PowerSpec 1720/*Clevo PB70EF-G* tuning!

Not talking about using XTU for any tuning.
Just running the XTU benchmark (which was designed to test CPU/RAM stability as it is changing loads instead of applying constants like other benches) and sharing results.

It also nets you more points on the bot than any other benchmark:

https://hwbot.org/hardware/processor/core_i7_8750h/

 

Unfortunately having XTU installed on top of Throttle Stop is not something I will do just to run the benchmark. The implications of having them running at the same time are not good, and I cannot keep the system running at peak performance without TS running. Given it runs games for several hours on end, and about half a dozen other legitimate benchmark programs I have no doubts it would be equally stable with XTU. Still, by tomorrow I should have at minimum 8-10 hours of consistent play time to further verify.

 

XTU doesn't make any setting changes on its own by just loading it up...it loads whatever values where set in BIOS/TS and makes suggestions, but doesn't change them as long as we don't hit the apply button.

You got 6.7 points for your CB20 run and get roughly a 150 plus 25 points for being first in XTU:

https://hwbot.org/submission/3982788_kill4l_xtu_core_i7_8750h_1979_marks

 

I'll give it a shot! I'm just stubborn haha.

Currently found an instability going to Sleep at -0.350v core so that is no longer a good setting on my machine. I went back to -0.250v and am increasing slowly as I go in and out of Sleep to find the breaking point. That is always an area I found touchy, as it puts the processor in very low power and undervolting even more can be too much.

I'll post back with the most it will undervolt while going and and out of Sleep and then I can try the XTU benchmark!

 

Sorry for the delay! I've been gaming all night, currently 1 AM here. I did manage to download and install XTU. To my surprise it has a ton more settings than my previous laptop.

Anyways, XTU score: 1887

https://hwbot.org/submission/4178576_t0bimaru_xtu_core_i7_8750h_1887_marks

3rd out of 1800+ on a first attempt! Going to keep trying for a bit more

 

Yes they are deliberately exposed settings so XTU can hook in.

 

Well done!
Sometimes the switch from being stubborn to being persistent is a small, yet rewarding one.

 

I picked up one of these the other day (open box deal was kinda irresistible), and must have got a bit lucky with the hardware. CPU ran hot, so I under volted with XTU by 150mv. GPU runs very cool though, I haven't seen it go over 75 yet, seems to hover around 70. Got a 2960 for Cinebench R20 with just the under volt.

Overclocked the GPU by 200Mhz with the preinstalled software and broke 8000 on TimeSpy: https://www.3dmark.com/spy/7477492

 

There is supposed to be a nice bit of play in the GPU clocks on these I hope you enjoy your machine.

 

I've just noticed something. Despite me setting Throttlestop to use less voltage, there's a minimum that the CPU will go down to. Just for fun, I set my UV to my core to -500mV and it still used the same voltage when I had -335mV. So there's a limit to how low you can set your voltage to. Shame, I wanted to see if I get under 1V when under full load.

 

what program do you use to stress test?

 

The voltage circuitry itself has certain limits on voltage ranges under load.

 

There are several. Mainly I use Cinebench, TS Bench, 3DMark, memtest, XTU. There are many others you can use, most free. It depends on what you are wanting to stress and test, and in what manner.

Beyond those very specific tests, I run the gamut of regular usage. Gaming for several hours, idle for several hours, entering and waking from sleep, shutting down, charging devices, connecting to external displays, bluetooth, etc. Those tests sometimes take just as long as a benchmark, but ensure actual usability of the laptop when overclocked or undervolted.

For example, running the memory at 3066MHz runs successfully every benchmark program I have, even the memory tests. The PC functions fine, acts normal. BUT...in games using the Vulkan API...it will crash after an extended load. A pretty specific scenario as I could just change API or play a different game and be "stable". But because I found 1 instability in 1 scenario, there are bound to be others hiding. So, using the breakdown of tests mentioned, you can find any instability in your system even if it appears stable.

What does your voltage settle at under load?

TS Bench 1024M 12 Threads pulls 57.5W on my system and sits between 0.9482 - 0.950v roughly.

That's with -0.225v core which seems to be the limit for my particular machine before it will not wake from Sleep mode.

It will benchmark and game all the way down to -0.350v (I did not try further) and appear stable. But because it cannot recover from Sleep I won't call it stable there.

 

Mine settles at 1.0162v and 60W. I use a -350mV on core, -125mV on cache and -100mV on GPU and slice.

 

Hey guys, I just wanted to say thanks for all the tips- I’ve learned a lot about computers in just a few days thanks to this thread and also YouTubing random computer videos lol.

Another question though - various people say to keep the CPU core and cache voltages the same, otherwise your computer might crash. Yet I see you guys (and some others) have different voltages for the core and cache. What gives?

 

As far as I can tell this comes down to the chipset. When I had a 7300HQ on Kaby Lake platform, the cpu core and cache were linked in XTU. If you undervolted one, it would match to the other. I believe this is why people say/said to keep them even.

However, as new platforms come out, the link between core and cache seems to have changed and the undervolt control no longer matches them. For the 8750H specifically, the cores seem to undervolt very well.

Someone may be able to explain it better, if I am even correct.

 

Sounds like your stock voltage is quite high but the chip is capable of a fair bit less.

 

Either that or Im wayyyyy past the point of lowering the core being actually affected.

 

That's easy to find out, see if it has an impact rasing it a bit.

 

Okay my CPUs gone weird. Im getting a PL 1 limit reason and my PKG power is dropping to 55W, forcing my clocks to 3.3 - 3.5GHz :\

 

Have you checked the FIVR - Disable and Lock Turbo Power Limits option? There are multiple power limits so for some laptops, you will need to check this option to disable the secondary power limits.

Sometimes ThrottleStop does not report the BCLK speed correctly. One of your previous screenshots shows a BCLK speed of 97.428 MHz.



When this bug happens, the multiplier will not be displayed correctly either. For the 8750H, in a test with all cores active, it should be showing at most 39.00 for the multiplier column. When this happens, click on the BCLK button a few times and you will usually see 99.768 MHz or 99.769 MHz. Each time you click the BCLK button, TS takes another stab at figuring out the correct BCLK.

This screenshot in your previous post looks OK.

Except for the throttling.

Edit - There is no need to set the turbo power time limit to 3670016 seconds. Default is 28 seconds. ThrottleStop may let you set this sky high but the CPU might ignore that request. This time limit controls how long the CPU will run at the Short term power limit before it switches to the Long term power limit. If you want the CPU to use the same power limit indefinitely, just set the long and the short power limits to the same value.

Glad to see that ThrottleStop gets the important stuff right! Nice work. Thanks for sharing all of your results.

Great to see a laptop with the 8750H that actually has proper cooling. I was just reading about a Dell G7-7588 laptop, same CPU, and it could not even break through the 1000 point barrier in Cinebench 15. What a joke.

 

Guys I’m very confused...

So last night I ran Cinebench and scored around 2500. This was my stock score without any modifications at all.

Today I took off the old thermal paste and applied some of the Noctua NTH1 to the CPU and GPU. I applied around a pea-sized amount to the CPU and a thin line to the GPU since it’s a little bigger. I also installed 16 GB Crucial RAM.

I ran Cinebench and only got around 1800! I then undervolted the core and cache to -100mv, which brought it up to 2100. I then realized I didn’t have the battery plugged in, but after I plugged it in it only went up to 2300.

All my temperatures are lower so I’m very confused. Did I mess up my computer somehow?

 

@initialize - Your CPU is throttling. If you are using ThrottleStop, have a look at the pictures attached to the first post in this thread for some examples. Open up Limit Reasons when your CPU is loaded and have a look for anything lighting up in red which will show you why your CPU is throttling. You can also use XTU to watch for throttling reasons. If you do not post any pics or examples of your CPU not running at full speed when benching, it is going to be hard for anyone to help you.

 

Ok I’ll take some pics later, but as far as I know my CPU isn’t throttling. I’ll double check later (I’m currently not on my computer) but throttlestop was showing that my cores were only reaching around 60-70 degrees Celsius while running Cinebench

 

Checked it, still not solved. I'm somewhat countering the issue by UVing my cache a bit more. Managed to bring it up to 3.65Ghz now.

EDIT: It doesn't seem to happen when I'm in games? It only happens whenever I benchmark...

 

Do you also have XTU installed? If so, try removing it, completely shut down and start up again.

Another thing to try is to do a sleep resume cycle. On some laptops, this will unlock the maximum power limit.

 

A lightbulb has just clicked on in my head. All this time I did not fully understand the relationship between TTL and PL1, PL2. Resetting to 28.

I'm watching reviews of 9th gen Intels that bench 2500-2800 out of the box. I'm glad the 8th gen is still viable and it's a good cost savings paired with an RTX card in a properly cooled laptop!

 

Help me guys lol. So for some reason I'm throttling at idle and while running cinebench. Max temps only reached 61 degrees C

 

What are your TPL window settings? You are only pulling 30W during the Cinebench run which is low for the 45W TDP rated 8750H. Other than not having Turbo & Lock Limits Disabled, the FIVR windows looks fine to me for average use!

 

I think this started happening right after I applied the thermal paste and installed more RAM... would this have caused this problem somehow?

 

A bad thermal paste application would make itself known in temperature. You'd see a difference core to core, or an overall rise in temperature across all cores. If you are not thermal throttling, but power throttling; I don't think it's thermal paste related.

Are you running on battery power? Where is your battery percentage indicator in TS? Running off battery would cause this kind of limit I think.

 

Sorry to be a pain, but how do I check this?

 

@initialize - In ThrottleStop, EDP OTHER throttling is usually caused by the current limit being set too low. In the first post of this thread, @4W4K3 has IccMax maxed out at 255.75. You have yours set to 45. This needs to be fixed. I would also check the FIVR - Disable and Lock Turbo Power Limits option in ThrottleStop.

Also change the Turbo Boost Long Power Max to at least 60 and do not check the Clamp option. The default setting of 45 will limit max performance.

Your temps are OK and appear to be consistent core to core so I think your paste job is good.

I like seeing your very low C0% number when your laptop is idle. That is a sign of not too much background crap using up CPU cycles.

In the ThrottleStop Options window you can turn on Battery monitoring and Nvidia GPU monitoring if you want to look at that info.

 

Thank you guys a ton! I’m off my computer now - on an unrelated note, I’m having surgery tomorrow so I might not be able to do these things for a couple days until I’m feeling better - so I’ll let you guys know if I (hopefully) fix these issues soon!

 

Poor application of paste can be checked by removing the heatsink and checking the pattern it leaves behind.

Good luck with surgery

 

Tried some more tweaking of GPU clocks and managed to score a 8201 in Time Spy (5th best for the 8750H/2070 notebook combo): https://www.3dmark.com/spy/7521034

Disabling gsync seems to bump Fire Strike by about 200pts or so (probably in the margin of error) but I didn't see much difference with Time Spy, although that 8201 was with gsync disabled.

I also tried disabling the "GPU Performance Scaling" in the bios to see what effect it had. According to some other threads here, it tries to balance the power in the system by slightly scaling back the CPU and letting the GPU scale up a bit more. Disabling did seem to reduce GPU scores a bit and increase CPU scores a bit, but it wasn't much and overall scores were similar. I wasn't diligent enough to run more than a couple benchmarks though, so that all could be within the margin of error as well. =)

 

Thanks for that testing! I had not tried disabling the GPU scaling before. Might be worth some more testing for ultimate CPU benchmarking. Your graphics scores are really high, I cannot consistently break 8K but hit high 7K's consistently. I want to try a better thermal paste that will keep me in the 60C's rather than normally seeing 70C or so.

I wanted to report; MEMORY update! I got it to boot at 3200MHz! I have a lot of testing to do, but I think I figured out the wall I was hitting in regards to timings. tREFI, tRCD, and CL are all very sensitive. I have AIDA64, Geekbench, and mmetest all to verify stability beyond normal usage. Should be back this evening with more solid numbers.

 

I've also updated the original post of this thread to include two videos. I will link them here so you don't have to scroll back as well.

 

It still seems PC3000 speed is the most stable for my mismatched memory sticks. I am convinced if I had a real "kit" of performance memory it would be very easy to overclock on this platform. I'm still pretty tickled with the overclock I was able to achieve on these non-paired sticks.

My 'GPU Performance Scaling" testing has resulted in a 300-400 point change using 3DMark Time Spy. Enabled I can consistently score 7900 or so, and Disabled I consistently score 7600 or so. What I'm not seeing is the CPU difference. I believe because I undervolted effectively, the CPU can draw full power regardless of the Performance Scaling. However, it does make an overhead difference for the RTX 2070 to have it enabled and occasionally draw that extra power.

 

Interesting. I checked your profile Time Spy result, and your CPU score was quite a bit higher than mine. (7100 to 6600 for my best Time Spy). My best CPU score was 6900 or so.

I also noticed that all of the overall scores higher than mine had a lower CPU score and a higher GPU score. So at least for Time Spy, it appears that more power/resources/TDP to the GPU is better.

 

I'd be curious to see what the actual GPU power draw was during benchmarking. You can watch the CPU pkg power through TS, but I don't have a program like that to show how much of the 115W RTX 2070 is actually being drawn.

Unrelated; I did a quick test comparing -0.230v core and -0.350v core running TS 1024M 12-thread test and each test scored within margin of error while the voltage remained exactly the same (0.9450v-0.950v).
This should verify there is some point of no return on each processor that varies a bit. It will actually benchmark and game all day at -0.500v but it is an illusion as the voltage is the same as if I were at -0.230v. So definitely go slow incrementally and find that sweet spot when undervolting!

 

Each chip will be slightly different too, just watch for the frequency possibly being different.

 

I managed to eek a few more points in Time Spy, up to 8293: https://www.3dmark.com/spy/7540518 .

I tried restricting the CPU a bit (disabled short term turbo, reduced long term turbo TDP to 45W, along with the undervolt, and still ended up with pretty high CPU scores. I managed one run with a CPU score of 7100+, despite all that.

The 3 scores of ahead me(for that CPU/GPU) all show 2080 eGPUs (with 2070 Max-Q strangely) in the details, so I'm declaring a moral victory. =)

 

Thank you guys, I fixed my power throttling problem!

Should I change my CPU cache IccMax to 255A as well?

 

It should not overdraw.

 

Well I think I found what's causing my throttling issue thanks to HWINFO. No matter how I change my TPL in Throttlestop, HWINFO says it's always at 55W and 90W.
It might be due to the PPM driver I installed when I reinstalled windows (since it says it changes the power settings) but I have zero clue how to uninstall it.

 

- delete -

 

That's very odd. Wasn't it working previously?

I can confirm changing the TPL settings in TS changes the HWiNFO for the 1720 model. I've had it everywhere from 45-90W.

 

When you set it like that, can it use 70W indefinitely? Really cool if TS lets you override the default power limits on this machine