Undervolting in XTU, best configurations for i7-4710HQ

I'm looking for the best undervolt configuration for Intel i7-4710HQ, I would like also to compare with yours. Currently I'm trying with -0,085V for dynamic CPU Voltage offset and -0,075V for Procesor Cache Voltage Offset and turbo boost power max for 60W. I've tried stability tests in furmark+cpu stress test in xtu simultaneously, 3dmark, metro 2033 benchmark, Rome II benchmark, Resident Evil benchmark(after lowering cache voltage from -0,085 to -0,075 seems to be ok) and only in Heaven Benchmark I've have bluescreen in windows 8.1 64 bit. Is it something wrong with unreal engine benchmarks, anyone had similar problem? Please write your fully stable configurations for this processor.

 

I have a P35G V2 with the 4710HQ.

I've set Dynamic CPU Voltage Offset to -80mv

I had it at -90mv, and it was stable in benchmarks and DOTA2, but it crashed 5 minutes into CS:GO

The rest are default settings

It helped my temps quite a bit, especially with Turbo disabled (I still get 60fps on everything) and the fans don't come on as loud

 

Did you try to get also the Processor Cache Voltage Offset down?

 

I have a 4710MQ with -80mv on CPU and other voltages set at -100mv. I can use the extra 2 multipliers for the CPU when undervolting, so up to x37 instead of x35. If i set the CPU to -90mv then I get random rare crashes in games so I leave it at -80mv. This has allowed me to get higher clocks under load whilst gaming. I have had other voltages undervolted by alot more, eg -130mv on iGPU, or I can set the iGPU to 1250 clocks on stock volts. The other voltages seem to have no real benefit for temperature reduction so they are not worth worrying about.
I have found it best to leave Windows 8.1 on Balanced power profile otherwise you can see throttling on the current in XTU if in Full power mode. Setting the current limit above the stock 70A reduces this throttling but I don't think it effects performance in any noticeable way.
So my advice is to just set the cpu to about -80mv and power mode to balanced. If you have time to test the other settings then set around -100mv.

 

I have a 4720HQ which is prime95 + intel xtu stable at stock clocks with -70mV, but running heaven caused a bsod after a few minutes. Currently testming if prime95+heaven is stable at -60mV, and after 10 minutes or so it seems fine. Trying to see if -65 will work.

 

Just a matter of fact: each cpu will be little different, mine is set at -85mV and works fine so far, but I'm not sure if it's even worth to bother with lowering cache voltage at all. This is not the first time I'm undervolting CPU, but I remember few years back, when undervolting core2duo, with normal voltages and large overclock C2D would run fine until about 95c and then start throttling, but when undervolted and overclocked it would run fine as long as temperatures were bellow 80c, but once you I hit 80c it would start crashing. I wonder if this is what happens to OP and if improving cooling a little bit wouldn't solve his crashing. The easiest way to test it would be to monitor temp. and maybe take it outside and run benchmark in colder air, if no crash then OP could repaste CPU or lower temp. some other way and keep it at -85mv. But keep other voltages close to stock, since not much benefit and less variables to go wrong, it could be cache crashing not CPU even at lower undervolt.

 

I seem to be one of the luckier ones.

On my 4700MQ, I have a stable -100mV dynamic UV, never had any stability issues. Haven't tried pushing any further, maybe there is a tiny bit of room left.

 

Im going to try to run my 4720HQ at lower clocks and see how far i an UV then. Maybe i can hit -100mV at 4700 speeds too! All the 4720HQ is is a better binned 4700 with a more refined manufacturing process, so it theoretically should be a beter chip.

 

Interested to see what you can achieve.

 

I also have a 4720 but I've never used xtu before.
So dynamic cpu voltage is the only parameter that I need to change for undervolting? Also how do I monitor whether the undervolt took effect. Ty.

Sent from a Galaxy far, far away

 

Look at the cpu volts, before and after, but since voltage is dynamic, best to check on idle. Another way, get hwmonitor, write down max/min cpu voltage, undervolt, reset monitor and look again. No matter what, after changing volts, make sure you press apply (it will be yellow after change, but before apply) and it will work. You have to redo all xtu settings after computer crash/blue screen, since they'll be reset for safety, otherwise new settings will "stick" even after reboot.

 

First things first... you all might need to raise your current limit (if that's even possible for you; there's little way to tell) if you're undervolting so heavily. I don't know if it'll help in heaven, though.

As was said earlier in the thread, your chip is your own and your mileage may vary. My first 4800MQ could do -80mV offset at 3.9GHz; my current one will BSOD at -10mV at 3.9GHz

The setting I've highlighted does nothing. Your CPU will never pull over 57W under any circumstances. Read my mobile i7 CPU guide for more information.

Nothing is "wrong" with unreal engine benchmarks. It depends on the load it puts and whether the CPU can handle it with the voltage. See when I said I'd BSOD at -10mV above? Idle/stream watching/etc doesn't crash it, neither does XTU stress test or rendering videos in Handbrake... what crashes it is Hearthstone for some reason. If *ANYTHING* causes you a BSOD, you need more voltage. No getting around that.

That statement isn't the best thing to do. You should raise your current limit instead of trying to gimp the CPU which is already "gimped" (though manually) by an undervolt. Set your current limit to 95A and forget about it. Also, please note your 4710MQ is *FAR* different than a 4710HQ. There's all sorts of limitations baked into (we speculate) the EC of all machines using them (as dictated to do so by intel) to screw with the power consumption and such of the machine. I don't even know if it's possible to properly raise the current limit on those things.

 

Will undervolting lessen the power draw of the chip?

Sent from a Galaxy far, far away

 

Yes, however it also causes more amps to be pulled. If you undervolt a lot and notice current limit throttling, you will need to raise the processor current limit. If raising the processor current limit does not help, then your motherboard/CPU cannot raise the current limit, and you would need to lessen the undervolt until it does not current throttle anymore.

Less voltage also reduces the heat output, but if you drop voltage too low and the chip needs too much amps, temps can start going back up.

 

Whole idea of undervolting is to lower power draw of cpu. Dissipated power for cmos gates = c*v2*f, where c=capacitance, v2=voltage square and f =frequency. The formula is not exact, due to current leakage and additional circuits inside cpu, but is close enough. I'm not sure what D2ultima is talking about when he said and I quote: " First things first... you all might need to raise your current limit (if that's even possible for you; there's little way to tell) if you're undervolting so heavily", which would suggest when you lower the volts you would increase current, but that is wrong, decreasing voltage will decrease current more or less, to the point that when voltage is zero no current will flow.
When you undervolt your cpu you will get less current throttling, but you still get, especially when overclocked, since 47w TDP on all i7, is not realistic especially top end models.

 

Lowering voltage means the CPU needs to draw more current to perform. Higher currents usually do not affect heat as much as higher voltage does, so the tradeoff is USUALLY reduced wattage and reduced heat. It is possible to undervolt TOO much, such that your current limit (let's say 71A) is too low, and the CPU begins current throttling. If you can raise this, then fine.

It is also possible that after undervolting beyond a certain point, that current required gets so high that temperatures start going back up (for example: -80mV on a chip with the current limit slider maxed out might grant you 80c in test X, and -120mV, while stable, might grant you 85c in test X, because of how much more Amps of current is required for the high clockspeeds).

 

d2ultima, I don't like to argue, but I think you're wrong on this one: I did some testing with xtu, when I undervolted -100mv and run benchmark, I had no current throttling at all, as soon as I overvolted by 25mv i immediately started having current throttling during same benchmark, everything else being the same. Modern cpu are very complex and Intel build a lot of safeguards inside, maybe something else is going on? I know for a fact that if I max my CPU on frequency and run some heavy duty benchmark, my CPU will not stay at maximum frequency for long, it will throttle down to lower frequency despite no thermal, current limit or power limit throttling (at least XTU doesn't show it). Setting lower frequency will make throttling down happen less often and therefore higher score, but also could make maximum temp. higher etc. Something similar may be happening to you, at larger undervolt CPU is throttling less, working harder and showing higher temp. But it doesn't mean undervolting itself caused larger current being drawn, it could mean undervolting allowed CPU to do much more work and that caused higher temps. I know undervolting works, because I do it to every computer I own and I always get cooler running and longer battery life.

 

Are you current limit throttling or power limit throttling? Because power limit is something else.

Set your turbo boost power max to 100W and overvolt and see if you "throttle". Of course, short power max must be set to 100W as well.

Higher voltage = higher TDP = chance to hit TDP limit easier.

And you weren't running "XTU's benchmark" you were running its "stress test", unless you have an extremely old version of XTU, because XTU's benchmark (NOT its stress test) will current throttle anybody who has current limit set under 256A.

 

I was running xtu benchmark and I have current throttling specifically displayed on the graph, so can't miss it, maybe it was throttling but not showing as such, I don't know. I have simple test in mind, which can prove the point, but kids won't let me do much now, so later I'll post more details.

 

You know... actually... if you have a 4700MQ... you probably wouldn't see current throttling in XTU bench.
My 4800MQ throttles to about 3.3GHz, so theoretically if you are below that... you should try OCing to 3.4GHz and seeing if you get current throttle and see.
I completely didn't think about what I usually throttled TO rather than the fact that the throttling occurred.

But rest assured, it does happen. Everyone who sets their CPU above that limit sees it, I'm sure. With the 4710MQs (especially with their turbo bins maxed out) and the 4720HQs that are rampant now (3.4GHz default 4-core turbo, 3.6GHz max OC) they're likely to be more apparent with the current throttle. There's also the fact of what your default current limit is. There was a user whose default limit was 53A, and my model of machine comes default at 71A for it.

 

So here is simple test I did to see how much undervolting decreases current: I open XTU and set multiplier to x26 on all cores. The reason to do so was to eliminate any kind of possible throttling and make CPU work at steady pace, then I undervolted cpu by 100mV and run XTU benchmark. My results were: 456 points, max frequency 2.6GHz, max temp. 70C. I also took few screen shots, most of the time at full speed my voltage was 0.773v and CPU total TDP 24W. CPU slowdown few times, but I think it was to load new set of tests. Then I reset voltage to overvolt by 40mV and rerun the same benchmark again with all other settings the same. This time my score again was 456, 2.6GHz max frequency and max temp 81C. Most of the time at full load my voltage was 0.865 and total CPU TDP 32W. The simple formula says Power=Volts*Amps, so when undervolting my CPU current was about 32.7 Amps (24W/.733V) and when overvolting about 36.9 Amps (32W/.865v). So when you undervolt your current drawn is less as well, but even I was surprised that increasing CPU by about 140mV, which I did, will increase power draw by 8W and increase temp by 11c. You should be able to do same test on your machine and have similar results, undervolting does not increase current, it decreases it. Now, running this test at much higher frequencies will invoke a lot of different throtling and make whole thing more confusing, but principle stays the same. Sorry for long winded post, but I hope it's clear.

 

What you've said makes sense, but hear what I'm saying. You're using "power=volts*amp" right? What about "amp=power/volts" as a re-writing of the statement?

Think of this. What if at 1v flat, to do task X, I draw 45W on my CPU. Then I would require 45A of current, no? Now lowering volts should lower that TDP. But if I lower volts too much and TDP required doesn't go down proportionately... then I'm going to require more current. What if I dropped my volts to 0.9v, but my TDP only hits 42W. Then I'd have 46.7A of current required. There's been people on this forum who have had current limit throttling show up after undervolting a whole lot, where it isn't there at stock.

BUT there's also been people who've had it show up at stock and it went away with a small undervolt. I was speaking from experience when I spoke about requiring more amps.

Blah I wish I knew more/had a better way to test.

 

Now I see where are you getting confused: When the system is getting pushed to the limit it will throttle, there are at least 3 throttle down scenarios: current throttling, TDP throttling and temperature throttling. In your scenario at 1 volt cpu is TDP throttling at 45W-47W, to do so it will lower frequency. When you undervolt, now CPU is not throttling on TDP anymore, so it runs at higher frequency and now will pull more current and instead of hitting TDP limit, it hits current limit, but this is due to running at higher frequency and doing more work, not because of undervolting itself. Run the test I described and you'll see for yourself, but if you run much higher frequencies, the throttling will confuse hell out of you. But it doesn't matter, undervolted CPU, even when current throttling, will get higher scores and do more work, than overvolted CPU. I'll give you an extreme example hopefully to make it more clear: if you overvolt by let's say 500mV to 1.5v, you'll hit 47TDP at about 2.5GHz and no matter what you do, at full load CPU will not go over 2.5GHz, because it hit the power limit despite current being only 30amps. Go back to 1 volt and your CPU will run 3.9GHz, 47 TDP and 47 amps, but it will be almost 50% faster. Due to throttling, sometimes you'll get higher scores and do more work at lower frequency, than higher frequency, but principle always stays the same, everything else being equal lower voltage=lower current, always.

 

Please consider checking out this thread for my current results. And to answer my question haha

http://forum.notebookreview.com/threads/minimum-haswell-cpu-clockspeed-for-3k-60fps-gaming.776158/

 

I can't find CPU current limit setting in XTU, where can I edit it? I have a P35x V3

I dont notice current limit throttling, but I do notice Power limit throttling when using aida64, even when turbo boost power max and short power max are set to 100W. total TDP goes to 58 watts sometimes, so the laptop seems capable, but after (i suspect) 28 seconds it throttles to 47 watts. How can i fix that? Also, is it dangerous to raise the tdp limit above stock?

 

Read my mobile i7 CPU guide for the answers to all your questions. It's in my signature.

Your BIOS is locked and won't let you see/adjust the current limit, is a likely thing.

 

Not all laptops have current limit setting, mine doesn't. Yes it could be dangerous, if the power supply is designed for 47 W and you set it to 100w it could burn up, but as you already noticed, hopefully it will not actually set itself above what it is designed for, even if the program setting says otherwise. There is some safety margin build in, where you could push it a little more, but I don't think it's double the rating.

 

What? No, my example was using made-up values.

In my actual experience, when I undervolted and ran Linpack LinX benchmark, I would draw 80W on the CPU and sit at 3.3GHz. When I raised voltage, my clockspeed would increase in the benchmark, and I would draw more power. In particular, 1.086v (the voltage I checked with, compared to my undervolt of 0.9927v) would let me sit at 3.5GHz and draw 94W of power.

In both instances, current limit was set to 100A, and I was current limit throttling. That's MY experience specifically. Others have had experiences where undervolting beyond a certain point caused them to experience current limit throttling, and they had to raise their current limit to hit their default turbo boost clockspeeds.

His chip will never pass 47W under extended load because he has a HQ CPU.

Setting it to 100W does nothing. It's simply a limit. If he was able to raise his power limit and he managed to overdraw his power brick, he'd simply shut off due to lack of power from the brick. If he couldn't cool the extra power, he'd hit 95c and start thermal throttling. Raising wattage limit does very little in the way of harm. Some programs/loads exceed the designed TDP for the CPU however, and would benefit from a TDP increase, such as rendering, or gaming + livestreaming, or certain benchmarks, etc. It's a problem we have with the HQ chips, that they're all TDP locked by design. You really should read my mobile i7 CPU guide. Like I told you to read before. >_<. You'd understand all of this already if you had.

 

I read it after I posted lol. So the 4720HQ is hard limited in the chip itself to 47W and I can't change it at all? Seems strange since my PSU is rated for 180W, so that should be able to handle cpu only tasks just fine. I cant imagine the cpu going above 47W when gaming either so thats fine too. It's not limited by my bois or motherboard either, right?

 

It is possible that Intel demanded that EC be locked to 47W for all HQ chips, but I don't know. Even on the GT80 Titan those chips are locked. You cannot change it.

 

Bummer.

Haven't checked whether the chip is power limit throttling during normal tasks like gaming. Did you notice your CPU going above 47W when clocked at 3.2-3.6 GHZ?

 

At 3.5GHz at stock voltage I sit at about 60W rendering videos etc. With an undervolt I hit about 50W doing x264 rendering in handbrake. I've passed that limit while streaming but I haven't streamed in a while. Evolve can use a decent bit of CPU power but I have unlocked framerate, so.. that's a different thing. But if you ever plan to game above 60fps or anything in a game like say... BF4? Good luck =D.

 

Wow, so if I'm able to eliminate the thermal throttling with a repaste I might run into power throttling when gaming at 3.2ghz?

 

Unlikely if 1080/60fps gaming, but it is indeed possible. Undervolting generally helps this TDP issue though.

 

I find it atrocious that intel is selling these chips as capable of running at 3.6ghz, when clearly that is only possible for 28 seconds, just long enough to render an extremely short video or unzipping a large file

 

The limit is actually ~2.5 minutes at max. The 28s may or may not work for you; it doesn't work in MSI and Clevo machines.

Yes, it's annoying that Intel sells those chips so gimped.

 

Newb question, do I need to have xtu running for the changes to have effect?

 

No. You can right click it and select "exit" XTU from the system tray and it'll have applied. The XTU service is likely always running though, so that if you BSOD it can reset your settings to default, so that you don't go into a BSOD loop if say... you dropped voltage too low.

 

So I don't need to start up xtu every time I turn on the notebook?

Also, is the reset default after a bsod an auto feature or do I have to set it up?

Sent from a Galaxy far, far away

 

Correct

Automatic. You can't turn it off actually, I think.

 

I've save like 8 different profiles on the xtu, how do I delete the old profiles after my new profile is sufficiently stable? Stupid question but I can't find the option.

Also what utility to use to find the gpu temp?

 

You can enter the profiles tab in Xtu and remove all the profiles you do not want. Select the profile you want to remove and click the x that appears. Then you get the choice whether you want to remove the profile or not. For monitoring of GPU temperature you can use Hwinfo64, GPU-Z and Hwmonitor + other similar software. Use the best one; Hwinfo64 because you can monitor almost everything of hardware in your computer. http://www.guru3d.com/files-details/hwinfo64-download.html But everyone should also have GPU-Z installed whatever other software you use for gpu temperatures. http://www.guru3d.com/files-details/gpu-z-download-techpowerup.html

 

Can you link the guide please

 

It is in my forum signature

Sent from my OnePlus 1 using a coconut