Nice thank you very much for the best OC tool!
Small suggestion for futur version: showing average and max value with OSD in addition to the curent value (CPU: current, average, max)
Well, this is a strange issue. When I initially started the program, the interface was tiny so increased scaling to match Windows 200% at 4K and the size is back normal. Previously 100% scaling was the correct size. Next, I can click on the Title bar and move the app around the the desktop, but if I click on a empty area of the desktop, the Title Bar becomes INACTIVE, in other words and cannot click and hold the title bar and move the app. I have to click a button on the interface to make the Title Bar active.....
Never mind, I decided to leave the above there but to add, if you have an issue with a skin, tying changing skins then changing back or essentially reloading the skin. That worked. So the Title bar is fine after reloading the skin. The scaling at 200% is normal size at 4K. I'mm not sure if I read the notes correctly on whether 200% scaling at 4K would be normal sized instead of still magnified some percentage.
So, one new feature of this is automated OC Scanner for Pascal cards, not just limited to Turing cards like it was before. I tested it this afternoon on my GTX 1070, I get higher overclocks just by using my own manual overclocking efforts. OC Scanner gives me +62Mhz on the core, but my manual highest stable overclock is +100Mhz. Interestingly they also have a "Test" feature in the OC Scanner where you can just test your manual overclock, by clicking the "Test" button - interestingly it says at stock clocks my card is "Test completed, confidence level is 90%", so it thinks it's not 100% certain that my card is stable even at stock clocks (which is not true). At my max manual overclock of +100Mhz it also says "confidence level 90%", but the most interesting part is that if I run +113Mhz through the test which I know to be unstable then this returns "confidence level 85%" - so to me it seems that for Pascal cards you might want to aim for a "confidence level 90%" when testing your overclocks using that application. Here's a screenshot of the test, and you can see it uses a hell of lot power through the GPU at times, blips lasting a few seconds of 250W on just a GTX 1070!
http://forum.notebookreview.com/attachments/oc-scanner-jpg.167427/
Well, this is a strange issue. When I initially started the program, the interface was tiny so increased scaling to match Windows 200% at 4K and the size is back normal. Previously 100% scaling was the correct size. Next, I can click on the Title bar and move the app around the the desktop, but if I click on a empty area of the desktop, the Title Bar becomes INACTIVE, in other words and cannot click and hold the title bar and move the app. I have to click a button on the interface to make the Title Bar active.....
Never mind, I decided to leave the above there but to add, if you have an issue with a skin, tying changing skins then changing back or essentially reloading the skin. That worked. So the Title bar is fine after reloading the skin. The scaling at 200% is normal size at 4K. I'mm not sure if I read the notes correctly on whether 200% scaling at 4K would be normal sized instead of still magnified some percentage.
From the release notes:
- Multilanguage user interface library is DPI aware now. Main window skinned interface is no longer being scaled by OS and no longer looks blurred by default when greater than 100% DPI is selected. Now main skinned window interface can be manually zoomed with skin scaling slider in “User interface” tab, and the properties are automatically scaled by OS according to selected DPI
So, one new feature of this is automated OC Scanner for Pascal cards, not just limited to Turing cards like it was before. I tested it this afternoon on my GTX 1070, I get higher overclocks just by using my own manual overclocking efforts. OC Scanner gives me +62Mhz on the core, but my manual highest stable overclock is +100Mhz.
Wrong. It is bas idea to compare _average_ overlocking returned after scanning (62MHz) to fixed +100Mhz which you manually applied before. Scanner creates _non_linear_ overclocked VF curve, so the offsets can be different for different V/F curve points. For example you may have +10MHz on one point and +114Mhz on another one with average result of 62MHz. That's why it is also bad idea to use average overclocking reported by scanner as manual overclocking offset for whole curve.
Robbo9999:
Interestingly they also have a "Test" feature in the OC Scanner where you can just test your manual overclock, by clicking the "Test" button - interestingly it says at stock clocks my card is "Test completed, confidence level is 90%", so it thinks it's not 100% certain that my card is stable even at stock clocks (which is not true).
It was documented a few times that NVIDIA thinks that it is too brave to declare that the system is 100% stable, so 90% is an absolute maximum you can see. Even on stock clocks. That's info directly from NV.
Wrong. It is bas idea to compare _average_ overlocking returned after scanning (62MHz) to fixed +100Mhz which you manually applied before. Scanner creates _non_linear_ overclocked VF curve, so the offsets can be different for different V/F curve points. For example you may have +10MHz on one point and +114Mhz on another one with average result of 62MHz. That's why it is also bad idea to use average overclocking reported by scanner as manual overclocking offset for whole curve.
It was documented a few times that NVIDIA thinks that it is too brave to declare that the system is 100% stable, so 90% is an absolute maximum you can see. Even on stock clocks. That's info directly from NV.
Now that last part is very cool info to hear from you - that 90% is the absolute maxium confidence level you can see, very pleased to hear that & it sheds a lot of light on it, thanks. I'm also impressed that it then detected my known unstable overclock of +113Mhz as 85% confidence level, so it did notice the distinction.
The first part of your response, fair enough (I understand), but I did get a higher manual overclock than what the OC Scanner could deliver - the actual clocks during benchmarking were 50Mhz higher with my manual overclock and the benchmark scores were higher too. But, it's good that OC Scanner has been included for Pascal cards, it's a way for noobs to hit a button and get a sizeable overclock, just with Pascal (at least) you can get a higher stable overclock by going old school with a manual overclock - for those enthusiasts that like to take the time to overclock & test themselves manually.
but I did get a higher manual overclock than what the OC Scanner could deliver - the actual clocks during benchmarking were 50Mhz higher with my manual overclock and the benchmark scores were higher too.
But didn't you kinda expect your own OC results to be better than the OC Scanner suggestion? I can't imagine any Auto Overclock tool that wouldn't be very conservative with the suggested settings.
But didn't you kinda expect your own OC results to be better than the OC Scanner suggestion? I can't imagine any Auto Overclock tool that wouldn't be very conservative with the suggested settings.
Well, yes, I did expect that, although one or two reviews of Hilberts of some Turing cards have shown higher benchmark scores from the OC Scanner overclock when compared to the manual overclock, but it was variable in his reviews, most showed higher results from manual OC though - made me curious to see what the OC Scanner could do for Pascal!
Well, yes, I did expect that, although one or two reviews of Hilberts of some Turing cards have shown higher benchmark scores from the OC Scanner overclock when compared to the manual overclock, but it was variable in his reviews, most showed higher results from manual OC though - made me curious to see what the OC Scanner could do for Pascal!
When it comes to clocking, most GPU oems will err on the conservative side, without a doubt. The specter of mass warranty returns causes them to lose sleep, I'd imagine...;)
OK, unlike my experience with beta 9, I'm having no problems so far with AB beta 10--which is nice!
When it comes to clocking, most GPU oems will err on the conservative side, without a doubt. The specter of mass warranty returns causes them to lose sleep, I'd imagine...;)
Yes, I imagine the first priority for an NVidia created OC Scanner is that any overclocks created by the tool would not cause instability.
Well this OC Scanner did its "SCAN" and it clocked my card over 2000 mhz with a curve which was NOT stable in BF 5 DX12... And i knew it wasn't going to be stable since i have clocked the card on my own @ slightly less than 2000mhz ever since i got it on the release date of GTX 1080 ti. On top of that, after it did its auto clock i did a "TEST" and it showed 90% confidence. So in the end, i would stick with manual and then trial and error in games. 🙂
Well usually things just work for me. But all I get trying to start the scanner is
19:51:57 Connected to MSI Afterburner control interface v2.3
19:52:01 GPU1 : VEN_10DE&DEV_1B06&SUBSYS_63903842&REV_A1&BUS_1&DEV_0&FN_0
19:52:01 Memory clock +0MHz
19:52:01 Overvoltage 100%
19:52:01 Power limit 120%
19:52:01 Thermal limit 90 °C
19:52:01 Fan speed 1 Auto
19:52:01 Start scanning, please wait a few minutes
19:52:01 Failed to start scanning!
If I click the test button without closing the program, I get Runtime Error.
If I just start the Scanner and hit the 'Test' button, I get
19:54:13 Connected to MSI Afterburner control interface v2.3
19:54:17 GPU1 : VEN_10DE&DEV_1B06&SUBSYS_63903842&REV_A1&BUS_1&DEV_0&FN_0
19:54:17 Core clock +0MHz
19:54:17 Failed to start testing!
Nice.
This screenshot perfectly demonstrates why average overclock reported by scanner can be used for estimating the result only and why it shouldn't be specified manually. Your GPU is able to take approximately 200MHz stable offset in lower voltages but it decreases to 100MHz on higher voltage range. So if you try to specify that average 160 manually, you'll get too low overclocking on low voltages but can get beyond the stability limits on higher voltages.
Nice.
This screenshot perfectly demonstrates why average overclock reported by scanner can be used for estimating the result only and why it shouldn't be specified manually. Your GPU is able to take approximately 200MHz stable offset in lower voltages but it decreases to 100MHz on higher voltage range. So if you try to specify that average 160 manually, you'll get too low overclocking on low voltages but can get beyond the stability limits on higher voltages.
Am I right in thinking that the main advantage of this variable overclock through each of the voltage points is that if you have a card that is often power limited (and perhaps also applying to temperature limited) then it could mean you could end up with an overclock that is higher performing than if you had just applied manual testing and a straight offset overclock? Because during power limited or temperature limited scenarios the card will drop to lower voltage points, and if the curve has detected higher overclock offsets at lower voltages (like in ultraex2003's post above) then this will end up in a higher clock at those lower voltages in comparison to a straight fixed offset overclock. Most desktop cards aren't often power or temperature limited during gaming though, so I suppose this effect I talk about is not particularly relevant, although I suppose it could be relevant to folks who decide to lower the TDP of their GPU for whatever reason. Thinking about it, I imagine this automated OC Scanner curve overclock would be more beneficial in laptops, they're more power & temperature limited.
NV Scanner component is failing on your system. I'm afraid I cannot help with that, it is not my development.
Well, that is precisely why I did not post it in the Afterburner thread nor ask for your help specifically. But thanks any way.
For others, the error message I see in event viewer is..
Faulting application name: MSIOCScanner_x64.exe, version: 1.6.0.0, time stamp: 0x5c223ef4
Faulting module name: MSVCR90.dll, version: 9.0.30729.9518, time stamp: 0x5b6909cf
Exception code: 0x40000015
Fault offset: 0x0000000000042686
Faulting process id: 0xe78
Faulting application start time: 0x01d49e60ade564e7
Faulting application path: C:\Program Files (x86)\MSI Afterburner\Bundle\OCScanner\MSIOCScanner_x64.exe
Faulting module path: C:\WINDOWS\WinSxS\amd64_microsoft.vc90.crt_1fc8b3b9a1e18e3b_9.0.30729.9518_none_08e07c8fa840efbe\MSVCR90.dll
Report Id: 71cea28a-36ed-484f-98b6-05fb518aa209
Faulting package full name:
Faulting package-relative application ID:
I'm going to do a DDU driver nuke and go from there.
It was documented a few times that NVIDIA thinks that it is too brave to declare that the system is 100% stable, so 90% is an absolute maximum you can see. Even on stock clocks. That's info directly from NV.