic Silver Ceramique (2.5 g)
o RAM Kingston DDR3 4 GB 1600 MHz KHX1600C8D3K2/4GX (CL8)
o SVGA Sapphire Radeon HD 4670
o HDD Western Digital WD1001FALS
o HDD Western Digital WD1001FALS
o HDD Western Digital WD1001FALS
The three WD1001FALS are set on RAID0. As you can see access time and transfer speed rocks!
http://www.postimage.org/image.php?v=gxFBTq0
http://www.postimage.org/image.php?v=gxFBQVS
o CD-R - none
o CD-RW Sony AD-5240S-0B
o Sound card onboard
o Speakers QuickShot QS-835
o LAN onboard
o LAN Adaper SiS900 PCI
o Bracket to connect any SATA HDD at eSATA + external SATA input
o CASE Lian li PC-K7B
o PSU Fortron Blue Storm II 500
o Keyboard Benq i150 (keyboard from kitt Key+Mouse)
o Mouse A4Tech SWW-48 (PS2)
o UPS APC Back-UPS CS, 650VA/400W (connected on a Bandrige Power Cable and protection)
o Monitor Horizon 2206SW - 22" LCD
o Scanner Hewlett-Packard ScanJet 2400
o OS Microsoft Windows 7 Home Premium 64 bit English OEM

- The completed system:
http://www.postimage.org/image.php?v=gxFB81i
http://www.postimage.org/image.php?v=gxFBawr
http://www.postimage.org/image.php?v=gxFBd0A
http://www.postimage.org/image.php?v=gxFBfvJ
http://www.postimage.org/image.php?v=gxFBh_S

BIOS Setup: J36_170x20
It took me a lot to understand that I have to enter MANUALLY all default values on BIOS, because on Gigabyte motherboards [AUTO] means really weird things (for example Vtt voltage left on AUTO was set by BIOS to almost 1.4v !!! when the default value is 1.100v and maximum specified by Intel is 1.21v). I am not sure about the other manufacturers but I would recommend to all Gigabyte users to write the default values manually on BIOS to protect their components on the long therm.
UPDATE: Looks like with the latest bios update, namely F8C (motherboard_bios_ga-p55-ud3_f8c.exe) these issues were corrected by Gigabyte. However, I still prefer to set my options manually rather then trusting AUTO.

o BIOS update using QFLASH and 1Gb USB stick. I noticed USB stick of 4Gb (FAT32 same as stick of 1gb) can't be detected by QFlash as a valid drive.

o BIOS settings:
http://www.postimage.org/image.php?v=gxFAxC9
http://www.postimage.org/image.php?v=gxFAA6i
http://www.postimage.org/image.php?v=gxFACBr
http://www.postimage.org/image.php?v=gxFAF5A
http://www.postimage.org/image.php?v=gxFAHAJ
http://www.postimage.org/image.php?v=gxFAK4S
http://www.postimage.org/image.php?v=gxFAMA0
http://www.postimage.org/image.php?v=gxFAP49
http://www.postimage.org/image.php?v=gxFARzi
http://www.postimage.org/image.php?v=gxFAU3r
http://www.postimage.org/image.php?v=gxFAWyA
http://www.postimage.org/image.php?v=gxFAZ2J
http://www.postimage.org/image.php?v=gxFB0xS
http://www.postimage.org/image.php?v=gxFB320
http://www.postimage.org/image.php?v=gxFB5x9

o TurboBoost deactivated from BIOS since bumping the multiplier to x24 at a frequency of 170x24=4080 MHz, is a sure crash at these low voltages, I don’t even want to test that. All power saving features are explicitly ENABLED.
However if you are willing to increase Vcore to make it stable, TurboBoost is possible but I do not recommended it.

o System Memory Multiplier set at 8. This allow rising BCLK to 170 while keeping the memory frequency at only 1360 MHz at 7-7-7-20 timings. This allow me to set Performance Enhance on TURBO without causing instability. This apparently have an effect on Round Trip Latency (RTL) parameter and in my case the difference was clearly measurable in all windows applications. I pondered a lot between setting RAM at 1600 MHz with 8-8-8-20 and Performance Enhance on STANDARD as many users on different forums recommend, and the above settings. I underline again: RAM at 1360 MHz at 7-7-7-20 timings with Performance Enhance on TURBO -> windows "feel" much faster and everything is moving smoothly compared with setting RAM at 1600 MHz with 8-8-8-20 and Performance Enhance on STANDARD.
More about RTL here: http://www.anandtech.com/show/2869/6


- CPU-Z and Gigabyte Easy Tune 6 reports(CPU on idle and full load with Prime95)
http://www.postimage.org/image.php?v=gxFBkv0
http://www.postimage.org/image.php?v=gxFBm_9
http://www.postimage.org/image.php?v=gxFBpui
http://www.postimage.org/image.php?v=gxFBrZr
http://www.postimage.org/image.php?v=gxFButA
http://www.postimage.org/image.php?v=gxFBzsS
http://www.postimage.org/image.php?v=gxFBBY0
http://www.postimage.org/image.php?v=gxFBEs9
http://www.postimage.org/image.php?v=gxFBGXi


- POWER CONSUMPTION (including monitor and entire system connected to the UPS)
o idle (170x9=1530MHz) = 116W (CPU=idle, GPU=idle)
http://www.postimage.org/image.php?v=gxFC4TJ

o full load (170x20=3400MHz) = 180W (CPU=full load, GPU=idle)
http://www.postimage.org/image.php?v=gxFB_Ur

o full load CPU+GPU(170x20=3400MHz) = 232W (CPU=full load, GPU=full load) (Priome95+Furmark in same time)
http://www.postimage.org/image.php?v=gxFC2oA

- CPU TEMPERATURES (recorded after 23 minutes of INTEL BURNTEST - ambient temp = 22ºC)
o idle (170x9=1530MHz) = 31ºC (the highest core)
o full load (170x20=3400MHz) = 60ºC (60,58,58,58 for each core after 23 min. of IntelBurnTest at Maximum stress level)
o Note1: INTEL BURNTEST is the most stressful test for PC and in real programs you can’t get temps this high even if you are using CPU at full load for hours.
o Note2: With this computer at 3.8GHz the speed (Gigaflops) on IntelBurnTest was 53? GFlops but the CPU temperature hit over 80ºC and power consumption increased to 310W. This is not for me, thank you very much.


- TEST1 = Memtest86+ v4.00 (MS DOS) => 3 hours memory testing on MS-DOS mode, 5 time coverage => 0 errors, PASSED

- TEST2 = MEMTEST (Windows)
4 instance running 234% coverage => 0 errors, PASSED
http://www.postimage.org/image.php?v=gxFBYpi

- TEST3 = PRIME 95 running for 3 hours => 0 errors PASSED
http://www.postimage.org/image.php?v=gxFC7nS

- TEST4 = INTEL BURNTEST V.2.3
Standard 1024Mb, 3 times => 0 errors PASSED
Maximum 3461Mb, 10 times => 0 errors PASSED (1392 sec.)
http://www.postimage.org/image.php?v=gxFBVV9

- TEST5 = 3DMark 2006 (tests: select all, settings: 1680x1050, noAA, run each test 10 times, all the rest at default) => 0 errors PASSED
3DMark score 7425, SM2.0 = 2574, HDR/SM3.0 = 2973, CPU = 4964

- TEST6 = FURMARK v.1.8.2
o Benchmarking: fullscreen, 1680x1050, noAA, 600000ms (10 min.) => 0 errors PASSED
Average fps = 21 (min=18, max=33)
GPU temp = 77ºC (normal) -> 78ºC (max.) (22ºC ambient)
o Stability test: windowed, 1024x768: 600 sec => 0 errors PASSED
Average fps = 32
GPU temp = 71ºC (normal) -> 73ºC (max.) (22ºC ambient)
http://www.postimage.org/image.php?v=gxFBOqJ

- TEST6 = PRIME95 + FURMARK Stability test (running both on same time)
FURMARK Stability test: windowed, 1024x768: 600 sec => 0 errors PASSED
o Average fps = 32
o GPU temp = 76ºC (max.) (25ºC ambient)

- TEST7 = LONG TIME STABILITY TEST USING VARIOUS APLICATIONS AND MULTITASKING.
A long time stability test is needed for LGA 1156 platforms, as were seen computers fully stable that pass all stress tests flawlessly and then crash at random. This seems to happen because of a too low VTT - V Core voltages on overclocking. Since I tried to keep voltages (and temperatures) to a minimum for my modest overclock, this test is perhaps the most important indication that the computer is FULLY STABLE.
Playing different games for several hours on different days => no crash, all stable, 0 errors, PASSED

- TEST8 = Apex DC++ and uTorrent
Downloading/Uploading with high speed => no crash, all stable, 0 errors, PASSED

- TEST9 = VIDEO CONVERTING
Converting 104 video files (4 converter instances to use CPU 100%), 687minutes.; average 396sec per file (6.60min) =>0 errors, PASSED

~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~*~

 

[Nhut Pham]

Good work, you're gonna clock that sucker at 4 ghz in a year or two right?

 

[TyrAntilles]

Haha, probably you are right!
I hope I will have time today to upload all the photos and finish.

 

[TyrAntilles]

Finished

This is it, remain to be seen if anybody actually have the patience to open the links to the images.

 

[FloridaBear]

Living in Florida, I'm also looking at overclocking with an eye toward power consumption (and heat production). For my "summer O/C," I wound up with a similar configuration on my Core i7 920: 3.35 GHz (160x21), with VCore set to 1.05 (without VDroop). Under load, the VCore will climb to nearly 1.08, but temps are dramatically lower than my previous 3.8 [email protected] under load. All stability tests so far have passed. I have SpeedStep enabled, which drops the CPU back to 1.9 GHz and 0.75 VCore. Core temps at idle and 15% CPU fan quickly drop into the 40s. Under load, they rarely get into the 70s with moderate CPU fan.

I do need to consider memory timings as I am running 8-8-8@1600--7-7-7 at some lower speed may give better overall results.

In addition, I have dropped my GTX-260 back to "stock" speed of 626 (this is the "superclocked" EVGA card), which cuts temps and power consumption.

Even at full load, my fans are staying much quieter than at 3.8; this is actually mainly due to the PSU fan not winding up like it used to--an indication that my power consumption is significantly lower. I don't currently have a way to measure power consumption, but my UPS may be able to report it fairly accurately (I will have to hook up a serial cable...ugh).

Anyway, these chips sure seem to like low VCore if you don't crank up the speed too much. I'd be interested in other VCore settings at frequencies around this range. What's the lowest VCore at stock speeds, I wonder?

 

[TyrAntilles]

Nice job FloridaBear, the normal VCore on my motherboard for my i5 is 1.075, but this don't have too much importance since these CPU's (i7 and i5) have an adjustable VCore. I think you can find out your lowest VCore at stock speeds setting a low VCore value, running tests and see if it is stable, if is not stable increase it a bit, run tests again and so on till you find the right one. All depends of your patience and your time. A good indication would be to find the lowest voltages where your computer is stable (for your desired clock speed) and then increase them by a notch just to be sure the CPU have enough juice .

About RAM settings, I am not really sure, I found out that in my case Round Trip Latency (RTL) parameter (aka turbo on my BIOS) have bigger influence of how working with computer "feels" then the timings themselves.

I posted my setup on my blog too, with all images included: http://tyrantilles.blogspot.com/ I believe is a lot easier to read this way.

 

[Totto]

Hello to everybody in this great community. This is my first post, but for the last couple of weeks I’ve been doing quite a lot of reading here. First of all sorry for bumping the thread, but I have several questions for the author, or anyone else who would be kind enough to help me with some advice.
Thank you for the great guide, because it helped me understand a lot of things about overclocking. I’m frankly a newbie in this area, so it would be great if you guys could help me out.

I own basically the same rig as the author does. Differences are that my mobo is a UD3L model, the graphics card is a 5770 Vapor-x and the psu is a CM gx750. CPU and Ram are the same, but the processor is with stock cooler.

After deciding that I also want to squeeze that little bit extra from the system without ruining it, I read your guide and started applying the changes in my BIOS. After setting everything up I saved the changes and the system started booting. When it got to the windows 7 loading screen it stopped and restarted. After that it didn’t even get to the loading screen and just offered to start the error checking service which is of no use.

So I started checking everything and managed to spot one difference in your setup that grabbed my attention. I know that every cpu is different, which means that the vcore voltage would be different. Mine is about 1.287 (see pic) in default BIOS settings. That can’t possibly be normal, as opposed to the value that you’ve shown in the screenshots (1.075V).

So the question is – Is this voltage normal, or there’s something wrong? And mainly what changes should I apply to get a stable system running at the settings you’ve shown?

Thank you guys for the tons of valuable info! Greetings from Bulgaria!
P.S. Sorry for the bad English.

 

[TyrAntilles]

hiyas

Hi Totto, welcome to the forums, I am glad you found my guide useful.

First remark is the stock cooler you have. This is just no good. If you are planning to overclock even a little bit, get a real cooler. Otherwise temperature will increase rapidly making your system unstable or even damage it.

Now, about your problem. System instability can arise from many things, like CPU tolerance, as you said, RAM settings and timings, the quality of the motherboard and so on. These settings for my system I found after a LOT of testing and changes. Initially I started with CPU Voltage of 1.21, then lower it bit by bit, playing with RAM voltages and timings as well and with different combinations of settings in BIOS. You can start from the settings I made for my system except voltages.
My suggestion will be to start from my BIOS settings replacing for the beginning only the CPU VCore at 1.21 and Vtt at 1.20 or 1.25, plus set your RAM "Performance Enhance" on STANDARD to see if you can boot like that with your system at 3.4 GHz and get no errors. If it works, then set "Performance Enhance" on TURBO, test again, then again, if all is well, lower voltages bit by bit till you find the sweet spot where your system never crash.

For your question about default voltage assigned by Gigabyte mobo, I wrote on my guide: "It took me a lot to understand that I have to enter MANUALLY all default values on BIOS, because on Gigabyte motherboards [AUTO] means really weird things (for example Vtt voltage left on AUTO was set by BIOS to almost 1.4v !!! when the default value is 1.100v and maximum specified by Intel is 1.21v). I am not sure about the other manufacturers but I would recommend to all Gigabyte users to write the default values manually on BIOS to protect their components on the long therm."
So, as you can see i have the same problem with default voltage settings on my motherboard, that's why I recommend to just set all your settings manually (see my screenshots).

I hope all will be well and you won't fry anything

 

[TyrAntilles]

Ah, and my final settings for voltages where my system never crash and which i use everyday since then are CPU Vcore = 1.09375 and Vtt = 1.170. But it took me about a month of tests to discover these settings, and they will work only on my system, of course. I am sure are other i5's that works at even lower voltages and surely others that require more juice, this is a matter of luck.

 

[noxqzs]

nice work Tyrantilles. If I may make a suggestion, there is a section on this site that has Project Logs. You can create a thread and link to it in normal posts. In addition, many people reserve follow up posts to continue their article. This is the easy workaround to having all pictures shown instead of links.

Totto,

the cpu voltage you are setting is way too low. The only reason Tyran got away with it was because of his specific configuration. If you are on the stock cooler, I would feel comfortable setting it at 1.3v and doing tests from there. Once you find your sweet spot, then start lowering little by little and confirming stability.

 

[EarthDog]

His vcore is on auto...he didnt set anything, it will change depending on the bclk.

THis is a great guide, but let me link you to another one.. check out my signature. One of the best around. Follow that to help you out. Using other people's settings, especially in this case, may not work.

 

[TyrAntilles]

Thank you noxqzs for showing me Project Logs. Is a good idea and I will try to create a thread there when i will have some time.
And thank you EarthDog, yes, I know Miahallen's guide, but unfortunately for me I discover it after i finished all testing on my setup. And the picture Totto provided is refering at the weird CPU Vcore assigned by default by Gigabyte mobo.

 

[Totto]

Thank you for helping me out. I knew that the stock cooler would limit the system a lot, so I ordered a new one. Until it arrives i've been doing even more reading, so now I understand what needs to be done.

The default Vcore troubled me until I noticed that it changed drastically while the pc was working. When in idle, and light use of resources (like playing music and browsing) it went down to something like 0.83 V. When playng some moderately resource hungry games like CoD and FarCry 2 it went up to about 1.15 V. So I have no more worries there.

Next up for me is the overclock, but first I have to get the new cooler. Hope all goes well.

Good luck to everybody, and thaks again!

 

[TyrAntilles]

[EDIT: looks like this post is incorrect, sorry. Thank you MIAHALLEN for pointing it out.]

Your Vcore changes dynamically because you have Load-Line Calibration enabled (Auto means enabled). If you are going to overclock I recommend to manually disabling it because if you increase Vcore it can be dangerous for the CPU. LLCalibration is made initially to supply enough power to the CPU when demanding tasks draw more power and to reduce power when CPU don't work too much. However if you manually increase voltage for overclocking is dangerous to have LLC enabled as it can sometimes boost too much power on an already overvolted CPU. And even if you don't overclock I still recommend disabling it. If you disable LLC you will notice a difference in Vcore voltage between your BIOS voltage and the voltage shown in Windows by CPU-Z for example. This is called vdrop and is normal - Windows will show less voltage then the one you know you set on BIOS. If you are still confused you can search the net for infos about vdrop.

 

[TyrAntilles]

Now i was able to post the entire story on the Project Logs section (thanks noxqzs), with all photos included. You can check it out here: http://www.overclockers.com/forums/showthread.php?t=652549

 

[MIAHALLEN]

This is completely incorrect and misleading...please do not heed the advice given in this post!

Your Vcore changes dynamically because you have Load-Line Calibration enabled (Auto means enabled). If you are going to overclock I recommend to manually disabling it because if you increase Vcore it can be dangerous for the CPU. LLCalibration is made initially to supply enough power to the CPU when demanding tasks draw more power and to reduce power when CPU don't work too much. However if you manually increase voltage for overclocking is dangerous to have LLC enabled as it can sometimes boost too much power on an already overvolted CPU. And even if you don't overclock I still recommend disabling it. If you disable LLC you will notice a difference in Vcore voltage between your BIOS voltage and the voltage shown in Windows by CPU-Z for example. This is called vdrop and is normal - Windows will show less voltage then the one you know you set on BIOS. If you are still confused you can search the net for infos about vdrop.

@ TyrAntilles...please do a little more reading before posting about things you don't understand. You are confusing 4 different technologies/features; LLC is one thing, but you are mixing it up with Speedstep (aka EIST), vdroop and vdrop.

 

[MIAHALLEN]

Other things you have talked about are on spot

It took me a lot to understand that I have to enter MANUALLY all default values on BIOS, because on Gigabyte motherboards [AUTO] means really weird things (for example Vtt voltage left on AUTO was set by BIOS to almost 1.4v !!! when the default value is 1.100v and maximum specified by Intel is 1.21v). I am not sure about the other manufacturers but I would recommend to all Gigabyte users to write the default values manually on BIOS to protect their components on the long therm.
UPDATE: Looks like with the latest bios update, namely F8C (motherboard_bios_ga-p55-ud3_f8c.exe) these issues were corrected by Gigabyte. However, I still prefer to set my options manually rather then trusting AUTO.

Only one other comment:

I will see no benefits to run a 4GHz rocket, my games would not run much more faster and my work won't be done essentially quicker.

...................

- TEST9 = VIDEO CONVERTING
Converting 104 video files (4 converter instances to use CPU 100%), 687minutes.; average 396sec per file (6.60min) =>0 errors, PASSED

I bet if you tested it, you'd save a lot of time converting videos at 4GHz instead of 3.4, invalidating your claim that it won't be "essentially quicker"....it'll be MUCH quicker
Video transcording scales in an almost linear fasion to CPU speed If you were encoding entire movies with high bit rates, with an encoder that used all four cores...it could save quite a bit of time.

Overall though, a good write up, and a nice system, well done

 

[TyrAntilles]

Thank you MIAHALLEN for your correction. I though I got the things right, but it looks I have still much to learn. I must present my excuses to users for my mistake.

I did some more research about these things and I did learn some new things:

- I didn't know the difference between vdroop and vdrop . I did some research and i found out what user hank123 wrote on this forum:
"drop is when you set it as 1.40 in the bios and in windows it shows up as 1.375 droop is when it goes from 1.375 to 1.350 under load.
yes both suck. there are mods out there for most mobos to help this problem. bad droop can hold back a max OC."
However, I have read on many places that having LLC enabled when you overclock, causes dangerous voltage spikes on the CPU and should be disabled.

- About Speedstep (aka EIST). It looks like this was causing Totto's Vcore to change from 0.83 V on idle to 1.15 V on full load. Is that correct?

- And last > LoadLine Calibration reduces the vdroop only. On other words it compensate for the voltage difference between idle CPU and full load CPU. Right?

 

[TyrAntilles]

4Ghz

I bet if you tested it, you'd save a lot of time converting videos at 4GHz instead of 3.4, invalidating your claim that it won't be "essentially quicker"....it'll be MUCH quicker
Video transcording scales in an almost linear fasion to CPU speed If you were encoding entire movies with high bit rates, with an encoder that used all four cores...it could save quite a bit of time.

You are right again. It will be much quicker to encode at 4 GHz then at just 3.4, but I didn't think is worth the risk to stress the CPU and mobo. Even at 3.8 GHz, with my cooler the CPU hit 85 C.

 

[MIAHALLEN]

Thank you MIAHALLEN for your correction. I though I got the things right, but it looks I have still much to learn. I must present my excuses to users for my mistake.

I did some more research about these things and I did learn some new things:

- I didn't know the difference between vdroop and vdrop . I did some research and i found out what user hank123 wrote on this forum:
"drop is when you set it as 1.40 in the bios and in windows it shows up as 1.375 droop is when it goes from 1.375 to 1.350 under load.
yes both suck. there are mods out there for most mobos to help this problem. bad droop can hold back a max OC."
However, I have read on many places that having LLC enabled when you overclock, causes dangerous voltage spikes on the CPU and should be disabled.

- About Speedstep (aka EIST). It looks like this was causing Totto's Vcore to change from 0.83 V on idle to 1.15 V on full load. Is that correct?

- And last > LoadLine Calibration reduces the vdroop only. On other words it compensate for the voltage difference between idle CPU and full load CPU. Right?

Right on all count's except for Load Line Calibration.....in that case it varies from manufacturer to manufactures....there is no standard, and they each take their own stance on their methods of implementation. Some only effect, vdroop, some effect both vdroop & vdrop, and some even overvolt a bit).

Whether or not it's dagerous or causes dagerous spikes is all speculation, and theory. Very few have done any actual investigation into it.....until Bobnova wrote this very interesting peice last week:
http://www.overclockers.com/load-line-calibration/

Personnally, I think our best metric of the dangers of LLC are user reports of CPUs dieing. And in that regard I haven't seen any specific trends until the 32nm CPUs hit the stage. They seem to die pretty easily, although I have not seen anyone draw parrallels between their deaths and the use of LLC. But I have head plenty of reports of users using LLC on 32nm chips for months without problems.

Their's nothing wrong with being overly cautious though