No more overclocking on Nvidia mobile GPUs

Yeah, I don't buy that DX12 theory. My GPU handles overclocking fine with 99% GPU load in the most demanding games - DX12 can't demand more than 99%! Responsible overclockers should already know to test the stability of their clocks with games that use 99% GPU load, not stuff like Total War, which has a severe CPU bottleneck and uses only, say 40-60% (just for the sake of argument) of the GPU.

Yeah, DX12 will increase GPU load on average but only for games that were CPU bottlenecked to begin with.

Edit: I was wrong (see my next post)

Sent from my Nexus 5

 

It's okay, you don't have to buy anything. Why don't you go load up Furmark and tell me more about GPU utilization?

 

Well I can't actually as I have Prema's modded VBIOS, which literally shuts down the laptop if you try to run Furmark! But surely you'd see 99%? Can't think it'd be CPU bottlenecked!

Edit: I just read this: http://www.overclockers.com/forums/...a5d1639e826c&p=6869254&viewfull=1#post6869254

I see that 99% load in GPU-Z does not mean actual 99% load on the GPU. I didn't know this. So looks like I was wrong, sorry. You learn something new everyday!

Sent from my Nexus 5

 

Got some preliminary numbers from MySn with regards to their survey (bolded some important info):

 

Only 50%?

 

Found this article on Yahoo's front page. It may have already been linked but the more outlets the better and Yahoo is not exactly a small place with few readers.

http://www.digitaltrends.com/computing/nvidia-inexplicably-deactivates-geforce-laptop-overclocking/


I am still blown away that Nvidia said it was a bug after ALL these years of overclocking Notebook GPUs. That has to be on the worst PR comments I have read in a long time. Would have been better just to say we intentionally shut it down for various reasons ?.......

 

Yes. GPU load (like CPU load) is very difficult to actually read. As I keep saying to people; XTU bench for example is not nearly as demanding as a livestream with heavy encoding or rendering out a video whose source was created using extremely fast compression with high bitrate for quality (for example: Shadowplay recordings). The power draw fluctuates all the time. I can run BF4 and a livestream and get clear up to 85%+ CPU utilization, but my power draw on it sits under 60W at 3.8GHz. I render a shadowplay video out in Handbrake and I'll use 65-70W. I run LinX Linpack benchmark (on 4 cores only; ignoring hyperthreads) and I'll pull 94W under heavy load... WHILE THROTTLING DUE TO LACK OF VOLTAGE (I could fix this if I used static voltage though; but my TDP limit at the time was 100W anyway, so...).

GPUs are the same. I managed to overdraw my power brick streaming Mass Effect 2 with 3.8GHz on my CPU and stock GPUs; just using an unlocked FPS in the game and my GPU load was only ~50-70% on each card. I have never been able to recreate such a load on my system in any gaming/streaming situation before. I tried with Dark Souls 2, I tried with CPU-recording in BF: Hardline, I tried all sorts of things. I never got my brick to trip like that again, but I know now I have to remove that OC if I ever stream ME2 again XD.

It's kind of why I'm annoyed that software doesn't tell us how much Watts the card is actually using, unless you have an MSI laptop and their software. That can list your actual system usage for you.

 

Thanks for the info. I had been blissfully ignorant of this whole aspect of overclocking - which is annoying as I have now become one of the very people that NVIDIA are using to justify their actions

Sent from my Nexus 5

 

That reminds me of what a dell engineer said on a conference call with some NBR members looking for a fix for the overheating 580m. He got hit with a lot of questions about the heatsink and fan profiles, but when someone mentioned GPU utilization he said 'What you (us) see is not actual utilization but an approximation based on frame buffer usage. They are often analogous but not always.' He then went on to say, 'I wish you could see the stats I have but the tool is NV restricted'. True or not? Seems there are hooks waiting to be found?

 

Well it's definitely true... I mean think about it. Why DOES furmark heat up a GPU far more than any game? It's not "realistic load", sure, but why? What's to say a game doesn't eventually make such a demanding load on it? I mean, it must be doing SOMETHING SOMEWHERE SOMEHOW to induce such a load on things...

I'm just happy; my slave GPU's max temp on furmark is 82 degrees after like 5 minutes =D. All I need now is a new heatsink for my primary and a solid CLU repaste and we should be in the same boat there =D. Or better, even.

 

Although very late, TechSpot's article on this issue is one of the best I've read so far: http://www.techspot.com/news/59788-...king-900m-mobile-gpus-feature-advertised.html

It seems that later articles seem to "get it" more often than the ones that came out right when this story broke. A lot of those early reports were just blatant click-bait for funneling traffic to their sites. Guess this tells us which sites have journalistic integrity and which ones will rush out a piece ASAP without really thinking it through just to get more hits.

 

Desktop 980s are throttling as well. They require modded vbios that changes rail max wattage. I have yet to care though for obvious reasons. Unforunately I will be trvalleing a lot soon and require a good enough notebook.

 

It's because FurMark is a GPU power virus. Some other power virus programs: OCCT, MSI Kombustor, PlanetSide 2. That's right. PlanetSide 2. At 99% GPU load that game heats up my GPU about 15C higher than any other game. People using desktop cards with an unlocked vBIOS have noticed it consuming almost 200% TDP.

 

I don't doubt it. PS2 never heated up my GPU all that much though, but I didn't hit even close to 99% scaling on it I'm quite sure, so that probably explains why. Hell, maybe ME2 is a power virus =D

 

I have seen a game ramp up temps without anything appearing to be happening. In AC:Unity, if I go to the screens to change weapons or anything around there, the GPU temp just keeps climbing. Usage was high but it always has been on the 680m for me. On my 680m it hit 90c (normal running 72c) before I backed out to an immediate cool down. I know the code gets panned for being poor but had I not been monitoring it who knows how high it would have got before shutting down?

 

bam! 600 signatures, continues climbing

 

Kepler GPUs cannot cross 92 degrees; they WILL downclock. You would never have shut off.

 

Thanks, at least I can put away the vision of a meltdown .

 

Kepler throttled at 95, Maxwell throttles at 92. Unless throttle points are different between desktop and laptop parts. The modded vbios on my 880s would not thermal throttle until 95 at which point it was a tug of war between 954MHz and 405.

Sent from my Nexus 6 using Tapatalk

 

It's 92c for Kepler with any BIOS unless it's the severely broken 880M bios at release.

 

The throttle cutoff point is 92, but your card could still generate excess heat and temps could rise a little higher; which is why the throttle point was set so low, so that there would be no chance of hitting the TJMax. If you opened nVidia inspector you'd see the "temp target" point would read 92 degrees, and you COULD try putting it to 93 but it never made a difference. I have had my card hit 94 before; where it kept going between 770-850 on a super hot day when Deus Ex Human Revolution Director's Cut only used single GPU and I really needed a repaste. At 92 it starts to downclock until it cools then it clocks up when it figures the cooling process is good enough. 880Ms were overly hot as an anomaly; so that makes sense you saw 95 degrees a lot.

Or I could get ninja'd by Zero989 who might have a better answer than I =D

Also, desktop cards have different throttle points per card. The 780Ti throttle point is only 87 degrees. I have no idea why it is so low, but I can confirm from a friend who has two reference 780Tis that it DOES throttle at that point regardless of whatever I put in NVI or NCP for him. Rest of Kepler was 92 as far as I remember.

I didn't know if mobile maxwell was 92 or not, but thanks for clarifying it for me =D.

 

Every single news site has got it wrong: it's not only 9XXM GPUs that cannot overclock - all Maxwell GPUs cannot. Maxwell includes the 850M, 860M, 960M, 965M, 970M, 980M.

 

This is because most websites
1 - Don't do research beyond what someone tells them
2 - Don't know jack about mobile market
3 - Don't care about mobile market

I've seen a couple sites that specify "mobile maxwell", but none list all five currently released and thus currently affected chips.

 

who made those cards? Mine is an EVGA reference card and I can change the throttle point through their software. And the 780ti has such a massive die. could be why its got a lower temp target. Still a great card though its a power hog. Got no reason to upgrade right now. Waiting to see what Nvidia jacks up next before I purchase another card.

 

I believe the cards are EVGA but I don't remember. Either way, that's the standard throttle point for them; even though you can change it in your desktop.

 

good to know ty

 

2) and 3) are why NotebookReview and NotebookCheck exist, aren't they? To expound to the world the glorious convenience and power of notebooks. Although desktops will be perennially more powerful. Can't help it. But we can help annoying, rich desktop owners with excessively powerful, electricity-guzzling SLI, watercooling and tri-monitor setups who troll 'us noob laptop OCers'.

 

You assume they'd bother with NBR and NBC. Honestly, both websites are minuscule compared to sites with the amount of traffic Linus Tech Tips, Overclock.net, Anandtech and Tom's Hardware get. Far less Tech Inferno; which is tens of times smaller in active users than NBR/NBC (though likely multiple times larger in expertise of users compared to LTT/OCN/Anand/etc).

Also, again, not caring about the mobile market means they'll never bother learning things. It's up to them to make the effort. Just like it was up to them to make the effort to learn how to build their own desktops, and to learn how to overclock, and about hardware, etc.

At least the benefit of us is that we know desktop and mobile hardware inside out, which puts us leaps and bounds ahead of them in general understanding.

 

Why the hell haven`t we heard anything from Nvidia? They issued out a statement couple of days after the 970 drama began. Now its been what, 2 weeks soon?

I tried to get PCPer to write an article about overclocking disabled on mobile cards because they have direct contact with Nvidia and could get an answer pretty quick or atleast push forward a change. But Ryan didnt even respond to my email.

This is not good

 

Why I love it here. I love to build desktops, and know that for the money between a laptop and a desktop a desktop would cost less to build to get comparable laptop power. (My $2500 MSI GT72 w/ a 980M could trade almost blow for blow with my $1500 Desktop that had an Xeon e3-1230v3 and GTX 770 in it, the price didn't include monitor or anything else. Desktop was probably closer to $2000 after all said and done.) BUT nothing beats being able to take my MSI WHEREVER I want to go. Plus it takes up far less space. Yet I can still play all my Steam games without much compromise on graphics settings. I sadly had to get rid of my desktop due to space being a premium where I now live. And thanks to NBR and such I was able to do some research on an offered laptop. (Alienware 17 w/ GTX 770M) Anyhow, between energy efficiency, less space needed and portability, a gaming laptop will always be first in my mind when shopping for a new computer. Not to mention I managed to get a couple friends to go gaming laptop in the past. They always went cheap laptop and cheap gaming desktop, then wondered why they where replacing their stuff every year almost.) Finally they both got gaming laptops and said they won't go back and that the price is worth being able to carry around something that they can game on.

Anyhow, I am still staying on 344.75 for my MSI so I can do an OC should I want to, and to also make a statement. I also downloaded 344.75 for the Alienware despite it having a 770m which shouldn't be affected, but how do I know if the newest version will break overclocking on it someday, which I need more on than my 980M. It's the principle of the matter. Will Nvidia keep it to the Mobile Maxwell cards or are Mobile everything next? Desktop may be next after that. I have my GTX 770M Overclocked by +135 on core, and +576 on memory and it means the difference of being able to play some of my games at higher fps or even higher detail. Though the 770m doesn't fully need the overclock on every game yet it does on occasion. Guess what the temperature gets to? 70C in an Ambient room temperature of almost 80F (26C). Hah! Eat that Nvidia! Can't overclock mobile cards?! Bull! I'm doing it perfectly fine on an older Kepler card and a newer Maxwell card, as many others are. I know what my next gpu will be and I'll give you a hint. The color is Red.
Not meaning to de-rail the thread, but is anyone having issues with the avatar images coming up, it shows the 'broken image link' picture. I'm also having issues with the image NBR uses by the address in the tabs of Chrome and Firefox. I have 2 computers that are doing the same thing. Random refreshes don't seem to fix it, and neither does a full shutdown. Just wondering if I'm the only one.

 

Excellent point - which isn't a very good omen. Nvidia isn't going to give a rat's fart about our petition or our needs. We are a relatively minuscule population - the enthusiasts and power users who prefer laptops. We are those in the very small intersection between those who want power, and those who want mobility.

As I've already mentioned, Nvidia cares mostly about profits - it's too big to bother with customer opinion and feedback, and it exhibits behaviour typical of a monopoly. AMD is becoming increasingly irrelevant in the gaming space (although Radeons are unmatched in terms of computational performance) and it is spreading itself too thin. It needs a major restructuring and a new outlook and target - its current practices are unsustainable. Perhaps R9 M4XX will satisfy our needs?

 

Could you post this on the OCN clockblock thread as I've stated the same before and its exactly what my thoughts are too!

Sent from my GT-I9300

 

Sure! I'll even re-run benchmarks and take screenshots of everything and post it. It may take a little bit, but it can be done. What screenshots should I do? I usually just did any for fun, not to prove a point, but now it's time to prove a point. I have Heaven benchmark, 3D Mark, as will as GPU-Z, and MSI Afterburner. Just want to make sure the right stuff and meaningful stuff gets posted!

Could you link that site to me please? Make sure I have to right one.

 

1 - This is because we're not desktop users.
2 - I'm fairly sure PCPer wrote an article already... It should be somewhere in this thread. Here: http://www.pcper.com/news/General-Tech/Bring-back-mobile-overclock-NVIDIA-stick-warning-it

 

Sure! Ill edit a link in this post in about 5.

Some benchmark pics like you suggested would be nice, in addition to some 3dmark scores!

If you put in a few pics showing max temps of the cpu and gpu during benchies thet would help as well! Make sure to post your overclocking results as well as the difference between stock and OC benchies!

This is really helpful!

Heres the link http://www.overclock.net/t/1540917/forums-geforce-no-more-overclocking-on-nvidia-mobile-gpus/0_50

Sent from my GT-I9300

 

If AMD does r9 m300 on 20nm, then you have potential competitor. Let's hope!

 

I heard in AMD's plans to switch 28nm TNMS to 28nm HP (High Perormance) of another manufacturer for APUs. But who knows about AMF GPUs.

 

Let's hope so too.. AMD's new CEO looks like a visionary and she's pushing it to diversify and improve.. Time needs to be given.. Also, the main problem for AMD is revenue.. People here who say we support AMD need to buy more AMD products.. I regret selling my 7970M now because I could have got the R9 295MX.. However if the R9 3xxM series GPUs are good, I might consider upgrading... especially if they have HBM memory

http://www.sfgate.com/visionaryofth...e-Year-nominee-Lisa-Su-CEO-of-AMD-6070002.php

 

There are two rumors at work here:
1) 20nm wafer shortage
2) GloFo producing this years cards

First, the wafer shortage is b/c TSMC did not get them locked down before Apple and Samsung snapped up most of them currently in the market place. As AMD planned to have the GPU on 20nm originally, produced by TSMC, they were in a pickle. This is when the second rumor developed.

Second, GloFo, the partner of AMD for CPUs, has offered to pick up the slack. They have 20nm wafers for the production of APUs on the 20nm form factor. But that is the only thing that is guaranteed. They have space and time to produce the GPUs on the tried and true 28nm form factor. This fits why the switch potentially to 28nm HP for their GPUs over their traditional manufacturer for GPUs.

What isn't clear is whether no GPUs will be produced by TSMC on the 20nm form factor. A couple have speculated both AMD and NVIDIA will wait for 16nm. Others have speculated that the push back to April or May timeframe on the release of the GPUs signals that the 20nm wafers for the high end will be available, you need fewer of them for the highest end cards, and so TSMC may be able to provide that limited supply while GloFo picks up the slack producing the rest of the line on 28nm form factor. I am hoping that this last speculation is the correct one and that they produce the high-end mobile on 20nm just to piss on NVIDIA while they have the community pissed off.

Further, if you think about it, Prema said that NVIDIA purposefully had a 20% lower TDP in the vbios that he raised in his mod. This vbios limited TDP was removed with driver 347.xx. To have a limiter like this in the vbios removed by a driver, NVIDIA would have had to planned and written both to do this. The cards hit the market for the 900m around October. This means that they had to have this plan in place by or before the summer of 2014. They planned to give it to their mobile customers in the keister for the past 6 months, then tell customers - look at the performance boost that you got by going to the new driver. Nevermind that we removed all overclock ability, we gave you a higher TDP and more performance. Another of my speculations.

 

I'll get on it soon. Got a few things to do before I do so then I will work on it. I'm gonna download Valley as well. Thinking about doing a pre-OC and after OC of Valley, Heaven, and 3D Mark each. I was trying to find a comfortable OC that won't cause any issues as well. I've read some Benchmark scores where done without care to the gpu, as in even if there were artifacts it was left going. For me, as soon as I see anything abnormal I cancel the benchmark and start again with a slightly lower OC. Not that anyone here is guilty of doing so, but what's the point in Benchmarking if it's not at least stable? I'll get to work on very really soon as I am anxious to actually try this out!

Had a thought, should I also show my OC on my 770M as well? None of my OC's will have modified vBios, so it's stock max core clock will be +135 on core. But isn't that the point we are trying to make too? We shouldn't HAVE to run a modified vBios just to get an OC?

 

Holy crap!

From:

PeterS@Nvidia:

"As you know, we are constantly tuning and optimizing the performance of your GeForce PC.

We obsess over every possible optimization so that you can enjoy a perfectly stable machine that balances game, thermal, power, and acoustic performance.

Still, many of you enjoy pushing the system even further with overclocking.

Our recent driver update disabled overclocking on some GTX notebooks. We heard from many of you that you would like this feature enabled again. So, we will again be enabling overclocking in our upcoming driver release next month for those affected notebooks.

If you are eager to regain this capability right away, you can also revert back to 344.75."

https://forums.geforce.com/default/.../gtx-900m-overclocking-with-347-09-347-25/17/

Success Maybe by dragonwolf8504 posted Feb 19, 2015 at 11:33 AM

 

you mean they listened to us?????? get in!!!

 

Get out of here!!! XD

Sent from my Nexus 6 using Tapatalk

 

Yep, that's straight from Nvidia's forums. I about crapped my pants when I read it. Posted in on Overclock.net as well. Feel free to spread the word. Looks like we won!

He also edited in a link for the last driver that would allow overclocking, that part isn't reflected in my original post.

 

I am crying of joy right now :') day made! Best thing ever.

 

Hold on tight to your pants guys, let's wait until we get first hand confirmation when they release the next new driver.

Remember, this is the same PeterS that promised to help desktop 970 owners with refunds then did a 180 a few days later.

 

Before we jump for joy, this is the same PeterS that said drivers'd "help" the 970's memory issue...

Blast it got ninja'd by n=1

 

...well now what do I do?

...I'm all pumped up for a fight.


...and the war ended.


...frustrated, but happy.