My i5 2500K Build - Overclocking and Benchmarks

[Cilraaz]

Well, this week I got a new Core i5 2500K setup to upgrade my i5 750. Usually, I wouldn't make a one generation jump, but it upgraded my wife's computer to my old i5 and my webserver to her old C2D E8500 (up from a P4 2.66GHz). Anyway, here are the specs:

Mobo - ASRock P67 Extreme4
CPU - Core i5 2500K
RAM - G.Skill Ripjaws X Series DDR3 1600MHz CL9
PSU - Corsair TX650 (carry-over)
GPU - XFX Radeon HD 5770 (carry-over)
HDD - 1TB WD Caviar Black
Case - NZXT Apollo Black

Warning: This will be wordy and full of rambling/babbling about voltage and other adjustments.

I didn't think to take any pictures of the actual build while in process, so I don't have anything for you there. The new UEFI is pretty neat, though. UEFI is essentially a GUI BIOS that allows use of the mouse. It's a bit more intuitive than a standard blue-screen BIOS is. I will have screenshots of my settings, so you'll be able to see that a bit lower in the post. So, what did I do for overclocks? Well, here we go...

I put the build together yesterday (Tuesday), which happens to be one of my guild's raid nights, so I was already short on time. In order to get to raid on time, I hopped into the UEFI and bumped the CPU Turbo Multiplier to 44x and left everything else on auto. This gave me a CPU clock of 4.4GHz on roughly 1.28v vCore. It booted into Windows without a problem and ran WoW without a hitch all night. That's a 33% overclock just by changing the multiplier!

Tonight, I decided to try to get a bit more out of the CPU. I started by bumping the CPU Turbo Multiplier to 46x, forcing a vCore of 1.33v, and lowering the CPU PLL to 1.750v. I booted into Windows without a problem. Sitting idle, I noted a solid vCore of 1.33v, but upon load it dropped sharply to 1.248v (0.082v droop) and threw a 0x124 BSOD. I resolved this by raising the Load-Line Calibration setting one notch to "4". The ASRock P67 Extreme4 has LLC settings from 5 to 1, with 5 being off and 1 being Extreme LLC. Using a setting of 4 brought my voltages up to 1.272 under load (0.058v droop) and allowed a 5 run pass of IntelBurnTest (IBT) at Maximum Stress Level (roughly 6.5GB) at a maximum core temperature of 66°C.

Next, I tried bumping the CPU Turbo Multiplier to 47x (4.7GHz), keeping vCore at 1.33v, keeping CPU PLL at 1.750v, and keeping LLC at 4. This booted into Windows, but threw a 0x124 BSOD as I was starting IntelBurnTest. I decided to raise the LLC level to 3, which resulted in a load voltage of 1.30v (0.03v droop). This resulted in a successful 5 run pass of IBT at Maximum Stress Level (roughly 6.5GB) at a maximum core temperature of 69°C.

Next, I tried bumping the CPU Turbo Multiplier to 48x (4.8GHz), bumping vCore to 1.35v, keeping CPU PLL at 1.750v, and keeping LLC at level 3. This resulted in an idle voltage of 1.355v and a load voltage of 1.32v (0.03v droop). It also threw a 0x124 BSOD during IBT. I tried increasing the vCore to 1.355v and received the same result. I decided to put the vCore back to 1.35v and move LLC up to level 2. This resulted in an idle voltage of 1.36v and a load voltage of 1.352v. It also threw a 0x124 BSOD, but much later into the test. I bumped vCore back up to 1.355v and left LLC at level 2. This resulted in an idle voltage of 1.364v and a load voltage that varied between 1.352 and 1.359v. It also resulted in a successful 5 run pass of IBT at Maximum Stress Level (roughly 6.5GB) at a maximum core temperature of 77°C.

[edit: All of the above information was valid, but Prime95 smashed the overclock into BSOD-land. I'm currently working on stabilizing the overclock. See the "Changes" section below.]

Overclock Images
CPU-Z


Core Temp


AXTU


UEFI Screenshots





Benchmark Images (more to come?)
SuperPi


3DMark (CPU tests only)


Cinebench (CPU test only)



To-Do

• Try to get stability via offset voltage so that the CPU lowers its voltage when idle, as right now only the multiplier drops while the voltage stays the same. - Idea scratched. Too much hassle for too little benefit.
• Run a Prime95 Blend test for an extended period of time to confirm stability. - Complete!
• Run some actual performance benchmarks! I plan to run SuperPi, 3DMark Vantage's CPU tests, and Cinebench. If you have suggestions for others, please let me know. - Complete! However, if you can think of other benchmarks, I'll run them.
• Get UEFI screenshots of my current settings and hopefully of the "final" offset voltage settings. - Complete!

Changes

• I removed the images since they're no longer valid
• I have been fiddling with vCore and got a semi-stable 4.7GHz overclock at 1.360v
• I decided to raise the VTT voltage slightly to try for stability at a lower vCore. It worked.
• As of 3/31 at 9:05PM EST, I'm 50 minutes into a Prime95 blend with the following settings:
  
  • 47x CPU Turbo Multiplier
  • 1.345v vCore
  • 1.750v CPU PLL
  • 1.090v VTT
• I won't be completing a longer Prime95 run this evening, since I have raid tonight. I plan to let this run hit the 1 hour mark, then play around in WoW on an alt for a while to make sure WoW doesn't insta-BSOD. I will then let Prime95 run overnight.
• Crashed while playing WoW, found my vCore had actually been at 1.345v (not 1.340v), bumped it to 1.350v. Temp during a 5-man? 42°C max.
• I tried to get it stable at 4.7GHz, but was unable to. 1.360v allowed for WoW to remain stable throughout raid, but it again failed a Prime95 run.
• I backed off of the 4.7GHz test and moved to 4.6GHz. I eventually found stability (13 hour Prime95) at the following settings:
  
  • 46x CPU Turbo Multiplier
  • 1.330v vCore
  • 1.750v CPU PLL
  • Auto VTT

[Deafyx]

I am a first time OC'er. You can see my first OC I made from early this moring.

What voltage did you need to get to 4.8?

How are you telling your Ambient Temp?

[Cilraaz]

What voltage did you need to get to 4.8?

UEFI fixed vCore setting of 1.355v, which resulted in an idle voltage of 1.364v and a load voltage ranging between 1.352v and 1.359v.

[Deafyx]

UEFI fixed vCore setting of 1.355v, which resulted in an idle voltage of 1.364v and a load voltage ranging between 1.352v and 1.359v.

That makes me happy I should be able to more.

I am not done OC'ing yet. Well for today I am, so I can enjoy my newly built PC lol. I am running a 4.6 OC at 1.32v

[Adappy]

If you plan on switching to offset voltage mode, you might have to turn down/off LLC. I had to turn off LLC on my board (ASUS P8P67 Pro) to use offset mode, otherwise my core voltages went waay too high under load.

[Saithes]

You need the latest LinPacks for IBT... Then you'll see 120-130 GFLops and get a better stress test =P



See

[Cilraaz]

You need the latest LinPacks for IBT... Then you'll see 120-130 GFLops and get a better stress test =P

Where can one obtain those?

[edit: Oh, and I turned off LLC and set an offset of +0.115v. This gives me an idle voltage around 1.12-1.14v and a load voltage of 1.353-1.359v (my previous sweet spot). The only downside is that when load hits, it temporarily jumps to 1.380v and then slides back down to the range I want it in. I'm not sure if I like that extra 0.02v spike. I may just go back to the fixed voltage.]

[Lightstrike]

That's a nice setup, my i7 920 is a powerhungry maniac.. I got it up to 4.2ghz once, couldn't prime it though without a bsod. 4.0ghz I could almost get stable, but got frustrated with changing voltages little by little, so I just settled for a safe 3.8

[Cilraaz]

That's a nice setup, my i7 920 is a powerhungry maniac.. I got it up to 4.2ghz once, couldn't prime it though without a bsod. 4.0ghz I could almost get stable, but got frustrated with changing voltages little by little, so I just settled for a safe 3.8

With the headaches of the offset voltage at 4.8GHz, I'm tempted to just go back to a fixed voltage at 4.7GHz. It was cooler, used a lower voltage, and was less of a pain. I might have found something with the +0.115v and LLC off, though. IBT seemed fine, so I'm going to run Prime95 overnight on blend and see what happens.

[Lightstrike]

I think mine currently isn't even overclocked.. there's nothing I use to merit it to be honest, but it is nice to know I can get those speeds, it means that my pc will last a hell of a long time making it a good investment, currently this sort of hardware isn't stressed by any games out, I only ever maxed mine while video processing when i ripped my dvd collection

[Saithes]

Where can one obtain those?

http://software.intel.com/en-us/arti...pack-download/

Nothing will/should stress harder than LinX or IBT using those new LinPacks. They allow for AVX Instructions from the Processor.

Note: You need Service Pack 1 for Windows 7 as well

[Chronius]

Grats on your new rig.
How about some pictures?
I'll start off with a few pictures of my new beast

[Deleted]

reminds me of an article recently published in CT (german it-magazine) regarding i5 overclocking, they had a nice little graph showing about the same thing you experienced with the voltages: up to ~4.5ghz only minimal voltage adjustments are needed, but anything above requires a lot more core-voltage. also read similar reports in several other forums, so 4.4/4.5/4.6 GHz seems to be the sweet spot to aim for if you want solid 24/7 oc-settings.


its a bit hard to spot, but larger images are not available for free
x axis shows the clock speed, the lowest graph shows required core-voltage. the marker where the 3 graphs jump upwards is 4,5GHz clock speed.
the pretty much linear graph shows performance, the other 2 are power consumption and load-temps.

[Khalelia]

What cooler are u using? Not that your temps scare me, just as additional info...

[Cilraaz]

What cooler are u using? Not that your temps scare me, just as additional info...

I'm using a Xigmatek Dark Knight.

I'm finding that getting a successful Prime95 blend test is requiring much more vCore than a few passes of IBT, though. I'm already up to 1.35v at 47x.

[Nume]

I still havent gotten mine but I'll be getting my new i5 setup Friday. I think I'm just gonna set it up at 4.5 ghz and leave it alone for now though. Just seems safer :P. Good to know it has higher potential though. I've never done any OCing before so still a bit nervous about it heh (this is the first time I've even built a computer... even though I took a technician class and have assembled/disassembled PCs for class before).

[Asera]

I've been having huge performance issues with my virtualization server, and sales on 2500K's with boards have been popping up on NCIX rather often.

And now this.

Damn you people ruining my fangirl AMDism.

[Cilraaz]

I think I'm just gonna set it up at 4.5 ghz and leave it alone for now though. Just seems safer :P.

Well, I'm playing on the ultra-safe side, too. I don't have $225 to blow if my chip explodes, so I'm ignoring all of the 1.4v+ overclocks and trying to keep mine to an absolute maximum of 1.38v. Unfortunately, it's looking like 4.8GHz is unobtainable at that voltage with my chip. I've got it at 4.7GHz @ 1.355v and boosted my VTT by one notch to 1.064v to try to help stability. If it BSOD's again, I'll probably run a Small FFTs test instead of blend to see if it's vCore that's hurting me or VTT.

[Saithes]

Well, I'm playing on the ultra-safe side, too. I don't have $225 to blow if my chip explodes, so I'm ignoring all of the 1.4v+ overclocks and trying to keep mine to an absolute maximum of 1.38v. Unfortunately, it's looking like 4.8GHz is unobtainable at that voltage with my chip. I've got it at 4.7GHz @ 1.355v and boosted my VTT by one notch to 1.064v to try to help stability. If it BSOD's again, I'll probably run a Small FFTs test instead of blend to see if it's vCore that's hurting me or VTT.

The VTT shouldn't be causing any instability if you kept your BCLK at 100mhz.

[Sezuka]

Man, starting to hate my i7 920, needs about 1.312V to stay stable at 3.8GHz. :/