Clevo Overclocker's Lounge

Read the last 20 pages of this thread, very entertaining. I wish I had my monster for the weather I've been having. 18 degrees or -7.77c here. Coupled with the wind chill and it is single digits. Snow expected tonight. Right now my garage is an icebox and this Ice Bear is loving it!!!

 

and add an undervolt if that doesn't bring the temps down a lot

I run on a 0 offset though, just stock voltage, then bump the multipliers to 46 for both Core / Cache Ratio (thanks to @Mr. Fox for always helping me optimize my BIOS settings)

 

If you want higher temp. Uncheck all except Stress FPU

 

can j00 elaborate please?

 

but why wouldn't you stress other components? isn't that cheating?

What's FPU anyway

 

Youre welcome http://www.overclock.net/t/35211/what-is-the-fpu

https://forums.aida64.com/topic/1629-real-world-benefit-of-fpu-test/

 

I am going to post this one more time.....
And actually try doing it first before commenting. Thanks. That would be a very big help!

Edit: Matter of fact, Everyone should try this to see if it's the same. I may be 1 MHz off or take a few seconds longer to switch....Can't really say why though...

With this. I'm able to game at 1911 solid, but if I hit a temp of 66C it drops to 1898. Now, if i cool gpu to 62C, it will go back up to 1911 and hold.
Can you try this to see if you can hold it?. And keep an eye on when the change occures. My range is 62C(Clock raises back to normal) to 66C (Clock lowers)


Edit: When I raised the clock to 1949.
At 1949 it drops to 1936. And it's still holding at 70C
At 1936 it drops to 1924@76C and return once temp hits 72C.

So now you have your range.
42-47=13mhz
52-57=13mhz
62-67=13mhz
72-77=13mhz
82-87=13mhz

 

0.912 V is awfully low voltage and might cause graphical artifacts. But since I never went below 0.950, I will give it a try now and check if I see any artifacts.

 

Only 1911 was tested at .912V

 

1911 @ 0.9120 V is no good for me, Fire Strike freezes. I'll try lower the clock.

 

Does your screen look like mine?

 

Here's a joke to lighten the mood =D

I wrote another guide. I think it's the best guide I've ever actually written. Here it is, in all its glory:

 

You had me laughing in that one haha

 

It looks like this:



Interesting that for 1936 @ 0.912 it says +206 for you, however for 1911 it says +240 for me...

Also, how is your core voltage slider unlocked and how come MSI Afterburner detects your core voltage ?

When I increase it to 1936 @ 912 a bug triggers and it throttles under factory clocks.

Are you using Prema BIOS or vBIOS ?

 

Set it to 206 not 240

Here is your 240+ with my card

 

This is to make a point about temps and clocks...
The setting is 223+ for 2000 on average.
Room temp & Auto fan

AC

 

Damn, how is your GPU able to do 2038 @ .975V.

 

@bloodhawk I don't know that it can actally hold that. I was just making a point about how much temps make a difference.

I'll see if it will actually finish that bench in a minute. Also, that was single gpu.

 

What did you do to be able to display voltage/unlock voltage slider in MSI Afterburner ?

I don't understand, doesn't all Clevo GTX 1080 have the same core clock/voltage table? Why is mine different ?

 

3dmark said it's 2050, but on screen it held 2038 the whole time.

That is irrelevant. It's just for visual effect and that's it. It has no baring on anything. As to the rest....I don't know.

 

How much points did you loose in Time Spy (and Fire Strike) with UV compared to the factory voltage? With factory voltage I get 6860 graphics score, with UV (1885 @ 0.912) I get 6660 points.

 

I do not lose like you do. And this voltage exercise was to see if you could hold a steady clock, not bench for points.

 

Could you please do a fresh single card Time Spy run for me with the latest 3DMark while you are undervolting for gaming?

 

Ah gotcha.
Usually for me unstable clocks crash during the second half of the Graphics Test 2.

Those temps looks awesome though.

 

Jesus Christ .981V.

 

Temps Play a huge role! Once you get past temps, then you run into power. . I haven't did the power mod yet, because haven't felt it was needed, but from what Meaker says...It's suppose to help keep clocks steady.


@DRevan
I have already done that a few times already. Even posted it in this thread as well. Not with this new driver or version of 3dmark though.

 

Exactly

 

Well there goes my next weekend. Testing testing more testing

 

just did a Fire Srike stress test with 1898 @ 0.925
The result was 99% which is pretty good considering that with factory settings most of the times it is under 97% (and thus failing the test).

 

?
Mine works. So I'm kind of lost here?

I can get 90 percent and still do better. Do not rely on that. "One size" does not fit all here. And you can get 100% and do worse in scores. Over time you may figure this out, but keep shooting for that 100%

 

Just did a Fire Strike Ultra stress test and as I suspected, PWR throttle there kills it and core clock is jumping between 1750-1898 Mhz even with 0.925 V.
With Fire Strike stress test the pwr limit is under 90%, so the clock is stable with this voltage, however Fire Strike Ultra stress test is pushing the pwr limit to 95-105% and the crappy factory vBIOS (or EC?) has a limit at around 95% and when that is reached, it will throttle the core clock more than 100 Mhz.
GTX 1080 FE cards have a similar thing built in their vBIOS, seems Clevo followed NVidia's stupid policy and did not remove this power limit ... thank God Clevo even locked the pwr limit slider so it can not be increased with software.
Of course there is always the hardware modding way, but I am not brave enough to mod the shunt resistors to increase the power limit ...

But you guys are saying that P870DM3-G does not have this power limit and the core clock is super stable even during Fire Strike Ultra stress test (in other word when the 95% pwr limit is exceeded) ?

 

Ok well I got some USB 2.0 tonight and it looks like it will now work; however, it would not launch at first until I changed UEFI to disabled. Anyone else have to do this with windows 10? Otherwise the memtest86+ wont launch

 

It depends on how it is formatted. I can boot it either way. The default is DOS, which will not work with UEFI enabled. To boot in UEFI I have to press F7 and select the USB UEFI partition 1 from the list of bootable options.

 

Got ya I didn't realize that but was using 7 before this computer

 

F7 will not boot the USB either for me. USB shows up but won't boot. Only option for me is to disable UEFI. No big deal it still works after I disable was just curious if that was strange.

 

I have CSM enabled as well as UEFI, so maybe that has something to do with it. But, the DOS version with UEFI disabled works fine as you said.

 

Where exactly are you getting these numbers from? This 90% 95% and 105%?

Maybe when you get around to dropping your cards temperature you will realize then that it's not power at all.... Until you get to that stage in your learning...This will never truly make sense.
If it was true throttle, your card would drop to 291 mhz or 1200 mhz or even 1400 mhz, not 1750 mhz.
And since this is SLI which draws far more power than a single card. I'm going with it's working as intended. And if you watched my video of atempting to run the same clocks but with different temps, then this would make perfect sense... The clocks fluctuate.

Just my opinion is all....

http://www.3dmark.com/compare/fsst/250653/fsst/250639#
Stock VS 2050/5508

Stock

Over clocked

 

I think a lot of people do not understand this, and some of the videos on YouTube with people talking about Pascal power limits may cause further confusion. Pascal is greatly affected by temperatures, even when temperatures are not very high. A great example of this can be seen benching on AC versus air. I can run everything the same settings and scores are way higher on AC because Pascal starts pulling back on clock speeds before normal operating temperatures are reached.

To be clear for anyone reading this, the thing with GPU clock speeds is very different from the CPU Current Limit Throttling issues that some are experiencing while using a messed up EC.

 

You are right. Quite a few do not understand this. That's why I posted direct examples not stuff I found on the net. To give people the "real" answer in "real" time.
I guess that is just not going over well...

 

Yes, it would drop to as low like that if it was thermal throttle.
I am using my desktop GTX 1080 to compare performance and behavior. Power display and slider is locked for the mobile variant so the 95% is just a guess which I have got from observing the behavior of my desktop MSI GTX 1080 Gaming X. On the desktop Time Spy consumes mostly 80-90% power, however at a certain scene it jumps up for a second to 95%. Desktop VGA does not drop core clock at that point, however the laptop VGA drops core clock to under 1800 Mhz if not undervolted. Now, with undervolting I can lower the amount of drop or completely stabilize it.
However Fire Strike Ultra stress test is another thing, since there the consumed power by the VGA is mostly 90-105%. According to my observation the laptop VGA drops core clock at around the same time when the power consumption exceeds 95% on the desktop VGA so I am guessing the trigger is at around 95%.

I have run Fire Strike stress test with the UV I mentioned earlier and it gave a perfect 99% score, however Fire Strike ultra stress test with the higher power consumption crashed, meaning that 0.925 V is not 100% stable for me with 1898 Mhz.

Just tried 1885 @ 0.925 V and as you said the % increased in Fire Strike Ultra, however the core clock is far away from being stable. In the 10 minute run I saw drops to 16xx Mhz.

ps: I have to mention that even my desktop GTX 1080 has slight core clock fluctuation when the power limit is jumping between 95-105%, however I was able to stabilize this with increasing the default 95% power limit to 105%. Now there is no core clock drop for my desktop VGA and it is super stable at 2000 Mhz.

 

GSkill 3000mhz XMP1 appears to be stable so far at 4.2ghz. Don't plan on overclocking the CPU very much for the games I play. No errors in in memtest86+ and ran aida64 for around 6 hours. Thanks for helping me get that running @Mr. Fox & @afloyd. Any other tests I should run for the ram to ensure it is stable?

 

You should be good to go, bud. Thanks for testing!

 

Make sure to re test after any bios resets or bios updates.

 

Passing on a little info that may prove useful to some regarding the DDR4-3000 stability being inconsistent. We don't really understand everything that relates to this, but seems like we have established some degree of cause and effect that could be useful in the event one is experiencing instability. I discovered something about the SPD checksum errors and the errors running memory tests that I was encountering. I found that problem was present with all G.SKILL and Corsair DDR4-3000 modules regardless of the module capacity. It affected 8GB and 16GB modules in both brands.

See the information in the thread linked below. I don’t know why the behavior is different on the P870DM3 than it was on the P870DM-G, but at least we have an explanation now to understand that the possibility of SMBus conflicts with popular monitoring software is present and what to do to remedy the issue. For those affected, the solution is in post #5, but go ahead and read posts #1 to 4 for context.

https://www.hwinfo.com/forum/Thread-SPD-Checksum-Errors-w-HWiNFO64-Running

I still get a few errors in memory tests unless I load XMP, reboot and switch to Custom Profile and tweak a few things, but that’s not the end of the world now that I understand what steps need to be taken. I can count it as performance tuning rather than a defect or an unresolvable problem. Below is a link to a photo showing the memory settings that are more stable for me than G.SKILL’s default XMP 3000 values. If you boot with XMP1, save and exit, then go back in the BIOS and switch to Custom Profile this retains the original XMP1 values in the Custom Profile. From there the minor changes can be made in the Custom Profile.

http://i.imgur.com/kBaJH2x.jpg

@Prema and I have let @Ted@HIDevolution, @Donald@HIDevolution and @Eurocom Support know about this. Posting it here so you folks, and so @Prostar Computer, @SagerPR, @Meaker@Sager, @pat@XOTICPC, @XMG and @mythlogic can be aware of it as well. The same issue is present with stock Clevo BIOS based on my trial and error testing.

I just ran this test for this post to show the problem can be resolved once the appropriate steps are taken by the end user. Those that do not use any system monitoring software that accesses the SMBus may or may not experience any DDR4-3000 instability problems.

 

That is weird indeed.

Maybe if one of the above tagged resellers could try to mimic the same effect or not.

 

@Mr. Fox , is this solved the problem you Will reporting a few pages back regarding inestability when you Oc'ed with xmp1?

 

The more the merrier. Brother @Johnksss@iBUYPOWER was able to replicate the issue and he was having far fewer problems than I was at the time. There could be other things in play that we don't understand yet.

It seems like it. If I run HWiNFO64 with SMBus Support enabled the memory test errors return and the SPD checksum errors return. If I shut down HWiNFO64 with SMBus Support disabled, wait a few minutes, then problems are gone. I am able to run HWiNFO64 without problems as long as SMBus Support is disabled. Screen shots are shown in the HWiNFO64 thread linked above. It would be nice to understand where the "collisions" are occurring and correct them in the BIOS/EC (wherever the issue resides) because I have never needed to disable SMBus Support with HWiNFO64 in the past. This is the first and only machine I have owned where having that enabled has been an issue. I am not drawing any conclusions, just sharing observations and what I did to stop the problems I was having.

 

great thank you, i´ll do a few test as soon as i get my laptop (bough 2x16gb 3000mhz) and report it if that can help.