You have the ASRock Extreme4 too? I've been playing with it all afternoon and I can give you my feedback.....
1) GET the Hyper 212+. I paid $15 AR for it. Here are my OC results and temps.....temps WAY better than what your stock is getting.....
-I am 1.5 hours into the following Prime test - 44x multiplier, 1.26V, and temps are hovering around 58-61C and HWMonitor shows the max on one core hit 65C at some point in the past 1.5 hours.
2) for your throttling issue, I had the same problem. I saw in Anandtech review of the mobo what is the solution I believe. You need to change the Long Duration Power Limit higher than the 95W that the mobo starts with. I changed mine to 110W, and haven't had any throttling since. I've seen others who have put it to 255 (highest value I assume) but I figured I might as well keep some safeguards in place rather than just take out the seatbelts and airbags completely.
3) I'm kinda liking this OFFSET voltage setting rather than FIXED. Fixed means it stays at the spec amount all the time. Offset retains the throttle down functions, so when it idles it heads back to 16x and the voltage is around 0.95+/- for me. I'm trying to design a moderate OC that I can "set it and forget it" so I like minimizing the voltages. I am not gonna be a "here's my screenshot of 5.3ghz" ooops I fried my CPU after 2 months kinda guy.
So compare my 44x to yours - temps MUCH better and if I were you, I'd do your own voltage as it seems your mobo is being pretty volty on AUTO. Also DO NOT disable speedstep and all the other junk....just leave it on. It doesn't seem to hurt the OC, and it allows all the idling stuff to stay in place.
Also I figured out that ASRock Load Line Calibration is Level 5 = none, while Level 1 is Lots! I choose to keep LLC off - seems there is debate about LLC's benefits vs risks.
Also I trimmed CPU PLL to 1.79 (board defaulted to 1.83) and it appears others are trimming more to the 1.75ish area. Not sure if it helps any, but didn't seem to hurt, and in general I guess "lower voltage at a given speed" is fine as long as its stable.
Fun stuff!
Let me know if you have questions.
Dadofamunky
Platinum Member
- Jan 4, 2005
- 2,184
- 0
- 0
OP: Temps are waaaaay too high, and voltage is too high for a 4.4 GHz OC. You should need no more than 1.32V for 4.4V, and possibly less. Forget the offset and go with realistic base values and a realistic speed level and make that work first. Look at my system, which is very similar and passes Prime overnight tests and LinPack. It's at 1.32V. Make sure you're not overvolting the RAM! Set the PLL to 1.9V max, PLC to Ultra settings or higher in BIOS. I can post more details tomorrow if requested, time to hit the rack.
I have the ASRock Pro3.
I have this. I wanted to see how hard I could push my cpu on stock cooling before putting it on.
Huge thanks for this--fiddling with Long Duration Power Limit and Short Duration Power Limit fixed the issue. Admittedly, I don't know exactly what it does (and surprisingly no one I could find on Google does either). Based on the name, though, it sounds like it sets a maximum amount of power the CPU can draw at a time. If the CPU pierces that threshold, clock speed is reduced.
If this is the case, then it is redundant with the core voltage and maximum core current settings in BIOS. This is because Power (in watts) = Voltage (volts) * Current (amps). In my case, 191.25 W = 150 amps (Core current limit) * 1.275 volts (core voltage).
Because I want to put caps on current and voltage independently, I decided to remove the cap on power. So, I set my power limits to the BIOS maxes:
Short duration power limit = 500
Long duration power limit = 500
Since I've made the change, I haven't seen anything crazy. Though, I haven't compared actual current or wattage draw before and after the change.
This was another very beneficial suggestion. Setting the voltage myself gave me a lot more control over my temperatures (thankfully, no more absurd 94 degree C temps). I opted for a fixed voltage, because it is simpler (though I appreciate in the long-run, your CPU will use less voltage if you take the offset approach).
Due to your help, here were my last stable iterations:
Test 5
Multiplier = 43
BCLK = 100
SpeedStep = on
Short Duration Power Limit = 500
Long Duration Power Limit = 500
CPU core voltage = 1.2 V (though I actually only get 1.192--I have found I tend to get about .010V less than I ask for)
CPU temperature: 66 degrees C
I was pleased with 4.3Ghz with (a little less than) stock voltage.
Test 6
Multiplier = 45
BCLK = 100
SpeedStep = on
Short Duration Power Limit = 500
Long Duration Power Limit = 500
CPU core voltage = 1.275 V (though I actually only get 1.266)
CPU temperature: 80 degrees C
I think this is the best I can do with the stock cooler given the thresholds I'm comfortable with. As far as I'm concerned this is a great result--36% faster!
My next step is installing my Hyper 212 and seeing if I get closer to my voltage threshold without piercing my temperature threshold.
Right now, because my constraint is temperature, I am going to hold off on fiddling with these. Though, it is nice to have these in my back pocket when I hit my next wall.
Last edited:
This strategy worked out well as I mentioned in the tests in my last post. Thanks!
Do you think 1.58V is too high? If so, why?
I wasn't planning to touch PLL as my understanding is that it impacts FSB/BCLK and I don't want to touch either.
What is PLC? If it was a typo and you meant LLC, I have seen mixed things about turning it on or off. What is your argument for setting it to ultra or higher?
Last edited:
Why put your power limit up to 500? Just wondering. I just put it to 110 like Anand did and I haven't seen the throttling. I guess I need to think more about what its doing - but I figured if its some form of safety belt, I might as well keep in in place (just loosen it).
I think people like LLC because then they don't see any Vdroop - so if you're fixing a voltage at 1.25 then under load it will stay there. I'm not quite convinced because of the microspikes - if your mobo is basically stepping on the voltage gas once the load comes, I can't see how it can't overvolt for a split second once the load is released until it figures out that it needs to step off the voltage. I guess lots of people say "well it never hurt MY chip" but then again, I doubt most people end up with a failed chip so they probably would never know if they're doing any damage.
I think people like LLC because then they don't see any Vdroop - so if you're fixing a voltage at 1.25 then under load it will stay there. I'm not quite convinced because of the microspikes - if your mobo is basically stepping on the voltage gas once the load comes, I can't see how it can't overvolt for a split second once the load is released until it figures out that it needs to step off the voltage. I guess lots of people say "well it never hurt MY chip" but then again, I doubt most people end up with a failed chip so they probably would never know if they're doing any damage.
On Cooling:
Firstly, ditch that stock cooler. They are 99% of the time crap, and loud (but "free"). Good aftermarket cooling doesn't have to be loud either. The guiding factor should be price! You can get a $100 air cooler that will keep your system below 70'C even at 1.5v with varying noise results, or you could spend the same $100 on a super silent cooler. My personal solution was a Thermalright Ultra 120 Extreme with a super high CFM fan and a fan controler to keep it in check. Stock (100%) the fan runs @ 2600 RPM, 122CFM, about 35Db. Use of a controler to limit this to a much quieter 1600 RPM (below ambient noise at my desk) when idle/low load and ramps up as load increases. The beauty of a custom HS and Fan is you can use what ever CFM / Db combo is comfortable to both your ears and your piece of mind. Long story short {too late}: Stock CPU cooling = Fail, Aftermarket = Win
On temps:
90'C is waaaaay too high. My safety zone is 70'C for constant load and about 80'C for intermitent load. Don't forget things like LinX/Linpack/Intel Burn Test will generate temps nearly 10'C higher than even the most intense gaming due to the sheer 100% stress level. For example, my system will hit about 70-75 on core 0 with IBT, but in my typical usage hardly breaks 60'C. It peaks at ~66'C on core 0 with prolonged full load on all four cores on my most demanding application.
On Volts/Vcore/Vcc:
In the case of overclocking, Speed doesn't kill. Voltage kills. So, as others have suggested, either adjust voltage as required to maintain stability, or at least decrease voltage until you reach instability once you have your core speed to find the min required voltage and run 1-2 steps above it as needed. The max safe voltage is an issue of debate, but to para-phrase another great AAT forum member, all over-volting has a negative effect on chip life. The question is what is acceptably unsafe, and what is unacceptably safe. Reducing life from ~20yrs to ~5yrs would be acceptable, but running a 2.0v Vcore suiccide bench at 6.9Ghz for 3days would probably be unacceptably safe!
On LLC
LLC is intended to correct that drop, known as vDroop, you've seen in the power delivered to your CPU. vDroop exists because when your CPU goes from full power drawing loads of Amps to a sudden idle state, it takes the power delivery system a few micro seconds to reduce the flow, so for an instant your CPU is being over-supplied with power. The drooping assures that this spike will not go above the originally intended voltage set in BIOS. With LLC enabled, this effect is negated, and your CPU is supplied closer to the intended voltage at load, but when that load is removed the resulting spike in momentary over-power goes above the intended voltage set in the BIOS. For example say your CPU needs 1.3625v in BIOS with LLC off to be stable. At load it might draw, say, 1.279V so that when the load is removed it might spike to say, 1.3614v.. within spec set in the BIOS. With LLC enabled only BIOS 1.3125v might be needed to be stable, yielding 1.29v load, thus allowing a lower overall settings for general idle/load conditions; but when load suddenly drops to zero the resulting spike might be say 1.44v, taking the system over the spec voltage of 1.3125v and over the non-LLC required voltage of 1.3625v for a few micro-seconds.
**Please note all of the above numbers are fictional, but the concept is accurate**
The debate about LLC is due to the lack of information regarding the length, severity, and overall effect the spikes mentioned above. My advice is this: LLC is a fine tool for the bleeding edge, imformed carefull use can have great benefits. Use it like you would a razor, cautiously, and it may prove to be just the tool for the job!
On the relationship between Volts, Speed, and Temps:
Two things cause the temps to rise. Speed causes the temps to rise indirectly because the chip needs more Amps at the same Voltage to run at the higher speed, thus increasing the total Watts drawn (hence your experience with Long/Short duration power limit). Voltage causes temps to rise directly [more] because the chip begins to draw more Amps at a higher Voltage causing a huge increase in Watts. The signal to noise ratio also degards as temps rise, which is another area cooling comes in. For example your CPU might be 100% stable at 1.4v @ 50'c, but increase the temp to 65'c and the chip might need 1.45v to be 100% stable.
Sorry, that became so much longer than I had first thought! This was typed at work with many interuptions so please excuse the mess! If you have any questions just ask!
Cheers mate, and good luck!
Firstly, ditch that stock cooler. They are 99% of the time crap, and loud (but "free"). Good aftermarket cooling doesn't have to be loud either. The guiding factor should be price! You can get a $100 air cooler that will keep your system below 70'C even at 1.5v with varying noise results, or you could spend the same $100 on a super silent cooler. My personal solution was a Thermalright Ultra 120 Extreme with a super high CFM fan and a fan controler to keep it in check. Stock (100%) the fan runs @ 2600 RPM, 122CFM, about 35Db. Use of a controler to limit this to a much quieter 1600 RPM (below ambient noise at my desk) when idle/low load and ramps up as load increases. The beauty of a custom HS and Fan is you can use what ever CFM / Db combo is comfortable to both your ears and your piece of mind. Long story short {too late}: Stock CPU cooling = Fail, Aftermarket = Win
On temps:
90'C is waaaaay too high. My safety zone is 70'C for constant load and about 80'C for intermitent load. Don't forget things like LinX/Linpack/Intel Burn Test will generate temps nearly 10'C higher than even the most intense gaming due to the sheer 100% stress level. For example, my system will hit about 70-75 on core 0 with IBT, but in my typical usage hardly breaks 60'C. It peaks at ~66'C on core 0 with prolonged full load on all four cores on my most demanding application.
On Volts/Vcore/Vcc:
In the case of overclocking, Speed doesn't kill. Voltage kills. So, as others have suggested, either adjust voltage as required to maintain stability, or at least decrease voltage until you reach instability once you have your core speed to find the min required voltage and run 1-2 steps above it as needed. The max safe voltage is an issue of debate, but to para-phrase another great AAT forum member, all over-volting has a negative effect on chip life. The question is what is acceptably unsafe, and what is unacceptably safe. Reducing life from ~20yrs to ~5yrs would be acceptable, but running a 2.0v Vcore suiccide bench at 6.9Ghz for 3days would probably be unacceptably safe!
On LLC
LLC is intended to correct that drop, known as vDroop, you've seen in the power delivered to your CPU. vDroop exists because when your CPU goes from full power drawing loads of Amps to a sudden idle state, it takes the power delivery system a few micro seconds to reduce the flow, so for an instant your CPU is being over-supplied with power. The drooping assures that this spike will not go above the originally intended voltage set in BIOS. With LLC enabled, this effect is negated, and your CPU is supplied closer to the intended voltage at load, but when that load is removed the resulting spike in momentary over-power goes above the intended voltage set in the BIOS. For example say your CPU needs 1.3625v in BIOS with LLC off to be stable. At load it might draw, say, 1.279V so that when the load is removed it might spike to say, 1.3614v.. within spec set in the BIOS. With LLC enabled only BIOS 1.3125v might be needed to be stable, yielding 1.29v load, thus allowing a lower overall settings for general idle/load conditions; but when load suddenly drops to zero the resulting spike might be say 1.44v, taking the system over the spec voltage of 1.3125v and over the non-LLC required voltage of 1.3625v for a few micro-seconds.
**Please note all of the above numbers are fictional, but the concept is accurate**
The debate about LLC is due to the lack of information regarding the length, severity, and overall effect the spikes mentioned above. My advice is this: LLC is a fine tool for the bleeding edge, imformed carefull use can have great benefits. Use it like you would a razor, cautiously, and it may prove to be just the tool for the job!
On the relationship between Volts, Speed, and Temps:
Two things cause the temps to rise. Speed causes the temps to rise indirectly because the chip needs more Amps at the same Voltage to run at the higher speed, thus increasing the total Watts drawn (hence your experience with Long/Short duration power limit). Voltage causes temps to rise directly [more] because the chip begins to draw more Amps at a higher Voltage causing a huge increase in Watts. The signal to noise ratio also degards as temps rise, which is another area cooling comes in. For example your CPU might be 100% stable at 1.4v @ 50'c, but increase the temp to 65'c and the chip might need 1.45v to be 100% stable.
Sorry, that became so much longer than I had first thought! This was typed at work with many interuptions so please excuse the mess! If you have any questions just ask!
Cheers mate, and good luck!
Last edited:
LiuKangBakinPie
Diamond Member
- Jan 31, 2011
- 3,903
- 0
- 0
Here is a summary of the stable iterations I've done so far.
Frequency / Vcore / Tcore / Heat-sink / BIOS settings
3300 Mhz / 1.216V / 60 C / Intel stock / Default
4400 Mhz / 1.354V / 94 C / Intel stock / Default
4300 Mhz / 1.200V / 66 C / Intel stock / Power limits = 500W
4500 Mhz / 1.275V / 80 C / Intel stock / Power limits = 500W
4700 Mhz / 1.350V / 66 C / Hyper 212 / Power limits = 500W
It's amazing to me what an improvement the Hyper 212 is over the Intel stock cooler. To be running .075 Vcore higher yet 14 degrees C cooler is remarkable.
Given the constraints I set at the outset, I think I'm stuck at 4.7 Ghz. While I'm well within my temperature threshold (66 vs. 80 degrees C), I've hit the voltage wall that I'm comfortable with (1.35 Vcore).
The next line of investigation is going to be geared towards minimizing temperature. I will be looking at what fan types and arrangements strike a good balance between low temperature and high noise. I may also fiddle with some of the more exotic cooling methods.
After that, I plan to branch out into GPU over-clocking and CPU memory over-clocking.
Frequency / Vcore / Tcore / Heat-sink / BIOS settings
3300 Mhz / 1.216V / 60 C / Intel stock / Default
4400 Mhz / 1.354V / 94 C / Intel stock / Default
4300 Mhz / 1.200V / 66 C / Intel stock / Power limits = 500W
4500 Mhz / 1.275V / 80 C / Intel stock / Power limits = 500W
4700 Mhz / 1.350V / 66 C / Hyper 212 / Power limits = 500W
It's amazing to me what an improvement the Hyper 212 is over the Intel stock cooler. To be running .075 Vcore higher yet 14 degrees C cooler is remarkable.
Given the constraints I set at the outset, I think I'm stuck at 4.7 Ghz. While I'm well within my temperature threshold (66 vs. 80 degrees C), I've hit the voltage wall that I'm comfortable with (1.35 Vcore).
The next line of investigation is going to be geared towards minimizing temperature. I will be looking at what fan types and arrangements strike a good balance between low temperature and high noise. I may also fiddle with some of the more exotic cooling methods.
After that, I plan to branch out into GPU over-clocking and CPU memory over-clocking.
I edited my post above to expand on my reasoning.
Blitz Krieger said:
Thanks, Blitz Krieger & Hogan773. I have a clear idea of what LLC is now. I have decided not to fiddle with it as I don't want the voltage to go above the threshold I've chosen--even if it isn't for very long.
Blitz Krieger said:
You weren't kidding. The result I described in the post above was fantastic.
Blitz Krieger said:
This is a good suggestion--I'll try this out.
This was a really helpful discussion. One of my main interests in fiddling with over-clocking is to learn more about electricity and temperature and this was very informative.Blitz Krieger said:
LiuKangBakinPie said:
What's your reasoning?
Are you still using Fixed Voltage or have you moved to Offset?
I've sort of grooved into a 42x, 1.21-ish Volt "mild OC" that I'm very comfortable using 24/7. It uses Offset Minus -.05. Idle voltage is in the 80s.
I'm still experimenting with some higher multipliers, but it seems there is a growing wall of voltage for each step. While I might be COMFORTABLE having it run at 1.35 for a time, I'm still skeptical that, for me at least, the extra speed of 45x vs 42x is worth running at a significantly higher voltage, even if it SHOULD be safe. Reality is this CPU is super fast already for me. Perhaps if I have a big video encoding task or something, I'll crank it up just for that. But for day to day stuff, I'm not likely to notice 42x vs 45x, and if there is ANY chance I'm shortening the life of my chip below my timeframe, which could be 4+ years potentially, then I'll be a voltage pansy on a 24/7 OC.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter