*** Official Clevo P770ZM / Sager NP9772 and P770ZM-G / Sager NP9773 Owner's Lounge ***

OK so basically what you are saying is use that value (the 1.105V) Multiply is times the stock multipliers of (44,44,43,42) and that will give you the exact voltage for each core? I always thought the multipliers refereed to the multiplier used based on number of active cores.

 

NO, give me a second to update this post with more info.

Edit:

Read this guide for a rough idea on what to do. Do reverse to underclock and undervolt
http://www.overclock.net/t/1411077/haswell-overclocking-guide-with-statistics

See this for multiplier and voltage table half way down:
http://www.anandtech.com/show/8227/devils-canyon-review-intel-core-i7-4790k-and-i5-4690k/2



Look for the CPU core voltage amount in the bottom beside the graph. If your's doesn't show it, then click the wrench in the corner of that area and add it!!! This is my stock, not my stable underclock/ undervolt.




This is my stable undervolt at 4GHz without speedstep and turboboost disabled. 1V to CPU. Your results will differ.

 

@ajc9988 thank you very much for that I'll read into it and learn as much as I can. If you don't mind take a look at the pdf I posted and tell me what you think.

 

Great idle temps, but scant on info. Also, use xtu stress test, not benchmark, to get a better feel on stability!

 

One of the PDFs show the stress test but it was after repast. The GPU benchmarks seem slightly above average as well and excellent for laptop graphics I'm impressed

 

Once you fully tune this laptop, you will be amazed.! My stable gpu overclock matches your firestrike score (100 less about), and I have a 970m! Learn as much as you can then tune it properly m treat it well, it's your new beast of a baby!

 

Wow very impressive coming from a 970m. I'm trying to take it one step at a time. I don't expect overnight results but yet a steep learning curve. I find this kind of stuff fascinating so it is much easier for me to want to learn. First the CPU then the GPU. I'm doing a lot of reading on setting multipliers to 40 and eliminating turbo boost and if I want that or not. Also am reading a ton about how to fine tune everything. You guys have been a lot of help and that is much appreciated. Out of curiosity are you using prema mod for your GPU? When you did the vbios update did you uninstall video drivers? If so which ones?

 

I am. It blocks furmark and kombuster. download DDU, a fresh stock 344.75 driver from Nvidia, then the bios mod of choice. Boot into safe mode, run DDU with the box checked to disable windows auto driver search. After ran, reboot. On windows boot, go to control panel, device manager and disable display adapter. Flash following the instructions given by the person that modified vbios. Reboot. Enable display adapter. Install driver. Use your preferred overclock utility. Prema locks max voltage at 1.2, which you shouldn't need more than that.

 

Once you boot into safe mode and open the software there is not an option to disable windows auto driver search. I'm guessing perhaps this is done by default? Once it is open I would assume that I should click "clean and restart" then follow the rest of the instructions?

Don't worry I figured it out. Think it installed without a hitch. I usually use an idea inspector to do GPU over lock. Is that the best to use or is there something else out now that is more efficient?

 

I can tell you straight that under certain load that speed will throttle. Heavily. BUT for gaming it should not be much of an issue at all.

 

You're wrong. Paste could be better, but it doesn't throttle prime 95 large fft. Even at xtu stress test and wprime in mid 70s. Now small fft in prime 95,yes, you are right.

At 4.5,in xtu it throttles at most 20hz. Just runs high idle which means when I Crack it open this Friday, should do better plus when switched from mx-4 to gelid, should help. Read the p750zm thread with me and jaybee83 race to 4.5. I on Avg. Get 3 to 4 degrees more than him (6 in rare instances).

 

It's default checked, but i like to mention it. It is on there, but you have to look for it.

 

Haha you guys are trying to get the CPU to run at 4.5 no turbo? That's cooking!

 

Pushing the thermal limit. It won't see that again until delidded next month!

 

I wasn't talking about your temps... I mean that some load requires more voltage than you are feeding the chip, and the chip WILL throttle under heavy enough loads in those instances. But for low-voltage stuff like gaming or most of those stress tests? You're fine. I've seen my chip use up to 1.192v for some actions, but most of the time normally it's under 1.1v easily; even lower if I reduce my core voltage (I have it at -50mV at stock speeds, but at an OC it'll BSOD instantly. I haven't done checking to see how low I can reduce it at my OC'd speed yet).

 

trying? already done

Sent from my Nexus 5 using Tapatalk

 

That is autofan. I don't cross 1.17 until I am 4.4 ghz under extreme load (XTU is 1.124 at 4.4ghz, don't think I tried to prime qualify it b.c of thermal throttle, but did do BurnIn Test, wprime, and XTU, will prime qualify after delid) It's a close to gold standard chip!!!

 

You're over 1 volt though... XD

 

That is my 4.3 setting with autofan!!!
And I took the screen shot on the hottest fluctuation!

Edit: being playful with the emojis, I mean no offense! Just happy at my chips abilities. GO VIETNAM!!!

 

Impressive stuff coming from a laptop. What kind of temps were you hitting and did you tweak fan speeds or leave that alone?

 

Autofans and it is in the pic. 83Deg. hottest at 4.3 in XTU, wprime is about the same, but that is too hot for me to run all the time. I like the 4.0 which is much cooler. I had it prime stable at 4.0, but it was crashing due to me being a tab whore in firefox at 4.0 ghz, so I upped the voltage from prime stable to 1V at 4.0. Never crashed at 4.0 because of that again!

 

Went with Ballistix 1866 CL 10-10-10-30.

1866 is the sweet spot for this machine. Latency matters more than speed. 1866 CL10 latency is lower than 1600 CL9 and 2133 CL11.

 

Ok one more question. I read through everything that was given to me as well as some other things I found on Google. I like the work flow that you laid out for me as it makes the most sense. I plan to start fresh from default. Lower the cache voltage offset until unstable then raise the voltage in 5mV steps until stable through the 5 minute stress test in XTU. Once completed I will then move to the dynamic CPU offset lowering that until unstable and increasing as before. Once finished I should have an idea of lowest temps and core voltage for stock multipliers. Is this the correct work flow? Also XTU wants to automatically set the cache ratio to 40X despite the default value of 44. If the above is correct then I understand all of that but I'm not exactly sure what this value does or what is the better of the two settings. Hopefully all of this makes sense haha. Again thank you guys for having patience with me while I learn the ins and outs you guys are awesome. If you guys ever get into the tv calibration regime hit me up and I can return the favor haha

 

Sounds like I'll have some questions for you then! !! First,after cache value is xtu stable, perform prime 95 stable for large fft and ram in prime 95. This is to rule out future heavy stress crashes on this value. It is set to 40 because the processors stock value is 40.also, if you do not plan on using a core multiplier higher than 40, turn off turboboost if you would like, and speedstep if you don't want it to throttle lower. The offset must work at all throttle states and turbo states. But you cannot go above 4.0 without speedstep and turboboost enabled in XTU. This can allow for more stability also, but also more idle heat. After that, then move onto core.

Edit: many say you want cache and core the same value. But you can squeeze more out of the core of you don't to manage heat. The cache offset won't change (or will be increased by 5mv if worried about stability with core at higher clocks). Then just clock away on the core!

 

Ok perfect. According to XTU the default value is 44 not 40 which is why I'm confused. I plan to keep turbo boost on and leaving the multipliers alone. I also don't plan to over clock as I don't really need it. I'm mainly just trying to reduce temps and wasted energy as much as I can while leaving the speed stock.

 

Then if a lot of heat is generated with prime 95, do a low under volt of-30 to -50 on core and use function key + 1 key to turn on max fan to handle heat of prone while checking stability

 

Got ya and thank you. But 40 is correct and not 44 for my case?

 

For cache, yes! Leave it at 40 and never change the stable undervolt on it (unless more stability at higher overclocks on the core, usually 4.5 to 4.8 and above).

Nice comparison of 3 4790Ks:

http://www.tweaktown.com/guides/648...pec-sr219-cpu-overclocking-report/index4.html

 

I was actually just reading that and one other much like it. So I did some reason about the cache frequency and I'm getting missed readings. I know you say leave it at 40 but I read others saying to leave it set to the highest multiplier. If my multipliers are 44, 44.43,42 I should still use 40? I'm getting stable -200mV cache frequency. That just doesn't seem right haha

 

i have to disagree: http://www.anandtech.com/show/7364/memory-scaling-on-haswell/10

unless were talking about huge latency steps, frequency is ALWAYS more important than timings...

and yes, im aware that article covers desktop ram, but since were dealing with the same cpu arch it doesnt make any difference

Sent from my Nexus 5 using Tapatalk

 

-200 seems a little high. That first link I sent you showed the comparison of having a 1:1 core to cache ratio. Ideally it is one to one. But it also generates heat. You will get faster benching with higher core multiplier and stock cache at 40 than you will get if you put both at 46 or 47 multiplier. The extra heat from it can lag the core going faster because of thermal throttling. We have a very compact machine with low thermal head room. So, to allow more for the core, you lock the cache to 40 and find the lowest stable with it. Now that said, you have an incredibly high cache undervolt. I get -118- -119, use -115 for benching, run on everyday at -110. I use -110 at 45 multiplier core also for a little extra stability in case it needs it. HTWingnut gets about the same. Jaybee gets -145. I'd say, start undervolting the core now and see where it goes. I cannot help again until Friday. Goodbye community, I shall miss you!

 

wow at the heat generated with prime95 Is this safe? These are my numbers currently. It is staying consistently at 96-97 with max fans. Should I continue the test or undervolt the dynamic more to help with heat first?

 

@ajc9988 thank you for all your help. You will be missed. You have been invaluable in my journey of unleashing the power of the batman

 

I'd try to get 24/7 temps to around 85C or below if at all possible for longevity.

Edit: prime95 - esp. with small fft is a power virus, very few actual workloads approach those values. So test with your own workloads too...

 

Zymphad is correct, if you only look at CAS, but things are more interesting under the surface.

I am not criticizing Zymphad, as a firmware engineer this is a big debate area for us engineers, I wanted to show how looking at all of the timings can change the viewpoint.

In order to do that I took some time to crunch the numbers:

My criteria for module selection was te DDR3L 204 pin modules available from Newegg today 2/23/2015.
I found the following.
1600 @ 9-9-9-24
1600 @ 9-9-9-28
1866 @ 11-11-11-32
1866 @ 10-10-10-30
1866 @ 10-10-10-32
2133 @ 11-11-11-34
2133 @ 11-11-11-31
2133 @ 11-11-11-27
Now to calculate the time for each tick:
10e6=10000000
MemoryClock = half of the advertised speed for DDR so 1800 = 900
1/((10e6 * MemoyClock)/2) = tick

With that, each speed comes out to:
1600Mhz tick = 2.500e-10
1866Mhz tick = 2.1436227224008574490889603429796e-10
2133Mhz tick = 1.8752930145335208626347866854196e-10

Now the fastest CL time for each speed:
1600 CL9 = 1.9292604501607717041800643086816e-9
1866 CL10 = 2.500e-9
2133 CL11 = 2.0628223159868729488982653539616e-9

But now comes the next number, RAS this is the number that tells us how fast they can be expected to finish any request given.
This timing changes a lot more than CAS and speaks to how long we have to wait to request the next operation.
This is the number that can be tweaked the most IMO and will have a impact on performance.
1600 RAS24 = 6.00e-9
1600 RAS28 = 7.00e-9
1866 RAS32 = 6.430e-9
1866 RAS30 = 6.859e-9
2133 RAS34 = 6.375e-9
2133 RAS31 = 5.813e-9
2133 RAS27 = 5.063e-9

As you can see a CL11, RAS27 2133 memory module finishes 0.94 microseconds before the fastest 1600Mhz Cl9, RAS24.
This gives the 2133Mhz a 0.807 head start getting to the next request.
That is a Huge difference and makes the 2133Mhz CL11 in effect have a relative CL of 1.125e-9.
This is just one example and you would need to re-do these calculations for each comparison but I hope this gives an insight of how memory timings work.

Zymphad is correct, 1600Mhz can get to the data the fastest.
But that does not mean that 1600Mhz is the clear winner.
If someone has a different opinion or point of view I would love to hear it.

 

Yes with normal workloads I don't even come close to those temps. I am only using it to test the stability of the under volt. If it handles prime95 it should handle anything

 

Yeah. My favourite stress test workload is FAHbench, but yeah, for stability testing, prime95 for sure...

 

I'm 45 minutes in and so far everything is stable. I'm running prime95 small and able to achieve really good undervolt values. Right now my values are dynamic cpu offset of -90 cache -160 system agent -60. Even with the large offsets and repaste I'm hitting temps in the 98 range running stock multipliers of 44,44,43,42. Durning the run I'm maxing out on power limit throttle from time to time and bouncing around from 0 up to 56% thermal throttle but constantly around 20-30. I know @Jaybee83 was able to hit low cache values as well. I'm interested in your temps at these values using stock multipliers. I'm also interested if you adjusted analog and digital offset at all. I read earlier that it could cause an increase in temp instead of a decrease.

 

I've been looking for this panel: LP173WF4 since it's apparently approved for use in the P770ZM with the latest BIOS: 1.03.09. But all of the panels I see have different connectors. Is there some sort of adapter used? Anyone have any insight? I'd rather pay $70~ for this panel than 200-300 from some shop that sells Clevo stuff.

 

I haven't really paid attention since I made the 980m upgrade at the very last minute. What are y'alls 970m idle and peak temps usually. My 980m idles ~44-45* and peaks around 62-63* running seti@home with occasional spike to 65*. This is after gelid repaste. Add about 5-7* pre repaste.

 

Stock OC'd 980m and I have same temps as you with same type of paste.

 

Lol. I'm curious about 970m temps. Wondering if they're significantly lower. Still can't believe these cards run so cool. Cooler than my old 540m easily.

 

also goes to show how powerful the cooling system is in our ZM machines 980M gpus run way hotter in SM models for example, reaching mid to high 80s....

on another note: un-f***in-believeable!!! it seems like nvidia has started shipping the latest maxwell gpus with a vbios version that blocks overclocking....

Link: http://forum.techinferno.com/genera...t-overclocking-mobile-gpus-12.html#post124474

to anyone who hasnt signed the petition in my signature yet: NOW is the time to do so!!!

Sent from my Nexus 5 using Tapatalk

 

Yep, can confirm my stock bios was nice and locked when I got it. That might be my bios that Prema confirmed against.

 

Same here, I'm happy user of new NVIDIA policy. It's time to think about modded vbios.

 

unless, of course, they "update" their drivers to block all modded vbios versions, then were f***ed

Sent from my Nexus 5 using Tapatalk

 

My 980m is unlocked. Let´s see if my new P751ZM will have a locked one.... :/

 

now that would be just wonderful, especially in your case samot! u have the freedom of choice between a gimped machine with extreme lightbleed, super loud whining and warped heatsink or a properly functioning machine with an oc-blocked vbios

Sent from my Nexus 5 using Tapatalk

 

oh the irony...

 

Are y'all saying the vbios is locked down where you can't even do the +135mhz or mess with memory? I'm still on 344.75. Do I need NVIDIA inspector to check?