Please provide feedback on my 2500k/6950 over-clocking strategy

[JamesWatt1]

A few days ago, I bought the following machine.

Intel Core i5-2500K
COOLER MASTER Hyper 212 Plus
SAPPHIRE Radeon HD 6950 2GB
G.SKILL Ripjaws Series 4GB DDR3 1600 9-9-9-24
Crucial RealSSD C300 64GB SATA III
ASRock P67 PRO3
COOLER MASTER HAF 912
XFX P1-650X-CAG9 650W ATX12V 2.2
Microsoft Windows 7 Home Premium 64-bit

While I’ve been waiting for it to show up, I’ve come up with a plan for over-clocking it. I would appreciate it if you let me know how I can make my plan better.

I am looking to increase the performance of games like Mass Effect 2 and Left 4 Dead 2.

I am going to use these benchmarks to measure my improvement:
Far cry 2
Stalker call of Pripyat
Heaven

In principle, my strategy for the CPU, CPU memory, GPU, and GPU memory is:
1) Ensure that the component I bought is stable at advertised speeds.
2) Assuming it is, identify a maximum voltage given my risk tolerance.
3) Set the voltage to the highest value I will tolerate.
4) Identify a maximum temperature given my risk tolerance.
5) Increase the clock
6) Test whether the component is stable and that I am below my temperature tolerance.
7) If unstable or above my temperature tolerance, then decrease clock and test again.

My goal with the voltage and temperature tolerances is to hit a point where I am getting a good performance increase, but not seriously diminishing the longevity of the part. Please let me know if any of the values I provide below are too conservative or aggressive given this goal.

I don’t want to run stress tests for too long, because I don’t want over-clocking the machine to take forever. However, if I run them for too short, I may miss stability issues. I am inclined towards an hour per test. Is this a reasonable place for striking this balance?

 

[JamesWatt1]

Here is how this strategy plays out for each component.

CPU
1) Run prime95 for an hour to make sure I got a good processor. Then, before increasing voltage and clock, I will make the following bios changes: set LLC to ultra high and disable speed-step.
2) I am willing to go to a 1.4 core voltage.
3) Set the voltage to a 1.4 core voltage in BIOS.
4) I will tolerate a maximum of 60 degrees Celsius at load.
5) Increase the multiplier until I hit my temperature tolerance. I am going to start with 40 and then increase one at a time. I am not touching the base clock, as I am concerned about screwing up things like my SSD.
6) After each increase, I will run prime95 for an hour.
7) If unstable or above my temperature tolerance, then decrease clock and test again.

CPU memory
1) Run memtest86+ for an hour
2) I am willing to go up to a 1.6 memory voltage. I will increase VTT to 1.1 to ensure that it is within 0.5V of the memory voltage.
3) Make the above changes in bios.
4) I’m not worried about memory temperature. Should I be?
5) Increase the clock and/or decrease the latency.
6) After each increase, I will run memtest86+ for an hour.
7) If unstable or above my temperature tolerance, then decrease clock and test again.

I am not sure whether increasing the memory clock or lowering the memory latency will make a bigger difference in games. I would be curious to hear what you expect. For example, is a latency of 7 cycles @ 1600 Mhz better than 9 at 1866?

GPU
1) Run FurMark for an hour. Then, before over-clocking, I will increase power control in overdrive to +20% and flash my card with a 6950 bios that has the dormant shaders unlocked.
2) I am willing to increase VDDC to 1.175. For, the folks who bricked their 6950s by flashing them to 6970s seemed to do so because the graphics memory voltage was too high, not because of an issue with the graphics core voltage.
3) Set the voltage as described above in MSI After-burner.
4) I am willing to have the card run at 70 degrees C at load.
5) Increase the clock after unlocking the >840 adjustments in MSI Afterburner.
6) Run FurMark for an hour.
7) If unstable or above my temperature tolerance, then decrease clock and test again.

GPU memory
1) Run “Video memory stress test” for an hour.
2) I am going to leave the memory voltage where it is. For, it sounds like the cause of the 6950s that were flashed 6970 and bricked was over-voltaging the memory.
3) I am not changing the voltage.
4) I am not measuring the temperature on the graphics memory.
5) Increase the memory clock in MSI Afterburner.
6) Run FurMark for an hour.
7) If unstable, then decrease clock and test again.

As I write this, I am concerned that this GPU memory strategy is too aggressive. For, it is ambiguous whether the high memory voltage or high memory temperature was the cause of the bricked 6950s that were flashed to become 6970s. I haven’t been able to find guidance on what graphics memory temperatures are safe, so I am reluctant to proceed obliviously. Though, this wouldn’t be an issue if an increase in clock without an increase in voltage has a negligible impact on temperature. Does anyone know if this is the case for graphics memory? If not, do you know if it is the case for the CPU, the CPU memory or the GPU?

 

[GLeeM]

7) If unstable or above my temperature tolerance, then decrease clock and test again.

If it is stable but above temp tolerance, wouldn't you want to lower only the volts first to see if that will put it in temp tolerance?

 

[Skurge]

Sometimes a CPU will hit its clockspeed wall before it hits a voltage/temp wall. So its better You increase clockspeed THEN the voltage.

Just my 2c

 

[videoclone]

That system will play mass effect 2 and Left 4 dead 2 flawless maxX Everything... unless you game at 5760x1080x rez on 3 monitors

"When i overclock" i try and get the maxx stable overclock without having to change Valts/vcore making it a FREE overclock in power usage and the only change is more heat!
At some point down the road when the system requires increased performance i bump it up without too much worries about system death as a replacement "at that time" is at budget pricing.

It worked well for me so far since i started back in 1996 when i was using a -

Pentium1 133Mhz MMX
8MB's of EDO Ram
800MB Hard drive
S4 Trio 4MB Videocard
Windows 95

And now i have a
i7 2600
HD6950 2GB
16GB DDR3 1600Mhz Ram
120GB SSD + 2TB WD Harddrive

good luck with your system overclock .. PS the mod to change your 6950 into a 6970 can cause the cards memory to fail after a few weeks - months and when you flash it back to a 6950 its still rooted.. you may want to check out this links

http://www.techpowerup.com/forums/showthread.php?t=138247
http://www.techpowerup.com/forums/showthread.php?t=138287
http://forums.techpowerup.com/showpost.php?p=2155513&postcount=1027

 

[JamesWatt1]

If it is stable but above temp tolerance, wouldn't you want to lower only the volts first to see if that will put it in temp tolerance?

That's a good point. Thanks.

Sometimes a CPU will hit its clockspeed wall before it hits a voltage/temp wall. So its better You increase clockspeed THEN the voltage.

I hadn't considered that. To accommodate this point, when I find a stable clock at my volt tolerance, I will ramp down my voltage until it is unstable.

Based on you guys' feedback, the last step of my strategy is:

If unstable, then decrease clock until stable. If above temperature tolerance, then decrease voltage until I get within temperature tolerance. When I find the maximum clock within voltage and temperature tolerance, reduce voltage to the minimum that it is still stable.

One of the interesting puzzles in overclocking is minimizing the number of iterations you have to perform before you find the optimal overclock for your system.

 

[JamesWatt1]

That system will play mass effect 2 and Left 4 dead 2 flawless maxX Everything... unless you game at 5760x1080x rez on 3 monitors

"When i overclock" i try and get the maxx stable overclock without having to change Valts/vcore making it a FREE overclock in power usage and the only change is more heat!
At some point down the road when the system requires increased performance i bump it up without too much worries about system death as a replacement "at that time" is at budget pricing.

It worked well for me so far since i started back in 1996 when i was using a -

Pentium1 133Mhz MMX
8MB's of EDO Ram
800MB Hard drive
S4 Trio 4MB Videocard
Windows 95

And now i have a
i7 2600
HD6950 2GB
16GB DDR3 1600Mhz Ram
120GB SSD + 2TB WD Harddrive

This does seem like the optimal strategy for striking the balance between maximizing your machine's longevity and maintaining its relevance. For me, however, I am too impatient about seeing how fast this machine can go. Once there, I don't think I will be satisfied with anything slower.

good luck with your system overclock .. PS the mod to change your 6950 into a 6970 can cause the cards memory to fail after a few weeks - months and when you flash it back to a 6950 its still rooted.. you may want to check out this links

http://www.techpowerup.com/forums/showthread.php?t=138247
http://www.techpowerup.com/forums/showthread.php?t=138287
http://forums.techpowerup.com/showpost.php?p=2155513&postcount=1027

Thanks. I had read some of those articles. That's why I was planning to go with a 6950 bios that changes the shaders only without touching any of the voltages or clocks (http://forums.techpowerup.com/showpost.php?p=2160965&postcount=1234). Then, I would fiddle with the gpu core voltages and clocks manually, while leaving the sensitive gpu memory alone.

 

[JamesWatt1]

Post removed.

 

[SickBeast]

If you want to quickly test your stability, I recommend running LinX, and while it's completing its default test, fire up Prime95 on the "blend" test, along with SuperPi calculating to 32M.

It takes about 10 minutes and is a pretty good barometer in terms of stability. You can leave Prime95 on overnight if you're paranoid about it. IME if it runs for 10 minutes like that, it will run overnight.

In terms of how to overclock, it's an acquired skill that really can't be held down to such a regimented methodology. IMO it's not so much about how many volts you're comfortable as much as it has to do with how many volts the manufacturer says the CPU can take before it burns up. From there, personally, I like to find the optimal performance-per-watt. I got a little greedy with my current setup though and I leave it maxed out.

 

[Diogenes2]

Really, you can spend the rest of your life tweaking for 100 more mhz or 5 more fps, and never get anything done.

Memory = 1333 - Set to the rated params and forget it.
You might get 5% gain in real world performance if you use 2133 ( or anything in between ) instead of 1333
http://www.bit-tech.net/hardware/memory/2011/01/11/the-best-memory-for-sandy-bridge/9


CPU = Look around, see what other people are getting and match your settings based on what you feel you can live with, power draw and temp-wise..

GPU = I haven't done any GPU overclocking lately, so I really don't know what is to be gained there, but I suspect the gain won't be worth spending too much time on it..

 

[Athadeus]

Those temp and volt parameters you came up with are bonkers. 1.4v is way more risky than a meek 60C on the CPU. Keeping an unlocked and overclocked 6950 under 70C is like a total joke, it probably couldn't even do that at stock. You shouldn't really worry about it unless it's too loud, or over 90C.

 

[JamesWatt1]

If you want to quickly test your stability, I recommend running LinX, and while it's completing its default test, fire up Prime95 on the "blend" test, along with SuperPi calculating to 32M.

It takes about 10 minutes and is a pretty good barometer in terms of stability. You can leave Prime95 on overnight if you're paranoid about it. IME if it runs for 10 minutes like that, it will run overnight.

Thank you. This will make my overclocking efforts go dramatically faster. Is there an analog for the GPU such that I can run it for 10 minutes and get a good gauge of its stability?

In terms of how to overclock, it's an acquired skill that really can't be held down to such a regimented methodology.

I think you're right to say that it isn't possible to capture all of the possible cases in a simple if-then list. I guess the idea behind my organization was to present what tools I planned to use and what tolerances I thought I should be comfortable with.

IMO it's not so much about how many volts you're comfortable as much as it has to do with how many volts the manufacturer says the CPU can take before it burns up.

Intel says the sandy bridge cores can take up to 1.52V (see the cart in the first post of http://www.overclock.net/intel-cpus/913062-why-i-think-1-52v-sandy.html). However, I am concerned that if I run it there, it will burn up quickly. For, as voltage increases, electro-migration, a main cause of chip death, increases exponentially.

From there, personally, I like to find the optimal performance-per-watt. I got a little greedy with my current setup though and I leave it maxed out.

To be frank, the eco-friendliness of my overclock is not my top priority.

 

[JamesWatt1]

Really, you can spend the rest of your life tweaking for 100 more mhz or 5 more fps, and never get anything done.

My primary motivation for overclocking is learning more about electricity and cooling. Being able to play games at a higher fps is a distant second (though, I admit, this was unclear in my OP).

Memory = 1333 - Set to the rated params and forget it.
You might get 5% gain in real world performance if you use 2133 ( or anything in between ) instead of 1333
http://www.bit-tech.net/hardware/memory/2011/01/11/the-best-memory-for-sandy-bridge/9

Thanks--that encourages me to de-prioritize memory over-clocking and to get an idea for the impact I should expect from it.

 

[JamesWatt1]

Those temp and volt parameters you came up with are bonkers. 1.4v is way more risky than a meek 60C on the CPU.
Keeping an unlocked and overclocked 6950 under 70C is like a total joke, it probably couldn't even do that at stock. You shouldn't really worry about it unless it's too loud, or over 90C.

Thanks for calling these into question--I think that choosing good tolerances is the most important part of the process. Here's how I came up with them. I would be curious to hear more about what's behind your suspicion of them.

CPU voltage
My choice is based on Asus's testing (http://techreport.com/forums/viewtopic.php?f=4&p=1054545). It seemed to suggest that 1.4-1.425 strikes the balance between performance and temperature impact.

Overall a key item to note is the best voltage to oc scaling range potential for the turbo multiplier is 1.400 to 1.425 vcore. Using this voltage range with an LLC recommendation of ultra high will generally provide the best scaling potential with proper load temperatures*. We have generally found exceeding this voltage will not provide additional scaling or will increase load temperatures to a high level with synthetic load applications ( like Prime, Linx, OCCT )

Where have you seen that it is too high? I saw 1.35 V (http://forums.overclockers.co.uk/showthread.php?t=18227647), but it I couldn't tell how they came up with it.

CPU temperature
I read about 60 degrees max in a couple of places.

http://www.xbitlabs.com/articles/cpu/display/newbie-oc-guide_6.html
http://www.overclockersclub.com/pages/overclock_faq/

What do you think is reasonable for a maximum CPU temperature and why?

GPU temperature

I saw 70 degrees a couple of places.

http://forums.nvidia.com/index.php?showtopic=3881&st=0&p=28764&#entry28764
http://forums.guru3d.com/showthread.php?t=336168

How did you come up with 90 degrees?

 

[Hogan773]

I am also interested in this - but my view is to MINIMIZE VOLTAGE at a given clock speed.

How specifically do I set it so that my system will still throttle up and down with speed and voltage, BUT also ensure that it doesn't go crazy high on voltage when its stressed? Is there a setting for a "ceiling" on voltage but not a floor? Or do you just do a "voltage offset" which basically adds or subtracts X volts at every point along the spectrum (which could help keep top end voltage reigned in but might then result in undervolting when the system is idling at 1.6ghz and .9 volts normally)......

help? I'm new to all of this stuff having only had Dells up till now

 

[Athadeus]

James, you are right that tolerances are of key importance in the process. I have not had the time to read through the sources from all your links, but some seem particularly good, while others seem poor.

CPU Voltage
I would say both sources are good. However, they are targeting different audiences. The techreport thread is for enthusiasts who are willing to take serious stability and hardware life risks, whether using exotic cooling to reduce that risk or not.

The Anandtech member Aigomorla falls into that category, and he has tested 32nm SB and pre-SB chips. He basically came to the conclusion same conclusion as the second source, overclockers.co.uk. That is why 1.35 has become the recommended max voltage by most members on this forum, and 1.375v really the threshold of safety. However, these values have hardly been proven scientifically (through significant testing).

CPU Temperature

http://www.xbitlabs.com/articles/cpu/display/newbie-oc-guide_6.html
http://www.overclockersclub.com/pages/overclock_faq/
These are both too dated to be good references, especially the second one (referencing AMD Athlon XP 1500-3200s from 1.33-2.2ghz, and P2s). Since you are talking specifically about thermals, those don't change quit as fast. Neither really says 60C should be ceiling though, rather one says 70C for continuous, and other just says 80C max general.

I am generally willing to run higher CPU temps because I prefer not to spend as much, or have as loud of a CPU cooler. 70C would be a good maximum for a system that had to run 24/7, but for normal gaming use, I'd say even 80C is fine.

http://www.anandtech.com/show/4083/...-core-i7-2600k-i5-2500k-core-i3-2100-tested/3
Honestly, Anand was able to hit 4.4ghz with the stock cooler on both his chips. He doesn't mention temps, but I would probably just use the stock cooler if I bothered to upgrade to SB and used a similar overclock. If I was in the mid 80s with no HT while gaming, I might just drop the overclock a little, but even the supposedly good value CM Hyper 212+ does not seem like a good deal anymore to me.

GPU Temperature

From your link here:
http://forums.nvidia.com/index.php?showtopic=3881&st=0&p=28764&#entry28764
2 posts up, a guy says nvidia GPUs are rated to 120C.
4 posts down, someone is hitting 110C.
That's just a plain bad source, as they are different products, and the range they are suggesting is getting into absurdity.

From your link here:
http://forums.guru3d.com/showthread.php?t=336168
2nd guy says they (your actual 6950 product) get to 92C in furmark.

http://forums.anandtech.com/showthread.php?t=2139877
Now, from Vivithemage's post on these forums, he says his own X-Fire unlocked, overclock 6950s get to 88-90C.

http://www.anandtech.com/bench/Product/293?vs=299&i=228.232.233.231.229.230.234.235
Finally, one of the ultimate sources.

Basically says, don't worry about your GPU temp. You will have instability from other reasons before you can get it to overheat (barring catastrophic cooler failure).

 

[Athadeus]

Hogan, you have some good thoughts. Many boards have an offset voltage you can set, which basically is like using Auto + offset. However, a lot of Intel (based, not brand) boards have insane auto regulation when overclocking, and can put vcore at 1.6 or cpu/dram at 1.8.

I am not sure, but what I think a lot of users here like to do is use full manual voltage, load line calibration setting on, and then if they care about efficiency, enable the power features. Usually the power saving features will drop the settings below your manual ones while not under stress, and just return to your full power manual settings when you need it.

Personally, on my desktop, I have it on all manual, no power savings, on an X58 platform which isn't particularly efficient. However, my idle usage is only like 110w with i7 930 @ 3.7ghz 1.25 vcore, 1x GTX 460 1gb, 1 system SSD, 1 HDD (almost always not spinning).

 

[Hogan773]

Hogan, you have some good thoughts. Many boards have an offset voltage you can set, which basically is like using Auto + offset. However, a lot of Intel (based, not brand) boards have insane auto regulation when overclocking, and can put vcore at 1.6 or cpu/dram at 1.8.

I am not sure, but what I think a lot of users here like to do is use full manual voltage, load line calibration setting on, and then if they care about efficiency, enable the power features. Usually the power saving features will drop the settings below your manual ones while not under stress, and just return to your full power manual settings when you need it.

Personally, on my desktop, I have it on all manual, no power savings, on an X58 platform which isn't particularly efficient. However, my idle usage is only like 110w with i7 930 @ 3.7ghz 1.25 vcore, 1x GTX 460 1gb, 1 system SSD, 1 HDD (almost always not spinning).

Ok thanks. I guess I don't care about the POWER USAGE per se from a cost standpoint (its like an extra lightbulb). I DO care about whether having 1.3 volts running at ALL times thru the chip is going to degrade its life vs having it idling lots of times at 0.9V which it seems to do now when it idles at 1.6ghz. But I totally agree that I don't want my mobo just throttling the he11 out of it when on "auto"

I'm a complete noob to building and OCing. Would you mind outlining a couple of the actual steps I need to take to make these changes in my bios?

I know where to change the CPU multiplier, but

1) offset voltage vs just fixing the voltage -- from a practical standpoint if I choose the offset voltage, how and how much to subtract? Do I just set the multiplier then run Prime95 and WATCH my voltage using CPUZ or HWMonitor? And just see what my mobo actually does with it? And then if its spiking to 1.4 or 1.5, I try to quickly cancel Prime and then pray that my CPU wasn't damaged?

2) what is PLL voltage and do I need to adjust it or leave it on auto?

3) what is "load line" voltages and why do you need to make changes on that?

What I really want is a max ceiling on voltage of [1.3]......but anything below that, for the CPU to get as much voltage as it needs but be able to keep all the idling CPU multipliers and voltage. Then I will adjust my max CPU multiplier to whatever the highest (4.4?) that can stay stable at that max voltage.

So is the above scenario possible and if so, how do I do it?

 

[Diogenes2]

1. If you just ' set the voltage ', it will pretty much stay there., but tend to dip ( vdroop ) under high loads; this is where LLC ( Load line Calibration ) comes in - it will try to keep your voltage where you set it, but the downside is, LLC can spike your voltage past where you set it under certain condidtions. The adjustments just determine how aggressive LLC is . If you want a stable overclock over 4.5 you have to assume the risk of a more agressive LLC..

If you need 1.35 for your overclock, and you just set it there, while probably perfectly safe, you do have additional heat and power draw even while idle..

If you use offset, it will not bump the voltage past default + offset, but will lower it during idle time..

You have to experiment with the offset amount to determine what yields a stable overclock.

I have found that for my 4.9ghz, I have to set offset to +.35 with LLC set to high .
With these settings my max vcore has been 1.45, but it seems to hover around 1.43 under full load ..

I have left all the other voltages, PLL & etc., on default ..

I can't say if these settings will work for you, or be responsible for any voltage adjustments you may choose to make.

 

[Hogan773]

1. If you just ' set the voltage ', it will pretty much stay there., but tend to dip ( vdroop ) under high loads; this is where LLC ( Load line Calibration ) comes in - it will try to keep your voltage where you set it, but the downside is, LLC can spike your voltage past where you set it under certain condidtions. The adjustments just determine how aggressive LLC is . If you want a stable overclock over 4.5 you have to assume the risk of a more agressive LLC..

If you need 1.35 for your overclock, and you just set it there, while probably perfectly safe, you do have additional heat and power draw even while idle..

If you use offset, it will not bump the voltage past default + offset, but will lower it during idle time..

You have to experiment with the offset amount to determine what yields a stable overclock.

I have found that for my 4.9ghz, I have to set offset to +.35 with LLC set to high .
With these settings my max vcore has been 1.45, but it seems to hover around 1.43 under full load ..

I have left all the other voltages, PLL & etc., on default ..

I can't say if these settings will work for you, or be responsible for any voltage adjustments you may choose to make.

Don't worry - if I fry my chip I won't sue you

So I guess I was confused about offset. I thought I would be using a negative offset to lower the voltage from what the mobo would naturally want to kick out when I change the multi up to 44. But it sounds like actually it is a positive number to add extra voltage to allow stability at the higher multiplier? So if ALL I do is leave everything on AUTO, and just change the multiplier, then will the mobo be giving the same voltage at full load as it does under full load at the stock multiplier? I thought that was the dangerous part, that the mobo would just start jacking up voltage willy nilly? Or is the OFFSET basically just a governor that tells the mobo "no matter what, don't increase voltage by more than XYZ mV"?

My ASRock's mobo manual is pretty sparse with actual explanations of WHAT I'm actually adjusting, so I'm trying to understand before I just start go changing numbers.......

 

[Diogenes2]

You can also specify a negative offset- which would be for underclocking - but that doesn't apply here..

My settings are for the Asus board - What kind of LLC options do you have ?

Mine are standard - medium - high and extreme ..

I was able to do 4.8 with medium and offset +.25 - and my max vcore was 1.38 ..

If you have those options - I would proceed with adjusting LLC and Offset for an OC you are happy with.

When I tested with vcore on auto - it would rise to 1.5 and beyond for multipliers above 45 ..

 

[Hogan773]

You can also specify a negative offset- which would be for underclocking - but that doesn't apply here..

My settings are for the Asus board - What kind of LLC options do you have ?

Mine are standard - medium - high and extreme ..

I was able to do 4.8 with medium and offset +.25 - and my max vcore was 1.38 ..

If you have those options - I would proceed with adjusting LLC and Offset for an OC you are happy with.

When I tested with vcore on auto - it would rise to 1.5 and beyond for multipliers above 45 ..

So at offset of +.25, does that mean if voltage at Bios defaults at idle 1.6ghz was 0.9V, that it would then idle at 1.15V minimum? ie is it just a +.25 across the board bump?

 

[Diogenes2]

I haven't really figured it all out. It seems to limit the maximum more than it bumps the minimum. I still idle at about 0.9 - 1.1..
Before I started tweaking things I would see it as low as .88, but then, that was without LLC on also ..
It's easy for me to waste too much time fidling with things, so I'm happy to find something I'm comfortable with and leave it alone.
I may play with it some more later ..

You can do crazy stuff with this BIOS like increase " CPU current capability " to 150%.
Not sure what that does, but it sounds scary..

 

[Hogan773]

I haven't really figured it all out. It seems to limit the maximum more than it bumps the minimum. I still idle at about 0.9 - 1.1..
Before I started tweaking things I would see it as low as .88, but then, that was without LLC on also ..
It's easy for me to waste too much time fidling with things, so I'm happy to find something I'm comfortable with and leave it alone.
I may play with it some more later ..

You can do crazy stuff with this BIOS like increase " CPU current capability " to 150%.
Not sure what that does, but it sounds scary..

Yeah that's my fear on some of these things - that I'm just tweaking settings but I don't know what I'm actually doing. My ASRock manual isn't that helpful........for example when describing the LLC option it might say "Allows adjustment of LLC parameter" -- wow thanks for that insightful tutorial......

I guess I'll just play around a little, slowly, and see how things change.

EDIT: so I just came across this page in case anyone else is interested - gives me a better def of LLC

http://www.overclockers.com/load-line-calibration/

Here is another good article from our local friends at AT

http://www.anandtech.com/show/2404/5

sounds like using LLC and offset may not be as great an idea as I thought

 

[JamesWatt1]

Commentary on sources

Thanks for this. It was very clarifying.

I got some baseline temperature data and took my first stab at overclocking this thing. Ultimately, I want to evaluate the value of the Hyper 212, so I did all of this with the Intel stock heatsink/fan.

Here's how I measured temperature and voltage:

CPU temperature: After 10 minutes of Linx 0.6.4.0 & Prime95 25.11.1.0 Blend & Super pi 1.1e 32M / Degrees C as reported by ASRock eXtreme Tuner

CPU voltage: After 10 minutes of Linx 0.6.4.0 & Prime95 25.11.1.0 Blend & Super pi 1.1e 32M / Vcore as reported by ASRock eXtreme Tuner

GPU temperature: After 10 minutes of FurMark 1.6.0 / Stability test / Xtreme burning mode / Full screen / 1680 x 1050 / MSAA None / Degrees C as reported by FurMark

Test 1
Settings: default
CPU temperature: 60 degrees C
GPU temperature: 84 degrees C
CPU voltage 1.216 (I didn't touch this in bios, it automatically selected it)

Test 2
Settings: default, except CPU max ratio (what ASRock calls the multiplier) = 48
CPU Voltage: 1.354 (I didn't touch this in bios, it automatically selected it)
CPU temperature: 94 degrees C (terrifying!)

It crashed 9 minutes into testing--I've heard the PC restarts itself if the diode on the chip reads 100 degrees C.

Test 3
Settings: default, except CPU max ratio = 48 & speed step disabled

It crashed while windows was booting.

Test 4
Settings: default, except CPU max ratio = 44 & speed step disabled
CPU Voltage: 1.354 (I didn't touch this in bios, it automatically selected it)
CPU temperature: 94 degrees C

10 minutes of stress testing succeeded. Though, I don't plan to run it like this given the outrageous temperature.

Questions
One thing I found puzzling was that even with speed step off, when I set the multiplier to 44 and 48 in BIOS, the CPU throttled down to multiplier = 37 when running the stress tests. Why is this and how can I stop it?

I find the high temperatures surprising given the voltage. I wonder if I screwed up the HSF install--though I thought that if the four plastic clips (the clear pieces on the bottom right picture on http://www.geekwithlaptop.com/intel-redesigns-heatsink-for-sandy-bridge) snapped in to the holes on the motherboard, then I was golden. Any ideas as to why this might be?