Clevo Overclocker's Lounge

ok so i tested separately undervolted Cache from Core via TS, and guess what? my temps actually ROSE in Prime95 v26.6 1344k testing *rofl* same goes for switching from adaptive to static voltage, again higher temps (note to static: i knew from previous adaptive stress testing that 1203 mV was stable at 4.9 Ghz, but when I set 1205 mV static in Bios it gave me 1225 mV load voltage, so theres a +20mV offset when setting static. and nope, i made sure that no additional offset was set )

back to cache=core and VCore offset for me

 

Where is the LLC located at? On my way right now to the gym, but for checking out later

And @jaybee83 weird to see that you CPU gets hotter with static, maybe a skylake (mine) vs kabylake (yours) thing.

Gesendet von meinem Nexus 6P mit Tapatalk

 

Ouch why prime95?
Plus if it's one of the more recent versions ,AVX will result in even more heat.
Not to mention its way too overkill for laptops.

 

ill have to check and get back to you on that. on my way to the gym now myself ^^

v26.6 as i said. thats the most current version without avx. two reasons why i started prime95 testing with my 7700k: other tests, including 8h realbench proved not to be sufficient to ensure complete stability for my everyday tasks (includes video rranscoding using avx). plus, its a worst-case test for my cooling setup, thus ill be ready for this upcoming summer ^^

thus far, 30 min. of prime95 v26.6 and v29.1 (with avx) with in-place FFTs at 1344K proved to be just what the doc ordered, stability-wise for VCore testing

Sent from my HUAWEI NXT-AL10 using Tapatalk

 

double post

Sent from my HUAWEI NXT-AL10 using Tapatalk

 

This is just to confirm @Mr. Fox 's findings about GPU power draw.


Completely stock including bios, cpu, gpu and os settings.No AC or External fans. 621W. So now you have 2 verified results.

http://www.3dmark.com/3dm/20230160

 

Here is the rest.
If you add it all up not counting the ia core watts it comes to roughly 574W+
The balance left would be left for drives/screen/lights/fans/usb devices and so on.
This should help to understand the breakdown of total watts.....With in reason of course.

 

Maybe this small widget can help out power castrated 1060 and 1070's users.
Download and disclaimer from GitHub:
https://github.com/LaneLyng/MobilePascalTDPTweaker/releases/tag/1.0.0.0

@bloodhawk , nothing new, the same addresses that we tested before. Regular users might be fine with this; extreme users better wait for Prema mod.
A programmer is still needed to flash it in.

My updated daily clock (950~975; 1000~1025mv@+202MHz; else @+189MHz; mem @+450MHz; passed 6hr Heaven):
FireStrike: http://www.3dmark.com/fs/12765819
FS Extreme: http://www.3dmark.com/fs/12765858

 

Looks interesting, but it would not yield optimal results in the end.....

 

Good freaking stuff man! Though the addresses might change depending on the vBIOS version. IIRC.

 

almost 1000 points more than me, even when you only use 2050mhz when i used 2000-1987mhz, not bad, was there more secret sauce in that? cause i thought for a roundabout 1400mhz 980ti is something like 2000-2050mhz 1070 in terms of raw perf.

 

Okay, so I checked. 5GHz TSBench 1024M.

1.2822v core, 1.0508v cache


1.2822v core & cache:


I ran multiple 4.7GHz tests before this, and I have determined that slight fluctuations in ambient temperatures are the culprit of changed temps. My first 4.7GHz test with reduced cache voltage was hotter than the next run at 4.7GHz with same cache, and even though the runs were RIGHT after each other, the room got slightly cooler when I started the same test, and when I ran a cache-reduced test right after the temps actually went down from both the first and second runs, etc. I tried over and over and all the tests I made were within margin of error, likely from slightly changing ambients. With that, I can say it doesn't seem to matter.

Also, don't mind the better speed with equal cache/core in the pictures above. In my 4.7GHz tests, the first one I did, with the reduced core, was the fastest of all the 4.7GHz runs I did. In this test, the 5GHz run with the equivalent core/cache was best. So all in all, it appears to actually not make a lick of difference.

It is very possible that the lower cache doesn't actually apply, and core and cache both use the highest option. I have no way to check cache VID; if someone has a way I'd be glad to do some more testing.

 

@D2 Ultima Ambient temps are going to change things over time....

 

Yup. I managed to get a particularly cool night this one time. Temps were still 27c ambient but I suppose other atmospheric conditions made it nice and cool (to me anyway). I was able to play Overwatch, which often approaches 60c in the daytime on my GPU, at 48c. Just because it was a bit cooler.

That's a 12c drop at already-low temps. It's astounding. Reducing temps below 60c is already a pretty difficult thing to do. Ambients are huge. It baffles me when I see people in 70F temperatures overheating their computers and I'm like wot.

 

at 70 degrees you would die

 

I said 70F

aka 21c

 

A big part of the America's use lbs/F vs Metric/C

 

Right, not the best out there comparing with the Prema vBIOS.
But normal users if they only need, let's say ~180W for their 1070, or ~140W for 1060, they might be satisfied with this.

Thanks man! I tested with Asus, Clevo, and MSI mobile BIOS, they are working okay so far. It needed updated when more models come out.

The more "secret" is, keeping the ambient temp as low as possible, lol.

 

This is true, but it sure beats not having anything!
And then there is that little part of needing a programmer to program it.

I like it though.

When Prema gets done we should be seeing pulls like this one....290W+ (Whether they are optimal or wasted....only time will tell) And no power throttling...Voltage...Well....That's another story...

 

Something is up with my 3DM results in 7. Installed 10 in a second drive and the graphic scores are right on that OS compared to 7:

http://www.3dmark.com/compare/fs/12767877/fs/12767623

http://www.3dmark.com/compare/3dm11/12203142/3dm11/12203109



Everything was run with the same settings. drivers and nVidia control panel settings as well.

 

You need to use the SLI profile/Bits for Superposition to properly work with 1080 SLI.

 

I was just using superposition real quick to see if it's just 3DM or everything. The issue is my 3DM graphics scores for both benchmarks.

 

Very nice! That's like a shut mod implemented on vBIOS level. I might test out some more addresses to see if I can find some breakthrough.

 

http://forum.notebookreview.com/threads/clevo-overclockers-lounge.788975/page-1225#post-10533052

 

yeah shuntmod is more like a bandaid fix atm, but i like it so far, definitely no Prema vBios but now **** get's done more properly like i want and clocks are definitely much much much more stable. but about the "ambient temp" is there more than that besides 50mhz core clock? do other things scale with ambient temp which i can't monitor?

 

So after the mod. Did you rerun Superposition?

 

The shunt mod evens out the clocks when pushing them but yes it does not magically unlock everything.

 

so just tried the last 2 days to get my ram to 3200 and even i finally did it, thanks to @Mr. Fox and @Johnksss@iBUYPOWER @Prema and some other guys ofc which i forgot to mention, but it seems after 30mins of bootup or memory training it looks good so far, not stress tested yet but will do with linux using "stressapptest". But it survived 3 shut downs and POSTs so i guess i don't have to wait every time about 30mins lol. and especially to @Mr. Fox after i stumbled upon his trick to manually insert the tWTR value and i instantly noticed my book "tries" to POST differently not like just shutdown 10 sec shutdown 10 sec etc, more like shutdown ran 5 mins shutdown ran 2 mins yeah shorter every time. 3100 with "Odd Clock Ratio" worked quite instantly and only had to bump the voltage to 1.25v to get 30mins for a "quick dirty" test done successfully. Used 1.3v just for safety.



i know it's a very minor achievement again but free boost is free boost

 

nice, good job there buddy! call yourself lucky, 2 days to get a ram speed stable is actually pretty quick. ive been trying to get 3000 mhz stable on my sticks for months now!

and ha wow, first time ive seen a 30 min boot to be successful

Sent from my HUAWEI NXT-AL10 using Tapatalk

 

What voltage are you guys running 3200 Mhz at ?

@Prema @Mr. Fox @Johnksss@iBUYPOWER

 

@Scerate mentioned 1.25v/1.3v for his setup

Sent from my HUAWEI NXT-AL10 using Tapatalk

 

As low as 1.3 and as high as 1.42

 

Interesting. Thanks for that, his voltage seems to be rather nice.

My 2800Mhz kit boots at 3000Mhz @ 1.3v - 15-17-17-39-526 almost instantly without training (or fast training?) , ill see if i can get it to boot at 3200mhz @ 1.35v. I was able to boot previously at 3100Mhz @ 1.2V , but that was not stable at all.

 

Might be because you had it at 2800 already so your timings were already trained?

 

2800Mhz timings were way higher.

 

have to tinker around with some settings anyways, memory isn't really stable, tried 2x 1h "stressapptest" runs and first time it put out quite some errors, the second time too ofc (but used +20 offset on Uncore to be sure). gonna try to tinker with subtimings a bit

 

true. Though it's still far above the default timings i think

 

1.300V or 1.400V, depending on what kind of mood the machine is in. I reprogrammed the XMP to use 1.300V by default. Same for 3000. I could not get 3333 to boot with 1.300V so that was 1.400V minimum.

 

Yeah thats my next thing to do as well. Im sure it saves a lot of time down the line.

 

Well, I have to use custom tREFI to have 3000 or higher stable, so I still have to use a custom memory profile. There is no way to program the custom tREFI values with Thaiphoon Burner. So, I set XMP and once that boots change it to custom and tweak the tREFI to where it needs to be for stability. The default tREFI is too low and causes way too many BSOD. But, it is convenient having the voltage already set. One less thing to have to mess with.

 

Ah not even the pro version does that?

 

what values do you have for the tREFI? just curious

 

1.3v

Getting a lot of errors when upping it to the 1.35v-1.4v range.

 

For 3000 I use 12296 and for 3200 I use 12960.

 

No. It's totally missing in the XMP table of settings. I don't know why. I have been meaning to ask Vitality why that is. I think that value is supposed to be calculated automatically during memory training, but it does not get calculated to the most stable value for some reason. Maybe it's a Clevo glitch. But, I do see at places like overclock.net that setting custom tREFI is a common path to stability for many desktop jockeys as well. I also use custom tRAS (39) and tFAW (32) but I can and have programmed those values with Thaiphoon Burner. The default G.SKILL timings for tRAS and tFAW are not as stable for me.

 

is there a technical reason for that, i always thought that when the ICs are the same as the desktop ones, theoretically the stick can reach the same speed, with proper tweaking ofc

ok thank you will try them then, right now i upped the ram voltage from 1.3 to 1.35 cause last test ended up in a BSOD and maybe your trick helps

edit:
well still nothing

Spoiler

root@MEI:/mnt/c/Windows/System32# stressapptest -s 600
Log: Commandline - stressapptest -s 600
Stats: SAT revision 1.0.6_autoconf, 64 bit binary
Log: buildd @ kapok on Wed Jan 21 17:09:35 UTC 2015 from open source release
Log: 1 nodes, 8 cpus.
Log: Defaulting to 8 copy threads
Log: Total 16327 MB. Free 13379 MB. Hugepages 0 MB. Targeting 15318 MB (93%)
Log: Prefer plain malloc memory allocation.
Log: Using memaligned allocation at 0x7fc1ee311000.
Stats: Starting SAT, 15318M, 600 seconds
Log: Region mask: 0x1
Log: Seconds remaining: 590
Log: Seconds remaining: 580
Log: Seconds remaining: 570
Log: CrcCopyPage CRC mismatch ffffffff01ffffffff11ff7fffff0280ff7fffff1370 != ffffffff01ffffffff01ff7fffff0280ff7fffff02
80, but no miscompares found. Retrying with fresh data.
Report Error: miscompare : DIMM Unknown : 1 : 41s
Hardware Error: miscompare on CPU 4(0x80) at 0x7fc4bfe9b788(0x0 IMM Unknown): read:0x0000001000000000, reread:0x0000001
000000000 expected:0x0000000000000000
Log: CrcCopyPage CRC mismatch ffffffff02ffffffff01ff8bffff02bcff8bffff0270 != ffffffff01ffffffff01ff8bffff0270ff8bffff02
70, but no miscompares found. Retrying with fresh data.
Report Error: miscompare : DIMM Unknown : 1 : 43s
Hardware Error: miscompare on CPU 0(0x4) at 0x7fc49e253da0(0x0 IMM Unknown): read:0x0002000000020001, reread:0x00020000
00020001 expected:0x0002000000020000
Log: Seconds remaining: 560
Log: CrcCopyPage CRC mismatch ffffffff01fffffffef11027ffffeff801027ffffeff30 != ffffffff01ffffffff011027ffffeff801027fff
feff80, but no miscompares found. Retrying with fresh data.
Report Error: miscompare : DIMM Unknown : 1 : 50s
Hardware Error: miscompare on CPU 4(0x10) at 0x7fc3136e0fd8(0x0 IMM Unknown): read:0xffffffefffffffff, reread:0xffffffe
fffffffff expected:0xffffffffffffffff
Log: Seconds remaining: 550
Log: Seconds remaining: 540
Log: Seconds remaining: 530
Log: Seconds remaining: 520
Log: Seconds remaining: 510
Log: Seconds remaining: 500
Log: Seconds remaining: 490
Log: Seconds remaining: 480
Log: Seconds remaining: 470
Log: Seconds remaining: 460
Log: CrcCopyPage CRC mismatch fffffffe01ffffffff01fe97fffddf60fe97ffff0360 != ffffffff01ffffffff01fe97ffff0360fe97ffff0360, but no miscompares found. Retrying with fresh data.
Report Error: miscompare : DIMM Unknown : 1 : 155s
Hardware Error: miscompare on CPU 0(0x2) at 0x7fc22a7286e0(0x0 IMM Unknown): read:0xf7fffffff7fffeff, reread:0xf7fffffff7fffeff expected:0xf7fffffff7ffffff

but going to bed now, tomorrow is still another day

 

It could be a difference in memory modules rather than the computer itself. Some modules may behave differently than others with more than 1.300V. There seems to be a silicon lottery of sorts in RAM modules the same at everything else.

 

ram tweaking is total voodoo if u ask me its more like "massaging" speeds and timings into the sticks until they accept them and sometimes it just doesnt follow any reasoning whatsoever.

examples:

- tREFI is the only timing that actually increases performance / throughput when its relaxed (higher values). still, @Mr. Fox has already shown that in his usage scenarios, it helps to tighten it by upping the value with increasing speed!
- VDimm can not only be too low but also too high for a single speed! so u gotta find the sweet spot for each and every speed u set your sticks at. "sure" you say, "voltages for a cpu can also be too high or too low" nuh uh, not the same thing! say 1.3V is unstable at 2800 Mhz and throws up errors, that doesnt mean that at 3000 mhz it wouldnt be the perfect setting! to with each new speed u gotta reset everything uve found out thus far and dial in from scratch!
- a lower ram speed might require you to up uncore and VCCIO voltage, whereas a higher speed needs VDimm adjusted but is totally happy with uncore/VCCIO at stock voltages
- the simple fact that ram and mobo slots can be "trained"....i mean, seriously? try and train a cpu, gpu or display

 

What if i have a sodimm adapter for the programmer?