My experience with overclocking the 6820HK and Grizzly's 'Kryonaut' thermal compound.

[Ocmersh]

I wanted to share my experience with overclocking the 6820HK in the Aorus x7 v6 laptop, and my results of re-pasting the thermal compound on the CPU. I also did the GPU, but it already ran cool enough where it made less of an impact.

 

I originally shot for an overclock to 4.2 GHz, and while I was able to make and maintain that, the voltages were far to high for me to be comfortable with, and the temperatures were also very high. I also had stability problems, as the processor cache liked to be within 100-200 MHz of the core clock, but would get unstable above 4.0ghz.

 

So I dialed back the voltage, dropped the overclock to 4.1 GHz, and kept the cache multiplier to 40x. Also, I think it is important to note that although I don't understand what Iccmax is on the processor, it did greatly impact performance. In my XTU settings, 65A is the conservative power-saver default, 95A is the 'turbo' default, and 100A was my overclock setting. Anything over that simply generated more instability and heat. It seemed that when Iccmax is reduced, the cores do not actually reach their desired frequency, or are unable to hold it under load. (i.e. hitting only 3.8 on all cores on a 4.0 GHz overclock)

 

Power limit was maxed (to the power limit that the brick can provide of 230 watts) a +.030 V offset, and a multiplier of 41x on all four cores. To those interested I also overclocked the GPU using the same methodology of finding the ceiling and stepping it back a notch. I achieved an offset of +145 MHz on the core, and a +885 MHz overclock on the memory. It is a mobile GTX 1070. I'll add that I think I got lucky on mine, with the overclock it scratches around 2000 MHz and performance is great on it. (I get 80 FPS max settings on Witcher 3 ).

 

With 4.2 I was thermal throttling all the time, hitting the 100C ceiling on synthetic loads. Backing down to 4.1 and lowering the voltage helped with temps around 80-85 under a "real-world" load, but temps were far to high for my liking. I did an Aida64 stress test with the FPU instruction set, and I again was getting low 90's with spikes into 100C, throttling the cpu again

 

So, I re-pasted the CPU and GPU (while I was at it, temps were 65C under-load and not a concern), and the new compound worked very well. Temps in the mid 80's with a synthetic load, and mid 70's with a 'real-world' load and other synthetic loads that aren't as hard as aida64. Ive posted the results before and after below, and although not perfect, much more manageable. I did still see some spikes into the 90's which triggered some minor throttling in Aida64, but none ever reach 100C. With a working temperature in the mid 70's under heavy load, I think this is much better, especially with a decent overclock on a laptop. Thanks!

 

Oh and the GPU temps improved by a whopping 1C. Final side note, this was a replacement for my desktop, as I wanted my gaming but needed better mobility. So I sold my beast of a machine and got this about 6 weeks ago. I purchased this from Newegg as open box during one of their open box sales.

 

The benchmarks and the before and after temps using aida64 after re-pasting the CPU and GPU with Kryonaut:

 

 

 

BEFORE:

 

 

AFTER:

 

[genexis_x]

@Pendragon what do you think

[Nikola Gh]

how much do you get in firestrike normal?

curious because your system looks pretty similar to my SC17 1070

[Ocmersh]

7 minutes ago, Nikola Gh said:

how much do you get in firestrike normal?

curious because your system looks pretty similar to my SC17 1070

Here is one on firestrike (regular)

 

[Nikola Gh]

14 hours ago, Ocmersh said:

Here is one on firestrike (regular)

 

i see... very similar to mine

[phongle123]

55 minutes ago, Ocmersh said:

It is a mobile GTX 1070

Mobile Pascal cards are about the same as desktop variants. To keep them cooler, the clock speeds are lowered and they gain additional CUDAcores over the desktop variant to make them match the desktop performance. However, with the mobile 1070, being able to reach 2000Mhz is really good. That's on par with overclocked desktop variants and since mobile Pascal are clocked lower, meaning you got an even higher overclock than many desktop out of box variants. Meaning the overclock at the same clock in comparison, you should be getting more performance. Good for you.

[Pendragon]

16 minutes ago, Ocmersh said:

Here is one on firestrike (regular)

 

several things you should do.

 

CPU wise. I think u have a terrible chip there RIP. I'd say keep it even at 4.0ghz for rock solid 24/7 stability. Try to lower voltages more. Seems high. XTU could be the issue also because XTU is fucking terrible. I'd highly recommend you redo all ur CPU stuff on Throttlestop. Make sure you clear the shit out of all former XTU settings before using Throttlestop. You'll have far better control that way. 

 

GPU wise. Get MSI afterburner. Try to rerun those OCs running Unigine heaven + occt on loop. I'm 100% sure those clocks don't stay because the 1070 is TDP limited that way (i have a similar system). How you get the most out is getting MSI afterburner and set custom voltage curves. The goal here is to get as low as a voltage as possible with the highest clocks possible. The 1070 is a far more efficient chip than stock so you can drop voltage much lower.

 

Otherwise, I would recommend if you want to push your system further is to use liquid metal and replace the thermal pads with some fujipoly ones. At this point its no longer about performance because the laptop is hardcapped at 230w from the adapter but its all about the acoustics. I can have my AW15 run full bore 70C ish give or take and the fans are nice. Gaming its much lower and general use is amazing no audible fans essentially. 

[Guest]

To get lower temps, I'd suggest using liquid metal, this method has been known for significant temperature reductions, allowing more overclocking headroom for the CPU.

[Ocmersh]

8 minutes ago, Pendragon said:

several things you should do.

 

CPU wise. I think u have a terrible chip there RIP. I'd say keep it even at 4.0ghz for rock solid 24/7 stability. Try to lower voltages more. Seems high. XTU could be the issue also because XTU is fucking terrible. I'd highly recommend you redo all ur CPU stuff on Throttlestop. Make sure you clear the shit out of all former XTU settings before using Throttlestop. You'll have far better control that way. 

 

GPU wise. Get MSI afterburner. Try to rerun those OCs running Unigine heaven + occt on loop. I'm 100% sure those clocks don't stay because the 1070 is TDP limited that way (i have a similar system). How you get the most out is getting MSI afterburner and set custom voltage curves. The goal here is to get as low as a voltage as possible with the highest clocks possible. The 1070 is a far more efficient chip than stock so you can drop voltage much lower.

 

Otherwise, I would recommend if you want to push your system further is to use liquid metal and replace the thermal pads with some fujipoly ones. At this point its no longer about performance because the laptop is hardcapped at 230w from the adapter but its all about the acoustics. I can have my AW15 run full bore 70C ish give or take and the fans are nice. Gaming its much lower and general use is amazing no audible fans essentially. 

I do use MSI Afterburner. I used the firestrike stress (where it runs it for like 60k frames or something, it takes about a 20 minutes or so) test several times to verify, and haven't had issue thus far, been using this overclock for a month in various games. It was at +155 on clock and +975 on memory, got artifacts every once and awhile so I turned it down a touch. Though I haven't tried custom curves yet,

 

Ill try TS and see if that results in better control. I was disappointed to see that the BIOS on this laptop doesn't offer any OC control.

[Ocmersh]

3 minutes ago, TheBeastPC said:

To get lower temps, I'd suggest using liquid metal, this method has been known for significant temperature reductions, allowing more overclocking headroom for the CPU.

Thought about that. Worried about conductivity.

[Pendragon]

13 minutes ago, Ocmersh said:

It was at +155 on clock and +975 on memory, got artifacts every once and awhile so I turned it down a touch. Though I haven't tried custom curves yet,

The real question is can it sustain that. because my max OC is like past 2ghz as well, but it can't sustain at that level and drop down due to TDP restrictions. Use HwInfo on the side to test for sustained clocks, not just max clocks.

[Ocmersh]

9 minutes ago, Pendragon said:

The real question is can it sustain that. because my max OC is like past 2ghz as well, but it can't sustain at that level and drop down due to TDP restrictions. Use HwInfo on the side to test for sustained clocks, not just max clocks.

I use HWinfo, and CPU-Z/GPU-Z, it settles around like 1970mhz -1990mhz with peaks into 2005 mhz or so. Max package power on CPU ive seen is 68watts... even though 230 max watts isn't great, that still leaves 140ish for the gpu and 20ish for everything else. Even the crappy gigabyte "OC panel" was getting the GPU to 1940. Just FYI.

 

Ill loop it on Unigine for awhile, I use the valley one. Maybe I'll leave it running overnight just to be sure. Playing with the curve sounds like that could work well aswell, but right now under full load my GPU is around 65c, or around 73c with fans at like 60%.

 

I tried really hard to get it to 4.3 for the chip, but it just wouldn't have it, I am happy with the 4.1 though. Checking out TS now... seems less intuitive but ill figure it out. XTU is annoying in that it randomly changes your settings...

 

End of the day I still feel like I got an OK chip, and a pretty good GPU

[Pendragon]

4 minutes ago, Ocmersh said:

XTU is annoying in that it randomly changes your settings...

exactly. this is the most terrible part of XTU. 

http://forum.notebookreview.com/threads/the-throttlestop-guide.531329/

here is a guide by our own @Eason85. Will help you do everything you need to do with TS. 

 

4 minutes ago, Ocmersh said:

I use HWinfo, and CPU-Z/GPU-Z, it settles around like 1970mhz -1990mhz with peaks into 2005 mhz or so

That's actually pretty fucking solid. Maybe they uncapped the TDP limits on the Aorus. Interesting. Cause on my AW there is a hard cap. I need to bring down voltages to bring it up around 1800mhz 

 

Also be careful of your adapter. Running it at 230w or over is not good for it. Hence why I gave up trying to MAX oc everything, but more like max optimized on everything. 

 

p.s. this should be on the laptop subforum  

[Ocmersh]

27 minutes ago, Pendragon said:

exactly. this is the most terrible part of XTU. 

http://forum.notebookreview.com/threads/the-throttlestop-guide.531329/

here is a guide by our own @Eason85. Will help you do everything you need to do with TS. 

 

That's actually pretty fucking solid. Maybe they uncapped the TDP limits on the Aorus. Interesting. Cause on my AW there is a hard cap. I need to bring down voltages to bring it up around 1800mhz 

 

Also be careful of your adapter. Running it at 230w or over is not good for it. Hence why I gave up trying to MAX oc everything, but more like max optimized on everything. 

 

p.s. this should be on the laptop subforum  

Alright ill look into it thanks.

 

That's a good point with the adapter, i've got a kill-a-watt that I can use to test it, its buried around here somewhere lol. I really just love that OCing and optimizing laptops is a thing now.

 

Thanks for your help.

 

Just did a quick check with valley, but ill leave it running overnight tomorrow or tonight to make sure its all stable. But I ran the 18 scene loop, I am hoping messing with a custom curve I can lower the power draw with the same performance.

[Ocmersh]

2 hours ago, Pendragon said:

 

Also be careful of your adapter. Running it at 230w or over is not good for it. Hence why I gave up trying to MAX oc everything, but more like max optimized on everything. 

 

So i found my Kill A Watt, and during the valley benchmark it was pulling between 180-190 consistently, peak was 200 watts reading from the wall, with over clocks in place. So think power is good, if anything that could be a little extra headroom . I am not sure if there is an efficiency curve to take into account, for example I know that a 900w PSU under peak load will actually pull around 1000w from the wall, as its rating is for what it can put into the system, not pull from the wall. If you treated the adapter the same way and assumed it was 80% efficient, you could say I was actually pulling around 160 watts from the adapter. I remember reading on another forum elsewhere that they use the same 230w brick for the gtx 1080 x7's, and people were complaining because they couldn't overclock them more because it exceeded the 230w limit. It doesn't look like that is the case here. (part of the reason why I went for this one, I was nervous about a 1080 drawing to much power)

[Pendragon]

56 minutes ago, Ocmersh said:

snip

 

Damn that is a stable OC. At almost 2ghz. That's solid. Your GPU is a pretty sick chip. CPU is a dud though LOL. 

 

Also doesn't work that way for laptops. If you're pulling MORE than the rated amount, that's really really bad. Mostly because the adapters aren't full PSUs. If you're pulling 200w in a gaming load means you will pull damn near the limit under an actual heavy load (gaming is not a heavy load). BUT if the most you are going to push it is for gaming, its perfectly fine.

 

2 hours ago, Ocmersh said:

I really just love that OCing and optimizing laptops is a thing now.

its always been a thing  

 

alot of the hardcore laptop fans are on Notebookreview. Check it out. The site is fucking ancient but that's where are the notebook people congregate. 

[Ocmersh]

 

On 10/25/2017 at 12:14 AM, Pendragon said:

exactly. this is the most terrible part of XTU. 

http://forum.notebookreview.com/threads/the-throttlestop-guide.531329/

here is a guide by our own @Eason85. Will help you do everything you need to do with TS. 

 

 

 

 

I tried TS. And now I am having problems that I just can't explain. I dropped it to 4.0, with a 0 offset, and now I am getting my temp on one core literally pegged at 100C and throttling. I tried switching back to XTU, and it idles in the 30's and then when a load engages it jumps straight to 100C. Any thoughts? (its core 0 in particular).

 

As per my OP, it was working fine... I am wondering if there is lack of contact between the plate and the die... Can the thermal paste shift after placement? I tested it thoroughly and the temps a week ago were pretty consistent.

[Ocmersh]

I even ran a sensor test using prime95, this was only with the speed set to 3.5 GHz:

 

 

Anything feedback would be great. The only explanation I have for the massive disparity between core 1/2 and 1/3 is lack of contact between the die and the cooler... I just don't see how that could be though given that it was performing fine before and I haven't touched it.... (don't have anymore past available atm... otherwise I'd have just reset it.)

[Pendragon]

8 hours ago, Ocmersh said:

As per my OP, it was working fine... I am wondering if there is lack of contact between the plate and the die... Can the thermal paste shift after placement? I tested it thoroughly and the temps a week ago were pretty consistent.

yes and no. it means the heatsink contact is garbage. gotta figure which side it is and tighten the screws more. alienware's had a huge problem with uneven heatsinks and there are mods you can do to fix it, but all annoying as hell. 

 

http://forum.notebookreview.com/threads/alienware-17r4-15r3-disassembly-repaste-guide-results.797373/

^ you can refer to this guide for high level information on how to deal with uneven heatsink but you'll need to tailor it to your machine. 

 

First test a repaste first just to make sure you didnt accidentally leave an entire core without thermal paste, otherwise it's uneven. 

[Ocmersh]

12 hours ago, Pendragon said:

yes and no. it means the heatsink contact is garbage. gotta figure which side it is and tighten the screws more. alienware's had a huge problem with uneven heatsinks and there are mods you can do to fix it, but all annoying as hell. 

 

http://forum.notebookreview.com/threads/alienware-17r4-15r3-disassembly-repaste-guide-results.797373/

^ you can refer to this guide for high level information on how to deal with uneven heatsink but you'll need to tailor it to your machine. 

 

First test a repaste first just to make sure you didnt accidentally leave an entire core without thermal paste, otherwise it's uneven. 

Repasted with artic silver 5... didn't have any thermal grizzly left. Didn't change anything with the CPU, however. GPU is still as cool as a cucumber.

 

Well... now I gotta get more screws... nearly stripped the ones there, I wasn't using that much pressure. I don't want to over torque the screws and damage the die...

 

I checked out that review and it looks nice, might make me feel confident enough to use that liquid metal stuff. 

 

Right now I am running a stress test at 3.7 GHz, with a - 165 mV offset, hottest core is in the mid-seventies with fans at like 65-70%.