Clevo Overclocker's Lounge

In that case mx4 or icd but I'd rather wait for the shipment.
ICD is good but it is hard to remove it.

 

Extreme fusion, Mx4, and nt-h1 are the ones that stand out on that list... It depends on mounting pressure, but...

Edit: nt-h1 is usually the lowest scoring of those three, but is cheap and not bad. MX4 does well with low mounting pressures. Extreme fusion does well with medium and high pressures, but isn't that different from mx4 in low pressure situations...

Sent from my SM-G900P using Tapatalk

 

Arctic MX-2 is a very good paste too regarding price-performance ratio.

As an ICD user for years I simply don't get the complaints about scratches on the dies. You have to let ICD soak up the alcohol for a few minutes. Afterward it's fairly easy to clean it up. @D2 Ultima

 

MX-2 and MX-4 are worthless for laptops, though. The mounting pressure isn't high enough and their performance degrades rather quickly and they pump out.

I'd say Gelid, ICD, or my personal underdog, Arctic Céramiqué 2 for non-liquid-metal. Anybody who actually uses AC2 do let me know if it performs to specifications; I like being sure =D.

 

My temps are in the 60's at best with cooler and auto fans.

 

Hm, I got decent results with MX-2, given the heatsink isn't warped. I say decent not superior or excellent and it is cheap.

For casual gaming it is okay, for all other options (benching, heavy duty gaming) ICD, Kryonaut and GC Extreme is the way to go -> CLU ftw!

 

Here is a link to my album of image i made during the repaste. http://imgur.com/a/l4A01

Think the line on the cpu(last image) is fine? I made a banana the first time so i did my cpu over again. Problem with this paste is getting the stuff out in a straight line. When you push a bit nothing comes out but when you add a tiny bit of force the stuff suddenly comes out a bit too much. I also noticed it started to curl as soon as it came out tho thats probbably due to it coming out too much. Would heating it up next time will help with that?

 

@D2 Ultima, @CaerCadarn, @Johnksss@iBUYPOWER, thanks for your inputs... But... I could only buy what the shops here have. I ended up buying a tube of MX-4, and compare the results. No changes were made to the CPU profile except to increase current limits to 300 A on both tests. Now, the CPU doesn't thermally throttle, but undergoes power-limit throttling.

 

because 300A is to many amps...

 

100A and 100W (long), 108W (short) set in W230SS BIOS are enough here to run 3.9Ghz on all cores without power throttle. Beyond that the AC Adapter trips...

 

I tried setting those power options in Prema Mod BIOS on my notebook, and I get a black screen (but active power LED). Does the AC adapter need to be connected for these settings to work?

 

Amp value is 1/8, so you have to input 800 in order to get a value of 100A.

 

Isn't it the same for watts as well in the bios?

 

nope

 

I set my amp limit to 256A because I found that under 256A I was prone to current limit throttle in rare cases; @Dufus can echo these findings. It's just a limit though; it doesn't mean my chip is going to draw 256A for everything. That'd be way too much power.

On the other hand, 100A limit was more than ok for most situations.

Yeah as Prema said, set that to about 800. If you find your current limit doesn't stick in the BIOS though after turning the system on/off etc (like mine) you could just use throttlestop to set it in TPL, and keep it set. I don't even bother with my BIOS anymore I just save the profile in TS8 and keep it, and on windows boot I simply start it up. It starts minimized, turned on, and at stock speeds with my undervolt, power limit, and current limit changes applied, and no problems for operation.

 

Just set that to like 300000 and you'll be fine regardless of what the multipliers are set to. This is a limit, not a static value. All you need to do is set it higher than what the CPU is capable of consuming. The CPU will only use what it needs, so you can max it out with no adverse side effects. The same applies to the other power limits on the same screen further down the page you are on. Same also applies to current limits in the sub-menus of the CPU VR power settings sections of the BIOS. You need to limit voltage, but not power.

Edit: here... watch this video from my YouTube channel...

 

256A max limit for a i7-4800mq? For a 47W tdp cpu? Interesting....

You know....I have found that this doesn't always hold true. I use to be a firm believe, but now... Not so much....
When you raise that amp setting to high you also raise the starting vid voltage on some benchmarks as well. Only on some things, other benchmarks are more lenient on vid voltage. And of course this is not linear across all systems.

Speculations of course.

 

That's interesting. We both were firm believers in that and it used to be true. Are you noticing this change on certain CPUs or on all of them, new and old, or just newer processors?

Edit: I will tinker with that on the 4790K and see if I can detect any performance changes, whether good or bad.

 

Kills current limit throttle in programs like Linpack or XTU Bench (not stress; that works without 256A limit):


Don't mind the low score for the first run; I simply didn't close my other programs before running it.

 

Actually after I went back and tested the 1680 V2 I really started to notice it. I would be super stable for say FS, then I would crash or something and make a change to amps only. Then I notice the cpu test would not complete. it gave me the error code you don't have enough qpi vtt voltage. So i had to add more vid voltage to compensate. In some most cases that amounted to way much more heat and wasted watts. Remember when I said I dropped back down to dual psu's instead of 4? It was because of that right there. It was just generating way to much wasted power and benchmarks were suffering in the long run. Not all, but quite a few. The system was thinking it had what it didn't really have so it pushed it hard.( Think this is only true with the fully unlocked versions though) Still could not go past the hard coded 1.5V, but you could sure push it like it was. And this intern puts a heavy strain on the other parts because they are in place to try and tone the voltage and watts down across many parts. So instead of being optimal it now becomes wasted and parts perform worse.

Interesting. I guess if it helps you then that's cool. I know it's not going to help me unfortunately.....

 

How is a haswell quad core on laptop with less than 100w of tdp drawing 256A of current is beyond me......

 

I think it's more of a bug on haswell; at least mobile haswell.

Setting the limit to 256A seems to kill the current limit throttle for such scenarios. It doesn't actually use that. I've also noticed that Skylake generally doesn't have that bug in XTU Bench etc.

Even setting to 255A doesn't kill the current throttle in those rare scenarios; it's just something about 256A+ that does it.

 

Not a bug with haswell. It's a setting in the bios that may or may not be exposed to set.

 

If one of you have the time/want to test this for the community, could you go through with different multipliers, adjusting voltage as necessary for each multiplier, and try different amp limits between the 100A and 256A, recording stability, current throttling, wattage, etc., for each step in between? By making a written log of this type, it will no longer be observation or hypothesis, but closer to proven. That would be a boon for all moving forward...

Sent from my SM-G900P using Tapatalk

 

It's already been done and not current throttling for me on a 4930k/4960X/1680V2/6820HK/6700k/5960X
It only happens if i try to bench certain things at 4.9 ghz and above. And that is actually where it should happen because of the voltage cap of 1.5V

 

wtf mang? make it 15 days so I can buy/return laptops like candies.

I don't buy to keep, I buy to bench and return!

 

It's open to set in my BIOS... but that isn't my point.

I'm saying that without setting current limit to 256A or higher, that weird current throttle shows up. Dufus noted it on his MSI, and I noted it on my Clevo; he was the one who suggested I try 256A or higher and see if it killed it, as I used to keep 100A set, and sure enough 255A or less still lets it show up and 256A or higher it vanishes. This is seemingly irrelevant of how much power/amps it actually draws in those tests, and since I don't see the issue on skylake, and I've never heard of people on Ivy Bridge having any current throttle issues in XTU Benchmark (that and Linpack are the primary areas where I have seen the current limit throttle show up to any sustained degree) I deduced it's probably something specific to (at least mobile) haswell.

Note however that dumping more voltage onto the CPU to a certain point also got rid of the current limit throttle, but the amount needed for overclocked throttle killing was too much for my cooling system to finish a test (as I have a rather hot/leaky chip and a relatively poor cooling system).

 

I just played with it a little bit and don't see any difference at 4.3GHz or 4.5GHz or 4.7GHz with the power limits maxed out at 1023A (8191 in the BIOS) versus 300A (2400 in the BIOS). I have not tried under 256 to see if the current limit throttling surfaces. At 4.9GHz I get clock watchdog timeout issues without the power limits set to 8191 in the BIOS. Running wPrime 32M at 4.8GHz I am seeing over 200W CPU draw in HWiNFO64 and 4.3GHz about 118W. This is the same with 300A or 1023A. This is almost identical to what the 4930MX did in the Alienware 18.

But, I also do not seem to have the 1.500V limit that some of the other machines have that @Johnksss@iBUYPOWER mentioned. Why this one seems different I am not sure. Maybe Clevo accidentally forgot to cripple it with a 1.500V limit.

I have CPU input voltage at 2.200V and at 5.0GHz VID needs to be set to 1.600V to run wPrime 32M. It will validate CPU-Z at 5.0GHz with 1.550V, but this only works on AC with the system super cold, bottom cover removed and cold air blowing on everything. Without AC the CPU voltage droops too low and the machine freezes or blue screens at 4.9 or 5.0GHz under load.

Maybe you're thinking of stock. 4930MX and 4790K both pull in excess of 200W when pushed to 4.7GHz or higher. In the P570WM the 4930K and 4960X pulled almost 100W more at similar clock speeds. I also think there may be some kind of algorithm between what the amps are and how many watts get pulled. I doubt it is actually pulling as many amps as the limit is set for, but raising the amp limit may allow it to pull more watts under load.

 

but 4930mx is same as 4790k right? just on a different pcb and mounting hardware (pga vs lga socket)

 

They are certainly very similar. Behavior and performance are close to identical, at least it sure seems that way to me. I have only had one of each CPU, so I couldn't say it is a hard and fast rule that applies to all of them.

4940MX seems to be a piece of junk and doesn't hold a candle to 4930MX or 4790K.

 

start of the bga binning downfall

 

I cannot go past 47x4 without thermal throttling using nothing but max fans, but at 4.8GHz even with thermal throttling it almost reaches 200W. I've got in in my lap right now, not using AC cooling, but without thermal throttling when using AC cooling it goes over 200W.

See images in spoiler.

Spoiler: 4790K Power Draw

That certainly could be. It was the last of the Mohicans for socketed mobile Extreme CPUs, and it appears that Intel didn't really care whether it turned out well or not. And, it certainly didn't... or at least all the examples of 4940MX I have personally seen are garbage.

 

I was referring to at below 4ghz. It was a 4800mq or something. I know how much watts CPUs pull when higher than 4.5ghz, not referring to that.

As for algorithm, yes, P=VI is the general case I believe.

As for 200w+ 4790k, you should be pushing well past 1.4-1.5v for that kind of power consumption. Seems awfully high for package power.

 

Check the spoiler in my post above and you'll see an example of that for 4790K. But, 6700K barely rarely exceeds 100W even at 4.9GHz.

 

Ehe, I wasnt talking about the overclocked CPUs, I was talking about a 48xxmq etc. Still, awfully high power consumption. Within like 30-40w of my 5960x 4.4ghz OC.

 

hotwell

but the 6820HK would cross 80w on 4.3ghz

 

Yeah, that's not what I meant.
Example. Bclk can be seen or not seen in an unlocked bios, but it's not unlocked in a lot of machines. And making changes to it does nothing. Is more along what I mean.

I'm also not knocking the 256A thing either. It's just not for me is all.


Side note:
Yeah, my 6820hk does not need 256A to operate at 3.5 ghz

 

This is exactly what the 6700K would need to go higher. On desktop boards it's locked at 1.6V, with a setting in the bios to by pass and go higher than that, but then you run the risk of damage. So it makes sense why most are locked. makes me want to go find a machine like @Mr. Fox has for single gpu benching!

 

Only the finest of failed desktop chips were used in 6820hk! LOL. Jk. imo

Also, apparently Intel's reference vdroop settings were lowering my Vinput by like 0.1v.....

 

on asus boards you have a switch to unlock the voltage limit, else it will lock the cpu to lower voltages to prevent accidental damage

what motherboard? I'm running LLC7 and seeing this when fully loaded (target 1.295v)

 

x99 micro. I just have an option to disable or enable intel reference vdroop for VInput. No individual LLC tweaking like other boards.

My vinput dropped from like a target of 1.8ish into 1.725v on loads I think.

 

Are you asking me or telling me?

 

Ah, it's a never-ending loop with my notebook, guys. See what happens at stock settings (47 W power max, 58 W short power max, 58 A current limit, ratios 35,34,33,33) :

1. I run P95 small FFTs. 100% current limit throttling, CPU runs at 3.10 GHz.

2. Increase current limit to 300 A, current limit throttling goes away, power limit throttling reaches 100%, CPU runs at 2.8 GHz.

3. Increase power max limits to around 100 W, CPU draws 67-68 W, gets bloody hot and 40% thermal throttling happens. CPU runs at ~ 3 GHz.

I cannot escape this, damn it.

All I want my notebook to do, is not throttle at 100% load when plugged in. P95 small FFTs may be an unrealistic load, but it's very very useful for testing the cooling limits of a given thermal dissipation system, and it's obvious that the one in my notebook is insufficient. @Prema how did you get your notebook to run @ 3.9 GHz???

 

How is it useful for that? Its running a incredibly unrealistic load on CPU. The temperature you get is extremely elevated.

I dont believe I have runned Prime 95 Small FFT on my current processor at 4.4ghz/1.3v and I refuse to do so.

Stop running Prime95 and use more realistic load testing. IMO.

 

Right... any suggestions? IntelBurnTest, Linpack, P95 Blend?

 

None of those. Linpack can be just as bad as prime95. I believe burntest can be even more brutal.

Wprime is a good start. Run a couple loops of 1024m.

 

No, none of those. Try Cinebench R11.5, wPrime 32M and 1024M, ThrottleStop 32M and 1024M.

 

Also Fox, I think my 5960x only have package power draw of 200-210w on 4.4. So your power draw for quad is definitely very high.

 

As others have echoed, P95 is not realistic. Cinebench works or TSBench.

Also, I do want to point out that while you stated before that you need to stick with the thermal pastes in your country, you should at least attempt importing something better than the MX-2/MX-4 stuff. I don't know if liquid metal helps your situation; your contact point may not be copper for your heatsink. If it is though, then that's what you'll be wanting. If it isn't, you should probably try importing Gelid or ICD. Even if you have to ask one of your computer stores to special-order one; that should be fine.

But I share in your pain, I really do. I wish I could hold 3.8GHz 24/7 without ever overheating in any realistic load I put on my chip (gaming + CPU-based recording/streaming would be about the limit, or video rendering via Handbrake etc).

 

OCCT large data set