After a very basic overclock (+120 clock, +1000 memory and no fine-tuning) I can reach 8900 in time-spy graphics score. But I can't go higher with fine-tuning on the curve. I've seen that the top score on 3dmark timespy, with same CPU and GPU, it's over 10'000 on graphics score but I can't explain myself how is that possibleGPU overclock won't make any more heat. You are limited to 115W just like me. You'll just get more frequency at a lower voltage. In the GPU tests the CPU isn't taxed at all, and the GPU temperatures are easy to keep in control. On mine OCing the Vram was the most performance/watt boost by a lot.
The free 3DMark tests are annoying. Unigine tests are better, especially Heaven which loops... and can be run in windowed mode which is great for tweaking settings on the fly.
8500 is more than I expected. Someone told me the 2070 model stock was about the same as my 2060 with the OC (~10% faster than the 2060 stock). 8500 is ~30% faster! That would almost be worth paying real money for
-
Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
Laptop undervolting and overclocking results... i7-10750H RTX 2060... MSI Raider GE75
- Thread starter rruff
- Start date
@rruff made a (maybe obvious) discovery:
after shifting from 84 to 0 the Speedshift-EEP in Throttlestop, graphics scores jumped up from 9050 to 9227.
Suggestion: in the afterburner curve, try to rise the low voltage values before the high ones. A good start-point is to watch the 3dmark / other benchmark report and choose the MHz(V) values on the curve corresponding to the most used frequency steps during the benchmark itself
after shifting from 84 to 0 the Speedshift-EEP in Throttlestop, graphics scores jumped up from 9050 to 9227.
Suggestion: in the afterburner curve, try to rise the low voltage values before the high ones. A good start-point is to watch the 3dmark / other benchmark report and choose the MHz(V) values on the curve corresponding to the most used frequency steps during the benchmark itself
You're getting a very good score! Almost makes me wish I'd sprung for the Raider 2070 Super when it was on sale. You are now 26% faster than my OC score. CPU score is basically the same.
One thing I noticed is that your vram clock is quite slow... even slower than mine, and I have the crappy low voltage vram that Nvidia switched to in 2020 on the 2060s. The way Timespy measures it's ~1690 Mhz. Have you maxed that out? On mine it was a much bigger FPS/W bonus than the GPU clock, and since we are wattage constrained, that's very important.
Unigine Heaven is a good one to use for tweaking on the fly. You can run in windowed mode and it loops... so you can adjust your vram OC in real time til you get artifacts, then back off til there are none... ever. The free version of Timespy is annoying to use, because it makes you go through a demo each time... Superposition is another good one to check.
just a 3dmark error, the memory was at 2000 MHz!
Hello,
FIrstly, I would like to say thank you all for this informative thread. I followed all the references and was able to boost the performance of my 10750H by almost 20%, according to CinebenchR20.
The chassis is MSI GF75 with mediocre cooling ability, so I was only able to run it with full load up to 67W before the CPU temp reached thermal throttle stage (I think MSI set it at 95).
At 67W, the core clock maxed out at 4090MHz for the first 2/3 of the test before the chip automatically drop back to 54W and 3800MHz (I thought the TDP is 45W, so it's weird), so I set all cores in FIVR up to 41. The rest of the settings in FIVR as you can see are simply trial and error.
Before performing the run on CinebenchR20, I reboot the system, set the fan at max (6000RPM), select power option to Ultimate Performance (I dont think It does anything for my case), turn off all non-essential, even dim the screen to minimum... The best score I got is just below 3090 and I think I that's my limit in this setup.
Any suggestion / input is very welcome,
Regard,
FIrstly, I would like to say thank you all for this informative thread. I followed all the references and was able to boost the performance of my 10750H by almost 20%, according to CinebenchR20.
The chassis is MSI GF75 with mediocre cooling ability, so I was only able to run it with full load up to 67W before the CPU temp reached thermal throttle stage (I think MSI set it at 95).
At 67W, the core clock maxed out at 4090MHz for the first 2/3 of the test before the chip automatically drop back to 54W and 3800MHz (I thought the TDP is 45W, so it's weird), so I set all cores in FIVR up to 41. The rest of the settings in FIVR as you can see are simply trial and error.
Before performing the run on CinebenchR20, I reboot the system, set the fan at max (6000RPM), select power option to Ultimate Performance (I dont think It does anything for my case), turn off all non-essential, even dim the screen to minimum... The best score I got is just below 3090 and I think I that's my limit in this setup.
Any suggestion / input is very welcome,
Regard,
Sure that you have removed the power limits inside the bios?
1) completely uninstall dragon center & MSI SDK (you can gain A LOT uninstalling the two, dunno what it does on the CPU but it does it BAD) and restart the pc. If you want to tweak the fan, use silent option. If you want to tweak GPU, use afterburner. If you use the battery limit, you can set it before uninstall: it will stay the same after.
2) Start with a fresh folder/install of throttlestop
3) Keep the turbo as stock (50,49,47,46,45,43) , try -125mV on core and -90mV on cache and start from those values in -5mV steps if you want downvolt more(i suspect that -200 on core isn't applied at all from the proc.). Tick "Disable and Lock Turbo Power Limits".
4) SpeedShift EEP at 0, untick speedstep, tick c1e.
5) Run C23 with colerboost enabled, just to see the absolute maximum at fixed condition. C23 is a little more intensive on cpu compared to C20, so better for testing. Your goal is to see, during all the test, a stable clock of 4290 MHz. If all goes as it should, you should stay in the 8700-8850 range.
6) if you want to test CPU & GPU in game, @rruff suggested some good benchmark softwares. Personally for now I tried only the 3dmark suite (now on sale for a few dollars) and found that the port royal and dlss tests are the best (inside the suite) to find overclock limits
@unclewebb
A suggestion: in the past week, at the office, I found that I change a lot the prochot limit to make sure that the laptop stay absolutely in silence. It is possible to add a prochot value for every profile inside throttlestop? It could be very useful
Last edited:
It makes sense that the GF75 would have poorer cooling ability... but again there is the weirdness of you only being able to get 4.1GHz at 67W even with a bigger UV than mine.
The 45W TDP doesn't mean anything apparently. But I wonder if the 54W 3.8GHz limit (after ~60 sec?) is unique to your model. If you haven't tried yet, set the frequency to 3.9GHz and see if it still happens, and if it's always at the same time. Mine will power throttle (can't remember if it's 54W or something else) if I'm using the GPU and CPU at the same time, but it's after a couple minutes. On a pure CPU test it doesn't throttle.
Also, if you want to set your max frequency to avoid temperature and power throttling, you only need to adjust down the 6-core value to 4.1GHz. You can leave the 1 and 2 core values as high as possible and maybe drop the others down .1GHz.
unclewebb
ThrottleStop & RealTemp Author
- Joined
- Jun 1, 2008
- Messages
- 7,329 (1.26/day)
That sounds like a good idea. Maybe I will add that feature to ThrottleStop someday.
I tried your suggestions and here are a few things I found:
- At lower Watt setting (67W or less), tempt slowly crawl up to over 90, and it take longer for the 54W Power Throttle to kick in (around 45s ).
- At higher Watt setting (73W or more), tempt goes up fast to PROCHOT (95) and the 54W Power Throttle also kick in earlier (around 38s).
- I set the 6th core at 4.1 and the other cores at higher values (4.5 & 4.3). During the bench run, frequency seems to stuck at the lowest core (4.1) regardless.
Here is the screenshot of when the Power Throttle kick in, as you can see I only run CPU test only.
I did every step that you mentioned. MSI BIOS is definitely very user friendly.
Thank you for the tip on Silent Option. I got rid of Dragon Center as soon as I got this fan control alternative.
I found that Dragon Center took a damn long time to boot up (probably due to SDK loading). So it sure as hell has some negative effect on the CPU idle.
WIth some tweaks on the setting (4.2 GHz / 73W), I was able to break the 3100 bar with almost zero thermal limit, definitely a result I can go to sleep with.
Last edited:
Well... I just did the TSBench long test and discovered that my processor also power throttles.... but only down to 65W, which is close to the ~69W it uses at 12x 4.3GHz. It kicked in after ~110s when max temperature was ~90C. If I get bored I may look around in the bios and see if there is a way to change that. It's been long enough since I've been focused on this that I forget a lot of the things I've tried...
In Cinebench20 it never throttles with the single pass. Got a score of 3395 in that, which I think is the highest I've seen.
If you choose all cores in a benchmark then the CPU will use your settings for all cores. The reason for increasing the frequency at lower core counts is because you can... and you won't have thermal issues. A lot of software only uses one core, and many don't use them all. If you run TSBench with different #thread settings you'll see a difference... or C20 on single core.
Apologize ahead of time for continuously hijacking your thread, but the whole Nvidia forum doesn't seem to be as helpful as this thread alone lol. I spent hours trying to boost the performance of my GTX 1650 with little success (I know it's a crappy card but that's the only option came with the GF75 in my region, next best thing was 1660ti but only in a GF65 chassis).
So basically, I tried 4 different ways to unlock the Core Voltage control on MSI Afterburner and only the method that seems to work was modifying the "VEN_10DE&DEV_1341&SUBSYS_2281103C&REV_A2&BUS_10&DEV_0&FN_0" file in the Profile folder. However, after unlocking the voltae slider and moving it to any given position, as soon as I hit "apply", the setting went back to default (0). Bottom line is, I'm stuck with the first 3 unusable sliders.
Moving on to the funny part, I can let the Core Clock run at any position from -500 to +200 and the GPU would just ignore the value and crank it up to 1380MHz under light load and 1785MHz under heavy load. Core Clock will be be stabilized "forever" at 1785MHz if I give it a setting of at least +100 or so. I selected core +212, app crash under load. I selected core +220, everything crashed. If I uninstall the MSI Afterburner, I wont be able to monitor the core but I suspect it will automatically crank up to 1785MHz aswell. I was hoping to get close the 2000MHz for the Core Clock as some guys on youtube said they did on GTX 1650 desktop, but I highly doubt it's possible on this mobile version.
The only slider seem to work (fully under my command) is the Memory Clock. I can crank it up to maximum +2000 which give me total of 8000MHz. Since my GTX 1650 is the newer DDR6 version, I have 6000MHz base Memory Clock to start with instead of 4000MHz on the older DDR5 version. Last but not least, according to the spec sheets below, Core Clock 1785MHz is the boost frequency of GTX 1660, not 1650 DDR5 or 1650 DDR6, which does not make any sense to me ...
NVIDIA GeForce GTX 1650 Specs
NVIDIA TU117, 1665 MHz, 896 Cores, 56 TMUs, 32 ROPs, 4096 MB GDDR5, 2001 MHz, 128 bit
NVIDIA GeForce GTX 1650 GDDR6 Specs
NVIDIA TU117, 1590 MHz, 896 Cores, 56 TMUs, 32 ROPs, 4096 MB GDDR6, 1500 MHz, 128 bit
NVIDIA GeForce GTX 1660 Specs
NVIDIA TU116, 1785 MHz, 1408 Cores, 88 TMUs, 48 ROPs, 6144 MB GDDR5, 2001 MHz, 192 bit
Here is the screenshot of Afterburner when I give the GPU some load. I have seen the power usage up to around 50W under full load, not sure if It can go further without the working core voltage slider.
OC scan doesn't work for me. I tried a couple other MSI Afterburner versions and all show same error. I suspect it only works with RTX cards.
I can open up the curve but not sure what to do with it without a proper OC scan.
Google that error and see if you can learn anything.
From what I can see of the graph it appears to be applying your OC. Try running Heaven in windowed mode and use HWinfo for monitoring while it runs. I'd start with something smaller like a +100 OC. If +2000 is stable on vram, that's pretty awesome.
BTW... found several instances of people running the scan and OCing a 1650 laptop, but YMMV. AB isn't supported on laptops so it doesn't always work.
From what I can see of the graph it appears to be applying your OC. Try running Heaven in windowed mode and use HWinfo for monitoring while it runs. I'd start with something smaller like a +100 OC. If +2000 is stable on vram, that's pretty awesome.
BTW... found several instances of people running the scan and OCing a 1650 laptop, but YMMV. AB isn't supported on laptops so it doesn't always work.
Last edited:
After alot of reading on other threads, I can confirmed that the 55W long term power limit is enforced by MSI on (AFAIK) the GF Thin series. I've seen some GS users figure out a way to trick their BIOS to remove such limit ... Well, i'll dig into that at a later time.
Believe it or not, I thought about this line for a very long time. At the default settings of TS, I could not get all 6 cores to run 4.3GHz without at least 80W on PL2 (BIOS says I can push 91W so no problem there). Even then, as you can guess thermal was bad and R20 score didnt improve very much. So I was happy to keep my cache at around -102.5 for awhile.
Recently, I have been playing around with the Cache Ratio min/max and Speed Shift min/max on TS and I found out several interesting tweaks:
- If you set the Cache Ratio at a lower number, you get more Mhz out of the same Watt input. At the same time, stability also decrease (the opposite is true).
- Cache Ratio will go no lower than 9, even if you put 0-8 on min.
- Cache Ratio will go no lower than the min value on Speed Shift minus 3 (If I set Speed Shift min = 26 and Cache Ratio min = 1, the Cache Ratio = 23). There are many other threads on the webs about this topic of Cache Ratio vs Multiplier so I wont go into details here, but most of them intel users also recommend Cache Ratio to be no more than 300Mhz below the CPU operating frequency. "light bulb"
- More negative cache offset = more Mhz out of the same Watt input. At the same time, stability also decrease. (This one is nothing new, but It's related to the first tweak)
So after some deductive resasonings, I come up with a theory that: I can pull the Cache Offset slider further to the left to get more Mhz out of the same Watt, at the same time increase the Cache Ratio to get more stability. Now I can run 6 cores at the higher frequency that I want with the lower PL2 limit, therefore reduce alot of heat and delay the point of thermal throttling. Below is my initial testing and result:
Here is the screenshot before (Cache Offset = -102.5 / Cache Ratio = 12 / Speed Shift min = 12)
Here is the screenshot after (Cache Offset = -125.0 / Cache Ratio = 23 / Speed Shift min = 26)
As you can see, the R20 score hit a new record for me. I now can run 4.2 Ghz all core with less than 72W vs full 75W previously. System run very stable so far, both in or out of games, so I believe this is not the limit. We all know that we can not just pull the slide forever to the left, there is always a point of diminishing return, and I dont know if I'm there yet. I need a lot more testing as there are more variables on the table now, but this is very promising.
Enjoy,
- Joined
- Feb 18, 2012
- Messages
- 2,715 (0.61/day)
| System Name | MSI GP76 |
|---|---|
| Processor | intel i7 11800h |
| Cooling | 2 laptop fans |
| Memory | 32gb of 3000mhz DDR4 |
| Video Card(s) | Nvidia 3070 |
| Storage | x2 PNY 8tb cs2130 m.2 SSD--16tb of space |
| Display(s) | 17.3" IPS 1920x1080 240Hz |
| Power Supply | 280w laptop power supply |
| Mouse | Logitech m705 |
| Keyboard | laptop keyboard |
| Software | lots of movies and Windows 10 with win 7 shell |
| Benchmark Scores | Good enough for me |
On the ge75 with the 10750h I get 4.3ghz 100% at 62w with a .70v undervolt in the msi bios.
unclewebb
ThrottleStop & RealTemp Author
- Joined
- Jun 1, 2008
- Messages
- 7,329 (1.26/day)
I think you mean a -70 mV undervolt which is -0.070 V.
No CPU is going to be stable with a -0.70 undervolt.
- Joined
- Feb 18, 2012
- Messages
- 2,715 (0.61/day)
| System Name | MSI GP76 |
|---|---|
| Processor | intel i7 11800h |
| Cooling | 2 laptop fans |
| Memory | 32gb of 3000mhz DDR4 |
| Video Card(s) | Nvidia 3070 |
| Storage | x2 PNY 8tb cs2130 m.2 SSD--16tb of space |
| Display(s) | 17.3" IPS 1920x1080 240Hz |
| Power Supply | 280w laptop power supply |
| Mouse | Logitech m705 |
| Keyboard | laptop keyboard |
| Software | lots of movies and Windows 10 with win 7 shell |
| Benchmark Scores | Good enough for me |
I meant to say -70mv.
- Joined
- Feb 18, 2012
- Messages
- 2,715 (0.61/day)
| System Name | MSI GP76 |
|---|---|
| Processor | intel i7 11800h |
| Cooling | 2 laptop fans |
| Memory | 32gb of 3000mhz DDR4 |
| Video Card(s) | Nvidia 3070 |
| Storage | x2 PNY 8tb cs2130 m.2 SSD--16tb of space |
| Display(s) | 17.3" IPS 1920x1080 240Hz |
| Power Supply | 280w laptop power supply |
| Mouse | Logitech m705 |
| Keyboard | laptop keyboard |
| Software | lots of movies and Windows 10 with win 7 shell |
| Benchmark Scores | Good enough for me |
Not really, I use liquid metal for TIM. It helps with the temps. -.70mv is about average for undevolt. I saw a few people use .90mv and higher without freezing.
Good to hear. Im still on factory paste, so there is room for improvement after warranty period.
- Joined
- Feb 18, 2012
- Messages
- 2,715 (0.61/day)
| System Name | MSI GP76 |
|---|---|
| Processor | intel i7 11800h |
| Cooling | 2 laptop fans |
| Memory | 32gb of 3000mhz DDR4 |
| Video Card(s) | Nvidia 3070 |
| Storage | x2 PNY 8tb cs2130 m.2 SSD--16tb of space |
| Display(s) | 17.3" IPS 1920x1080 240Hz |
| Power Supply | 280w laptop power supply |
| Mouse | Logitech m705 |
| Keyboard | laptop keyboard |
| Software | lots of movies and Windows 10 with win 7 shell |
| Benchmark Scores | Good enough for me |
My laptop is still on warranty period.
Are you in the US?
- Joined
- Aug 6, 2010
- Messages
- 590 (0.12/day)
- Location
- Hà Nội, Việt Nam
| Processor | Intel Core i5 12400F Alder Lake Hex-Core Processor @ 2.5 GHz |
|---|---|
| Motherboard | Asrock B760M Pro RS/D4 |
| Cooling | Thermalright Assassin 120X Refined SE |
| Memory | 16GB DDR4 2666MHz Dual Channel Corsair Vengeance LPX |
| Video Card(s) | Asus RTX2060-DUAL-O6G EVO (NVIDIA GeForce RTX 2060) |
| Storage | 1TB Samsung PM9A1 NVMe + 500GB Crucial P3 NVMe + 2TB WD Black P10 External HDD |
| Display(s) | Asus VG278Q TN Full HD 144Hz Gaming Monitor |
| Case | MSI MAG Shield 110R |
| Audio Device(s) | Logitech G Pro X Wireless Gaming Headset |
| Power Supply | CoolerMaster MWE Bronze v2 450W |
| Mouse | Razer Viper V3 Hyperspeed w/ Hyperpolling Dongle + Razer Deathadder V3 Wired |
| Keyboard | VGN VXE ATK68 Gateron Mechanical Hall-Effect| Monka 3075 V2 Pro Cherry Powder |
| Software | AtlasOS v0.4 |
Glad to see a fellow countryman here in this forum. Nice little journey you have here in the topic, and a warm welcome to the forum as well (if you haven't got one!)
True. The 62W @4.3GHz is the unusual part, and the paste won't help with that... but as uncleweb has stated the self reported power isn't reliable.
Did you run benchmarks before/after to compare thermal performance?
Xin cảm ơn. Best tech forum I've been to in years.
Thank you again to rruff and unclewebb for all the contributions, you all helped turning my overpriced electric frying pan into an adequate gaming machine.