*OFFICIAL* Alienware 15 R4 Owner's Lounge

Hello Papusan. I have the exact same system so i post you my results. This is stock paste, undervolt 360 core and 160 cache in TS. I never was able to get more then 1130 with a constant 3.9GHZ in Cinebench R15.

 

Thanks and +rep 3.9GHz and with your massive undervolt the Cpu Package Power shouldnt exceed 45-48W. Could you Lock Package Power Short in TPL window to 50 and re-run test. See below... 3.9GHz and package power was barely 48W and the scores is a way higher. And as well take a look on load voltage and see what clocks run in middle of the bench. They shouldn't drop. And don't forget to disable the AV software. Thanks

(8700K running [email protected] clock speed all 6 cores). 48W max power consumption and without tweaked voltage for the clocks.


Edit. You could also re-run as your previous test a second time without changing Package Power Short. But this time put in your Cashe undervolt (-0.160mv) for both Core and Cashe. Then we can compare with your first run you did with -0.360mv for Core and -0.160mv on Cashe. Post pict as before. Thanks

 

View attachment 161253

-0.400mv uv on the cpu - 0.180mv on cache.
the temps are around 5c higher then usual since running on silent fans profile. The cpu speed is always dropping after 2/3 of benchmark to 3.3ghz no matter what settings or undervolt, I dont quite get why since this never happens in games.
This is one of the highest scores I got with 8750h (highest was low 1200, but couldnt replicate today.

 

Thanks.
First picture doesn’t show up. Could you then lock Package Power Short to 50 in ThrottleStop TPL Windows and re-run. There is no need for more than maximum 50W Package Power to feed 3.9GHz on all 6 cores in Cinebench with this undervolt(if it’s real). And what voltage could you see during the Benchmark? Thanks

Edit. Could you also try Performance mode to keep temps down as well see if scores being more normal for the clock speed.

 

Coffee Lake allows for a bigger undervolt of CPU over cache? I thought both are locked to the same value....if this so, then the CPU is not being undervolted by 360mV but by the cache value of 110mV.

 

I see significant temp drop if I compare my curren 0.415mv core 0.180mv cache undervolt to 0.180mv core+cache undervolt. Did several tests for over many hours in bf1 and pubg. And yes, if I undervolt the core any further the system gets unstable.
Cant say I am not suprised, since I could run my 6700hq only with -0.140 on both max.

Here one more run with 0.415mv on core and 0.180mv on cache and locked package power at 50. The CPU runs at way lower clock during the bench, which results at lower score and temps:

T

 

I know it worked with lower undervolt for Cashe vs. Core on my i7-6700K. Probably also on my 7700K. But not a good solution for max or high overclock. But the 0.360mv or even the -0.400mv undervolt we see here is most likely fake. The lower Cashe undervolt (-0.160mv / -0.180mv) will most likely override the higher Core undervolt.

With correct -0.400mv undervolt you would see below 444-45W power consumption in Cinebench R 15 with 3.9GHz for all 6 cores.

Run same 0.415mv on core and 0.180mv undervolt but use the default Package Power Short settings.

The lower temp is due the power starving.

 

I have run multiple test with different power short and different core voltage offset. For power short anything under 68 will make the cpu run under the 3.9GHz on all 6 cores. 2.4Ghz was the lowest on some cores during the test at 50 power short. I saw no gain in undervolting anything above 160mv core and cache so i kept it at this setting for all of my test. Amazingly the best results i obtain was on battery power (1160).

 

Have you the load voltage numbers during the tests? Load Voltage can tell a lot about the Power consumption. I don't talk about idle volt (no load) or voltage spikes if you get that in the tests. I talk about average load volt in the tests. + or minus a few mv ain't so important.

 

Default power:

Higher score.

Still the CPU Clock goes down after 2/3 of the bench. Which is weird, since this never happen under load in games.

Regarding the high core undervot - as menitoned before, I tested it temp-wise it really does impact the temperature. I will run few more test, maybe I oversee some other factors but at this point I am pretty sure it works.

 

I have tested this and the max power consuption was: 67.1W and the score was 1096CB.

 

Your cpu reach Turbo Time Limit and the processor will go back to PL1 limits. The games doesn't tax Cpu in same way.

You could also test max Cashe undervolt for both Core and Cashe (I mean you use 0.180mv). Follow TS monitoring in TS. Tell what load voltage you get. Take also screenshoot in the middle or 3/4 in the Cinebench run for different voltage settings. + end results.

Edit. Could you also tick the box Speed Shift - EPP and save. Remove the checkmark for C1E.

 

Here a screen at the moment of the drop:
default power
fans started
-0.180 cache & core
you see bit higher temps, they would significatly increase under heavy load over time compared to 0.415 uv on core. thats have been only few seconds.
end result was around the same 1147 points.
Max Power: 70,7W

sry saw the edit to late, on the next run then.

 

Here one more run with default voltage, no undervolt:
Full speed fan profile:

 

And finally the result with full fan profile and -0.415mv core -0.180mv cache:


Thats 4° cooler then -0.180mv core undervolt after few seconds under load.

 

As you say you run a massive -0.415mv undervolt who work as expected. Final test... Tell me the Default load voltage when you run the benchmarks (You can measure the voltage value in TS windows after 13 sec and before clocks drop below the fixed 3.9GHz for all 6 cores). Do the same test for your max undervolt.

You should see a difference like 0.415mv if your undervolt vorks. F.eks lets say you run default voltage that is 1.175mv with 3.9GHz on all 6 cores in Cinebench R15. The next round you test ecual but with your 0.415mv undervolt in same way and clock speed (always 3.9GHz on all 6 cores in full load for i7-8750H). This means your max undervolted load voltage in same bench test with same clock speed on all 6 cores should be 0.760mV if your undervolt worked as it hould. You See?

Or let us put a higher voltage value... Your Default load voltage was 1.250mV in same benchmark but with the 0.415mV undervolt this means your max load voltage should never been higher than 0.835mV for same 3.9GHz all 6 cores if it worked as it should.

 

The truth seem to be somewhere in the middle. Ran few more tests. If you study the screens, you see that the "effective" undervolt value is around -0.250mv with -0.415 core and -0.185 cache:

1: no undervolt:


2: the -0.415mw core undervolt

 

I put it this way... Can you do as the pict. Screen shoot after 15-20 sec or before clocks drops below stock(need steady 3.9GHz all 6 cores as this is default clock speed for i7-8750H all 6 cores in full load). With your Default voltage, Your Undervolt -0.415 core + -0.185 cache: And -0.185 core + -0.185 cache. You may need have to put Package Power Short around 55-60 to stop Clock drops. Or even higher. If you can, put -0.215 core and -0.215 cache as well. Thanks

Edit. Another option is put -0.200 core + -0.250 cache.



You could as well pst pict of Throttle'stop TPL settings as well. Need only Default.

 

Sure, I will do it later. Undervolting bellow -0.185mv on cache wont probably work, its my max working undervolt on cache. Rest I can do.

But I cant quite follow what for. Its pretty much proven that the voltage goes down by around -0.250mv with my undervolt of -0.185 cache & -0.415 core.

 

what do you guys use to fully test thermal stability?
I've applied a -120mv undervolt to my 8750 and in cinebench theres no issues with temps (score 1163) but in aida64 with CPU FPU and GPU its running hot

 

Good thing you can do is 2-3hours of Battlefield1 or Pubg. Those will stress your cpu the most under realistic conditions. My cpu gets around 80 average after 3h rarely spikes close to 90. Very rarely few spikes into low 90s - rarely 92-93c max. No throttling ofc. I consider it quite safe.

 

I'll give that a try.
Playing wildlands with medium settings and increasing the undervolt to 120mv gives me averages of 85/87 and spikes of 96

I can get the temps under control by keeping the undervolt and capping at 3.7 (spikes only 90/91) and averages of 79/80.

Guess I'll stick with the return.... As no one else is reducing multiplier

 

Try my undervolt. I dont think that my cpu is soo special in this regard. And keep in mind, its important to lift the back of the laptop properly so it can vent the air.

 

dasachmo, you should try undervolt cache to -160 (stable) -185 (unstable bsod) and core anything between -160 to 225 (stable) anything above is a risk of freeze and bsod. I use -160 on both.
When i test aida FPU GPU i get 84C GPU and 98C CPU thermal throtling with stock paste. With LM i get 95C CPU no thermal throtling but cores are not running constant 3.9GHz But varies between 2.5 to 3.9.

Envoyé de mon SM-G935W8 en utilisant Tapatalk

 

Yup. I'm going to try that tonight.
Going to take off the multiplier limit.
The new bios update reduces clock speeds when it thermal throttles. Before the tech Repasted I was dropping to 2.7ghz

Ill try - 160 and play at 3.9. Also will check the gpu and fpu in aida.

 

slowly work towards -0.180 cache and -0.400 on core.

 

okay, playing for the past 35 mins. -160 each, i get max of 93, average of 82.
will up it and keep trying

 

The cache is more sensitive to bsod but most effective for lowering temp on these CPU (Tested on 4 different motherboard so far). If you can have it stable at -180mv temps will reduce by about 3c. Test the cache first to find your max stable then your core.

Envoyé de mon SM-G935W8 en utilisant Tapatalk

 

As I increase the undervolt I've noticed that hwinfo stops responding or hangs for a few seconds. No bsod but just lags .
Ive gone to - 160 cache. - 195 cpu.
Keyboard deck is really warm tho. Which is strange because it's only happened after the Repaste?

 

Tested with [email protected] all cores and -0.245mv for Core and Cashe (41-41-41-41-41-39). Load voltage 0.882V during the test and the power consumption maxed out att 44.5-45.5W. Can You can run it the same way. Screenshoot in the middle of the Bench test.

 

How do you stop the cpu throttling down to lower speeds like 2.9-3.3 under load. It seems to do that when the temps are high to prevent throttling and in Hwinfo doesn't show throttling either once this kicks in.
This is built into the bios right?
I've changed multipliers but this new bios seems to keep temps down by reducing speed.

 

Technically, that's how it's supposed to work (same as every other CPU).

Base clock is 2.2GHz. If you're using all cores, it can boost to 3.9GHz if everything else is good. If you're using only 2 cores, it can reach 4.1GHz. If you are running too hot, you will get less.

 

You're probably encountering the same dynamic throttling that doofus99 has, which is triggered after you reach thermal throttling (this was probably implemented to reduce the nr of RMAs because of overheating and now users can't complain since temps don't rise above 93-94C):

http://forum.notebookreview.com/thr...5-owners-lounge.815492/page-136#post-10760367

See if the PL1 value changes during the stress test (in that post, PL1 was as low as 26.875w).

 

Your settings are wrong @Papusan, it should be 41 41 40 40 39 39 if you want to emulate the behaviour of 8750h (but even then I doubt that a desktop cpu will behave in the same way as a laptop cpu does)

Do we agree on the thing with undervolt ? -0.180mv cache and -0.415mv core result in approximate -0.250mv undervolt ? I thought my tests proved it, if you have any ideas why not, let me know. Thought it was a very interesting discovery.

Now to the screen you wanted, the very second of the drop:

 

I know but this really doesn't matter because all 6 cores will run exactlty 3.9GHz in same way as i7-8750H in Cinebench R15.

I want screenshoot before the clock drop come. You show 3.7GHz and not the stock 3.9GHz all 6 cores for i7-8750H. Remember voltage drop with lower clocks.

Regarding a massive higher Core voltage undervolt vs. Cashe, this have been tested in Msi Coffee lake models. After further testing the tester could shoot down this theory.

 

Sure, there you go the clock is sitting at 3890Mhz during the first phase of the bench. Package Power: 70.5W



May I see that undervolting test? Since The HWMonitor clearly states the -0.250mv Voltage with applied undervolt and 0.250mv more when no undervolt is applied.

 

@Papusan , @CanIbiteYou

Wouldn't it be easier to estimate the effect of these uv offsets by comparing the power draw at the same clocks in the same scenario?

Since 8750h is power limited at 45w and the time window for PL2 is relatively short, it might be easier to compare the data in tests where the power draw isn't too high to be limited.

Can't remember what is the cpu power draw/clocks in aida64 CPU test, but it should be below 45w for 3.9Ghz (I think).

Can you compare the CPU power draw from those different uv offsets (if the power draw is indeed below 45w at 3.9Ghz) in aida64 system stability test (2 minutes should be enough) with just the CPU box ticked?

U don't need hwinfo for these runs, but for better viewing, use the Unified tab in aida stability test (to view the clocks, temps power draw and voltage at the same time), stretch the test window to the length of the screen and drag downwards the left button from the stress test window, so the grid values are 0 to 100 instead of 0 to 200.

After that, we can compare the differences in power draw from those offset (if the higher cpu core offset does make a differnce, it shoul show up with less power draw for 3.9Ghz).

 

Already showed in the Cinebench R15 run. Just look at the load Voltage for 3.9GHz. +100Mv will need it's Power for same clock speed...

Thanks. Could you do exactly the same for stock voltage? Then it's good. If you can get it work, maybe try also -0.200mv on Core and Cashe if you have time for it. And if it don't crash.

Read previous posts in the Msi thread regarding High undevolt on Core.

A nice test for testing thermals... Aida64 Stability Stress Tests Guide

 

Yeah, I saw the voltage difference for the clocks (3.7 vs 3.9), but this run is -415/-180 and I couldn't find (or I missed it?) any run with the same cache and core offset (-180/-180) with full load and all the info displayed in hwinfo to compare with (so we can see if the higher core offset makes a considerable difference).

 

Only last post with pict was what I asked for. The rest of the results in revious posts (all the pictures) didn't tell anything. Hence you can see what I asked for in my last post. Probably due my not very best English

 

@Papusan , don't worry about your English, it is good enough for a non native speaker

Your run was 3.9Ghz at 0.882v with 45.2w vs 3.9Ghz at 0.980v with 70w, so we have this comparison (basically this is what you asked for and the same with stock voltage in load), but I still doubt that there is a considerable difference between this run (-415/-180) vs the same run with -180/-180 (that is why I mentioned a comparison with different offsets, but I believe that the comparison is better done in aida64 with just the CPU box ticked so it is not power limited (hopefully).

 

@Papusan, this is my best score for cinebench R15. I found that AWcommand center was making my system results inconsistent. I have uninstalled it, cleaned all the files even in the regedit. I tested without AWCC and reach a result of 1236CB (highest score) -160 mv core and cache, original paste. Then i reinstalle AWCC and OCcontrol and the picture shows my best results with AWCC installed correctly. My point is you were right when you said that the 8750H at 3.9GHZ constant was capable of more the 1120 1160CB. I am sure i can now worked around to reach even better results.

 

The score is good for 8750H, but you shouldn't rely only on a single CB run to estimate the performance. See what scores you get in 5 (or more) runs (WITHOUT PAUSE BETWEEN THEM) so you can better asses the performance in sustained loads.

You can use the script from the link (below) and it automatically outputs the scores to the CB15.txt file (modify the script according to your path settings/folder names and replace 13 with the nr. of runs you want to do):

http://forum.notebookreview.com/thr...-line-syntax-bat-loop-detailed-output.815101/

 

@Papusan There you go:

Stock voltage - moment of the drop (actually it drops almost immidiatly since its overheating and throttling on stock voltage:

-0.200mv on cache and -0.200mv on cores at the moment of the drop. As you see the max voltages are way higher then if I run -0.180 cache and -0.415 core undervolt.

 

tested both settings you discuss about...

-160 core/-160 -> ca 1,07v on mine CPU
-500 core/-160 -> ca. 1,07v on mine CPU
-1000 core/ -160 -> ca.1,07v on mine CPU
-180 core / -180 -> 1,05v on mine CPU...

CB Score the same with every setting. Core and Cache must be equal... everything above Cache uv has no effect.

 

Well no idea why its different on my 8750h. The voltage is definitly lower with higher core uv.

 

There is enough with 2 max 3 rounds on fully locked down processor due PL1 limits (long time power limit).

Take a screenshoot before the overheating and the Throttle Mess starts.

You still show screenshoots with clocks lower than the 3.9GHz. See your post #285

 

Dude you asked me several times to make a screenshot, when the clock goes down lol? But ye sure, wait a min.

 

Read. Or my English strikes again
http://forum.notebookreview.com/thr...r4-owners-lounge.816045/page-29#post-10768103

 

Ye maybe I misunderstood it too, not a native speaker neither, sorry. Anyways:

Full clock speed in three different test.

1. Stock voltage - it throttles almost instantly so cant get pic of full clock speed:



2. -0.200mv core -0.200mv cache Undervolt:

3. -0.400mv core -0.180 mv cache Undervolt:





The one with -0.400 on core gets clearly lowest max voltage on HWMonitor.