880m over clock score/problem narrowed down

Hi guys,

Over the past few days, some of you may remember that I had encountered problems with my 880m after I installed the latest beta driver. I installed all known 880m drivers, plus emulated the 780m. None fixed my problem. So yesterday, I decided to perform a clean install of Windows 8.1. Upon running 3D Mark 11, I got through the entire test without any problems.

My stock score - 3D Mark 11

After testing with my stock vbios, I decided to go ahead and flash Slv's and John's vbios. Bare in mind, this vbios was released a few weeks prior to the beta driver. Anyway, as some of you maybe aware, Nvidia recently added the Sky Diver profile to it's latest driver. I've run 4 different benchmarks. Only 1 out of those 4 times I tried running 3D Mark 11, did I manage to get through the entire test without a problem. I'll tell you why now.

Settings in Nvidia Inspector - all tests performed with driver 337.88

Core/Memory/Voltage - Power set to 160% and Temperature set to 93c (Prioritize Temperature - Unticked)
1101/3000/1.050v

Failed within 5 seconds of the first test starting (submarine test).

Core/Memory/Voltage - Power and Temperature set same as above
1083/2900/1.050v

Failed within the exact time frame of the first test I performed (submarine test).

Core/Memory/Voltage - Power and Temperature set same as above
1063/2800/1.050v

Failed within the exact time frame of the first test I performed (submarine test).

Core/Memory/Voltage - Power and Temperature set same as above
1101/2600/1.062v

Score - 3D Mark 11

I've come to the conclusion that this may be memory related. It seems that I'm having to up the voltage to an outstanding 1.062 v in order to get through benchmark testing with just a slight +100 over clock on memory. If I just boost the clock, I get no problems what so ever. I should also mention, that I found out why I'm unable to over clock after restarting my AW 17. In HWINFO, my GPU is up and running, but the core clock isn't even firing up without opening a 3D program (game, nvidia control panel). As soon as I fire up a 3D program, my core clock fires up to 993mhz, and then I'm able to over clock once again. Bare in mind, that this doesn't happen on my stock vbios. My core clock fires straight up as soon as I boot back into Windows. This only happens with the modified vbios. Obviously that leads on to this being the vbios at fault. But where did this issue come from? I'm placing my money on the beta driver adding something. It's to do with the memory as far as I can tell. My temps never reached over 79c during testing, so I know it's not temp related. I also noticed that my core clock never boosts over 954mhz when running with my stock vbios. However, before this latest beta update. My core clock always boosted to 993mhz during benching. Suffice to say, I place my money on Nvida doing something to the 880m with this latest beta driver, which in turn, has completely crippled our chances over clocking the memory since no one here want's to run with a voltage sky high.

This is just 1 example of how my single 880m performed before installing the latest beta driver.

Core/Memory/Voltage - Power and Temperature set to 160% and 93c (Prioritize Temperature - Unticked)
1093/3000/0.050v

Score - 3D Mark 11

So from that score, I can 100% say that the beta driver did something.

Also, take a look at my Sky Diver scores (modified vbios but with default settings)

Driver 337.88 - Sky Diver

Driver 340.43 - Sky Diver

Notice that there's a slight drop in score with new the beta driver. That shouldn't be happening since driver 340.43 was suppose to be optimised for Sky Diver. Nvidia definitely cut corners, or implemented something with this latest beta driver.

I'm now done with testing until a solution comes our way. I've put so many hours into testing to come up with a solution these past 4-5 days. I'm so tired at this point brothers. It's time for me to start actually enjoying this AW 17. Until John or Slv come up with a solution, there is nothing more any of us can do, unless you specialise in that specific modding department. It a shame that it came to this. I was really enjoying number chasing, and comparing scores with brother Fox. I'm unable to do that now, but I do hope that some day soon, we can all start fully enjoying our 880m's once again.

My thanks go out to - Mr. Fox, Ethrem, members over at TechInferno, and most of all, John and Slv for giving me the pleasure of over clocking these past few weeks. It's been pleasure working by your sides guys. I hope to do it again sometime soon.

 

nightdex, your testing is a little haphazard and all over the shop. You were having crashes in 3DMark11 because your overclock was unstable, that's not unusual, especially at the really high core overclocks you ran. It's probably therefore no coincidence that the only one that was stable was the 1.062V run. The 1.062V just refers to the core voltage, you can't change the voltage of the memory, so your conclusion that raising it to 1.062V made your memory overclock stable is not really possible. I didn't read the rest of your post, because I know it's related to the whole "880M latest driver issues" that I've decided not to be involved with, but having a better test methodology for your overclocking process is important for you to arrive at stable overclocks, you're changing too many things at once (core Mhz, core voltage, memory clock - just change one thing at a time & test stability), and I think it's causing you to draw invalid conclusions.

(And why are you not 'Prioritising Temperature' as a checkbox option when you overclock. That's the sensible choice, doesn't limit the current then, and will downclock at your temperature target).

 

No, your not understanding me. I've already run multiple tests before this. I've been doing this for weeks. I came up with those settings because they actually worked really well at a stable level before installing the latest beta driver. The settings above were suggested to me by multiple member on here around 3 weeks ago. So no, my settings are stable, and I've proven that multiple times before I installed this latest beta driver. However, I will enable (Prioritising Temperature) to see how thing's go.

Edit: Ok, so I tried your suggestion by 'Prioritising Temperature'. I changed just my core clock to 1013mhz, left voltage and memory as stock. That went fine. I then upped the memory by 2520mhz. 3D Mark 11 crashed straight away, Nvidia driver crash prompt appeared. I then added 12.5 voltage. I got through the entire test with the following.

1013/2520/1.012v

I never upped the voltage at those minimal settings before installing the update. My 880m was perfectly stable at those settings without +12.5 added in voltage before updating to the beta. Either way, I've been running with the settings at the top of the thread for 3 weeks. I never had a single problem until installing 340.43. So once again, it's something to do with Nvidia's latest driver. It's obviously had a knock-on affect on the vbios by John and Slv. Either way, I'm done with this whole thing now. It's giving me a headache. I'm sick and tired of messing around with it. It's done, the 880m is what one call "a bag of trash".

 

This is still helpful information. Thank you very much for posting it. The 1.062V required for 1100 on core is appropriate and about the same as 780M, so you're good there.

My 780M cards (all 4 of them) cannot go higher than 1500 (+500) on memory with the default memory voltage (which is not adjustable). Even 1525 is unstable for most things. Overclocking the GPU memory is also far more hazardous to the GPU than overclocking the core and the benefit from overclocking the GPU memory is far less significant that overclocking the core. The core speed has much greater influence on benchmark results. I normally prioritize Power Target over Temperature and link them with better results than I do from prioritizing Temperature. You should test both ways and decide what works best for your own needs. Unless you are having thermal control problems prioritizing Power Target (not temperature) should yield better performance and will most likely cause no issues with temperatures.

It looks to me like your 880M is doing a good job now.

 

It is, but it's definitely not up to the same standards that it was at before installing 340.43. I'm just glad that the stock speeds are running fine. I've noticed that my core clock is also crashing my driver when being altered now. In GPU-Z, it's showing as the PerfCap Reason is - Util, what ever that is. I do seriously think that the current modified vbios is funked all over for the time being. That driver must of done a number on the GPU in order for it to mess with the vbios that much. I'm guessing tables have been altered and what not.

 

You may need a little more core voltage then. I just checked and 1.062V is a little on the low side for 1100 on core for my 780M cards.

 

Wow. Can you give me your settings so I can test them my end please? This will tell us if the vbios is acting up.

 

That's weird, I thought that by 'Prioritising Temperature' (at the same time as increasing the power target), then it would use all the way up to your Power Target until the Target Temperature was reached. As long as you're below Target Temperature then logic would say you shouldn't get less performance by Prioritising Temperature. I thought Prioritising Temperature was just a convenient & safe way of inputting the maximum temperature you'd be happy with, and the GPU would only throttle when it reached that temperature. Is that incorrect?

 

Settings aren't applicable from one unit to the next, each GPU core will have different overclocking abilities in relation to voltage required. Other peoples results can only give a rough guide.

(I think your previously stable 1.05V 1100Mhz runs perhaps just weren't as stable as you thought, it can take hours of extensive testing, and gaming over a period of weeks to truly arrive at a 100% stable overclock when you're pushing right close to the limit of stability. Passing one run in a benchmark is not a reliable indication).

Although some people say that their max stable overclock can change with different drivers, so you might be seeing that too. I've never personally experienced the need to change my overclock with different drivers, maybe my overclock is not that close to the limit to warrant feeling that effect.

 

That's what I'm after, a rough guide.

 

Well, you've got a rough guide. 1.062V was a little low for 1100Mhz for his 780M's. How much voltage are you willing to put through them - 1.1V? In that case you've got 12.5mv steps all the way from 1.062V to 1.1V to experiment with to try to get your overclocks stable. That's only 3 steps of voltage to experiment with. Because overclocking results vary from unit to unit, then you've already got a rough guide, and you've got those 3 steps of voltage left to experiment with to fine tune things.

 

Wow - 3D Mark 11

I won't be keeping these settings for obvious reasons, but wow. Nice to see it come alive again.

1123/3020/1.087v

 

At extreme overclock levels prioritizing temps over power has resulted in throttling for me in the absence of high core temps. Once either GPU reaches 125-135% of TDP the core clock drops to a silly 2D level momentarily and then shoots back up. This is only visible through graphs or OSD, but it is enough of a moment to impact benchmark scores. Prioritizing Power Target minimizes this and normally prevents it from happening for me. It may not make any difference whatsoever at lower clock speeds, or it could help during extended gaming sessions... not sure of the effect under those conditions because with few exceptions I run stock clocks when gaming because that is plenty of horsepower for a great experience. So, as long as my gaming experience is good, I don't really expend any energy testing different settings or variations for anything. I only invest my time and energy on optimization and performance tweaks when I am number chasing.

 

I've been number chasing as of late. I have no idea why I'm doing it on my AW 17 and not on my AW 18. But, the 18 is sacred to me at this moment in time. The 17, well, I couldn't care less for. I kept this purely for number chasing. I guess a single card is the best way to go about it.

 

Hmmmmm...

Well you may well have narrowed down what my temperature issue was... I wasn't prioritizing temperature, I was prioritizing power.

Maybe it is worth going back to the modded vbios and doing some testing.

Project for when I get back from the doctor's appointment.

And dex, nice score man! You're so lucky, my RAM won't go over 2950 without throwing up pink artifacts and then freezing!

What should I set the power target to? Does it usually need a lot over 100%? It goes to 160% on the mod and I will admit I was jacking it up to 160% - I think my problem stems from eVGA's GUI not making it easy to see what is being prioritized (I didn't find out until yesterday that the arrow is clickable).

I may have totally had a huge blonde moment (and I'm not blonde lol) the entire time I was testing the vbios. With that said, even when I didn't touch the target slider, it was running hot. I guess more testing is in order. There's also the problem that I don't know what Clevo has the power target at on the stock vbios. Considering the cards don't pass 87C on it, its probably lower than 100% if I had to guess.

 

Set the power target to 160%. I've currently got a nice small over clock going on. I've run benchamarks 3D Mark 11 and Heaven. I've just run Murder Soul Suspect. Nvidia FPS counter is displaying 133 FPS on a 60hz screen? what the hell haha.

+60 core/+150 memory/+37.5mV

Score - 3D Mark 11

I have mine set at 160%. I personally use Nvidia Inspector as I find it nice and stable on Windows 8.1. MSI Afterburner won't even let me adjust the voltage slider, so that's out of the question, although, I would liked to actually retain my settings after each restart.

Good luck brother.

 

160% was causing run away temps though. I'll test out different settings though. I suppose it's entirely possible that Precision X is the problem too or even that flashing back to stock and flashing back to the mod will fix it.

Sent from my HTC One_M8 using Tapatalk

 

Open up your task manager. How many Voltage Tuners are running?

 

I'll check when I get home. I'm at doctors office right now. I know that I had no instances running when I went and had to close programs the other day to change a setting in the nVidia control panel. I haven't had any issues with it on either my desktop or my laptop unless that is what is causing the high temps.

Sent from my HTC One_M8 using Tapatalk

 

The thing to remember about power target is that if you reach it and it stays at the threshold for a little while the core value may drop (throttle) because the GPU is trying to exceed the limit. The GPU will try to correct itself to stay within the power target so it slows down to accomplish that because there is no other way it can stay within the power target you set.

Always try to make it at least a little more than what you see being utilized. For example, if at a given overclock you see 126% of TDP, set at least 130%. If you never see above 110% you can probably get away with 120% without encountering a decrease in performance. If you set it a lot higher I doubt that it does too much because the GPUs will only draw what they need in terms of watts. Voltage is different than watts and it's not so much on demand utilization. Essentially, whenever value is set for voltage is what the GPUs are going to be fed at 3D clock speeds. So, you can be giving them more voltage than necessary and this is more likely to increase temperatures than the power target being set higher than it needs to be.

If you are pushing the GPUs really hard having the power target set really high can coincide with increased temps and throttling, but that is a symptom of being pushed hard rather than a symptom of malfunction. In other words, don't blame the power target for overheating and throttling if your GPUs are doing what you have told them to do. Conceptually, this could also apply to CPU overclocking.

 

Mine needs +50 :/

NVIDIA GeForce GTX 880M video card benchmark result - Intel(R) Core(TM) i7-4940MX CPU @ 3.10GHz,Notebook P377SM-A

 

So what you're saying is, I don't need my power limit set to 160%?

What are you pushing +50? the power limit, or core? I don't understand why you are getting such low scores with a slight over clock. 1 single 880m does 8109 in 3D Mark 11 at stock speed - 954mhz. I've managed to get 9k out of my single 880m in 3D Mark 11. Bare in mind, this's also running an i7 4710MQ. With your specs, you should be at least hitting around 16000 in score at stock speeds.

 

The voltage, it froze at +37.5

I bumped the power target to 140% and it passed at +37.5 so I guess its a balancing act. With fans on max, one GPU hit 85C, the other hit 78C. I guess the cooling in this system just isn't adequate for overclocking because of the way they designed the secondary GPU's cooling :\

At this point I'm going to hang up the overclocking gloves and just use my system at stock GPU frequency. Its more than enough at stock anyway.

 

I use the 1.1v as a stock vbios lol never had an issue.. and thats on my 780s..

 

Etrem
You can maybe have a slight overclock (+50/100) . To reduce the lowest image flow during games.

 

It's a balancing act for sure. I followed the advice from Robbo and Mr. Fox. I'm seeing some nice scores to be fair.

1073/2650/1.037/110% - 3D Mark 11 Highest temperature - 75c

1093/2750/1.050/130% - 3D Mark 11 Highest temperature - 78c

1103/2770/1.062/120% - 3D Mark 11 Highest temperature - 82c

1128/2800/1.087/130% - 3D Mark 11 Highest temperature - 89c

I can push this further, but temps are against me at this point. So I shall leave benching for there for time being. I've just purchased some Liquid Ultra, so that should shave 4-5c off my temps. Being in the 80c's leaves nothing to desire. However, for benchmark purposes, I will push this single 880m as far as it will possibly go.

 

Maybe a more powerful CPU to your A17 ?. So you can push your A17 for a higher score. (Benching)? You've have a 330W power supply from your A18 "-)
I have no way to get a higher overclock my gpu than 78/200 Otherwise, my 240W power supply go to pieces. Already have maxed it out at 249.5 w already.

 

I've been in the market for a better cpu for a good month or so now. It's just getting a hold of one at a low cost. I'm not paying £400+ for a mobile CPU.

 

It looks like the price on Amazon and eBay are about +500 now. It is expensive. I originally had an i7-4900mq. There was something wrong with it and Dell could not and acquire a new one to me. They offered me a new machine but I wanted a new cpu. I nagged at them all the time. That's why I got an i7 4930mx from Dell instead "-). It looks like i7 4930mx overclock better in many cases than the i7-4940mx.

 

That's looking a lot better now for your 880M. Definitely need more CPU... here is a something to aspire to... 3DMark11 880M vs 880M Comparison (looks like you've got headroom for another ~1,800 3DMark graphics points)

 

Looks like another CPU's in order then. I'll take a look at some later today.

 

LOL!

Well I was able to match your clocks in Firestrike but it froze at +87.5 right when 11 started and +100mV popped the overload protection in my adapter. Pretty disappointing considering that I wasn't even overclocking the CPU, it was stock voltage and everything.

Anyway, this is what I got from Firestrike. I'm stuck until I get a converter box.

Result

That's a comparison between my new overclocked result and my previous - my physics score went up which was a surprise. This CPU pretty much decides when it wants to cooperate and when it doesn't.

Spoiler

Trying to run 3DMark 11 with 1108/2800/0.875 popped the other adapter.

3DMark 11 is much more stressful on power than 3DMark (2013) is, I'll be in contact with my reseller tomorrow about the converter box. If I can get it cheap or free, I'll do it. If not, I won't be using these clocks for daily use anyway so its not worth the extra money it would cost me.

Its nice to see that things are actually performing well now though. I agree with you nightdex that 340.38 was the problem. It seems like whatever it did was flushed out between going back to the stock vbios and 337.88 before reflashing the mod as my temps seem to be fine now and they freely overclock. Getting rid of EVGA probably helped too, I'm going to look at my desktop next.

Its too bad that we won't ever see our CPUs hitting those kinds of scores thanks to Haswell's heat but the CPU is generally not a big bottleneck for gaming once you pass 3.6GHz or so, I guess its not the end of the world.

Here's a comparison between my overclocked 780 Ti and my laptop: http://www.3dmark.com/compare/fs/2296599/fs/2343225

The laptop is 10.2% faster than the desktop in graphics score. It would take quite a significant additional overclock on top of the one my desktop card already has there to squeeze out that extra 10.2%, I'd probably need a modded vbios for my 780 Ti SC w/ ACX.

Going through my Firestrike scores, I forgot about this little gem

1153/2950 (most likely 1.1v as I had turbo boost disabled and I believe that right after this test I went and ran 3DMark 11 and thought I blew the adapter)

Graphics score was 14681 - overall would have been a lot higher if I could have turbo on for that overclock.

http://www.3dmark.com/fs/2327316

 

wait brother john has a new system???

 

Not that I know of, but check the date of the benchmark result. The link was for a run he posted with the 880M that he had for testing.

 

YEA but look at the cpu lol...

 

ah well im seeing 4940mx every where!

 

Somewhat off topic, but I couldn't help but laugh when I saw this after clicking on a link in a promo email I received from Alienware this afternoon. This is the first sentence of their description of the Alienware 18. Someone at Dell has a pretty twisted sense of humor.

 

That is kinda sadistic.....still, the 'humour factor' made me laugh even if the reality of what is going down doesn't.....and I don't even own an 880m....let alone two of 'em.