Geforce 280.28 OpenGL 4.2 Driver

I now switched measuring to Power Consumption im not sure now if not calling NVAPI alone is causing trouble in the Drivers Monitoring behavior, especially the long P0 states. Though the issue with Firefox is crazy and can't be alone Nvidias fault http://demos.hacks.mozilla.org/openweb/HWACCEL/ this should push more pressure on layers then smooth scrolling does though it behaves ok with 280.28.

Power measuring shows it switches very fast the P-states and holds Performance.

http://ie.microsoft.com/testdrive/Performance/FlyingImages/Default.html

Works also perfectly fine switches very fast to higher P-state when needing it and fast down if it doesn't (like it should be)

After some more tests im gonna do a Bug Report with Mozilla about this 280.xx driver changes and Layers behavior @ smooth scrolling)

 

Some information to cope with the Firefox Browser situation without losing the Driver and Firefox

http://forums.nvidia.com/index.php?showtopic=196885&view=findpost&p=1281587

There is also the way of using the Multi Display Saver by Orbmuk but it's not flexible enough to guarantee optimal Performance in different Browsing Scenarios.

 

Does this driver have the "run hot" bug that ppl report that the 280.26 driver has ?

 

that's not a fix but a lame workaround, sorry.

And yes its nv fault, fact it was fine before with older driver.

Again now that they changed this gpu P-state monitoring it is what it is, you can't argue with that., yes its the best way not to blame them and blame everything else like OS, browser and God knows what...Open your eyes and stop being so stubborn in favor of nvidia. And sorry your logic seems flawed at times..

And dont think nv is perfect, far from it.

-----
edit: don't get me wrong i have nothing against nvidia, im just unsatisfied customer who bought an expensive product and now the driver causes all kinds of issues. And I knew from the past that nv can be sloppy at driver dep., but i never expected such lame "fixes/hacks/whatever" that are happening now for quite a while, again..

Or what's worse, these famous driver stopped responding warnings..

phew at least i didnt have this problem when i used 280.xx., im still at 280.26 in winxp though.. with yea that nv inspector workaround, but when i play Fear Combat i have to disable it, because it didnt felt so smooth with it in the background as its monitoring your gpu all the time (also eats 25mb of ram)..c1:


So yea now they're gonna release this promised beta driver with P-state fix in early September :infinity: instead of now in August like it was first mentioned from the messenger ManuelG.

 

I partly agree but Software has to adapt also to so Massive changes from time to time.

Latest Nightly of Firefox improved the situation considerably (Smooth scrolling)



And the better response time of the 280.xx branch is overall a very good thing, if everyone fully adapts to it

I don't like it if Nvidia hacks every fix in via the Driver as this is uncontrollable behavior most of the times, better is to adapt to such changes accordingly via your Application directly.

 

ok yea i agree partially one thing needs to adopt to another, but still Full3D clock just for browsing is to much anyway you look at it, i mean seriously 30gpixel and 50gtexel powah just for browsing, heh nooo way.

 

For WebGL you could benefiet from very high clocks depending on the Application but yeah most stuff shouldn't be but see on my chart Firefoxs Layers Engine needs 50% GPU though can still render it ok @ 50 Mhz but how should the Driver know that if it's just taking a GPU load threshol into account, even the Graph doesn't tell fully what Nvidia uses exactly to define a P-state switch their must be more factors then just GPU load

As you can see GPU load is most of the Time High but it doesn't switch and when it's suddenly low it switches so that seems not logic for us if we only take GPU load into account

Though you can't Debug such things with Afterburner it's to limited same as the whole NVAPI for consumer

 

yea other factors like sloppy driver coding.


Anyway nv better bring this fixed beta already, this is getting ridiculous.

 

Ahh wait i think i was this who forced this behavior via a Nvida Profile change for Firefox let me see

 

Vsync + tripple buffering <- not sure though if this works under NT 6

 

btw dont think you will get any reasonable answers there over at nvidia forums.. Except from some clueless long time trolling members.

and about monitoring 270.xx spikes to full3d clock when i open Flash player(youtube video) but the gpu usage is at only 3%, compared to some spikes up to 55% at low3d power or low 2d..so yea this whole monitoring is one big messed up joke.


and i dont see any point in discussing it anymore, nv driver team has already been working on this "fix" for sometime now, what almost 15-20days.. And this 280.28 openGL 4.2 is still from the old branch without this 100% P-state fix.


Like i said if they went back to 275 with Quadro then you know what's the deal.. They did it in the past if a driver "sucked"..


Look


and before that a couple of times with 19x.xx branch.


edit: forcing triple buffer +vsync didnt do anything here, it goes to full3d clock anyway.

 

Not sure though im currently trying 275.89 but it doesn't behave much different then 280 with Firefox and Smooth scrolling all of them go straigth to full 3d clocks.
And i didn't mean thatit doesn't go to full 3d clocks anymore but it wont switch as fast anymore like it uses to without when just faster scrolling on the page with smooth scrolling and layers.

The major propblem just seems to be that without tripple buffering and vsync Firefox Layers is exploding (might be only Vsync related) on the GPU causing a immediate full 3d clock switch that's also why you see almost 0 GPU load if you enabled vsync you can see the GPU load and it switches in a much more sane way then without (seems logic it's rendering like a benchmark and it seems that makes not much sense for smooth scrolling after all so vsync should be enabled for Firefox in the Profiles and Tripple buffering for keeping the Performance high) .

Without Vsync enabled you wont see the Graph above with the GPU load just the jump into Full 3D clock (not sure but maybe Nvidias sensors get overloaded by the load of Firefox Layers it causes when smooth scrolling)

 

yes all 270.xx and 275.27, 275.33, 275.65, 275.89 go to full 3d for a brief second or two, 3.. but they revert back (keeping gpu cooler) and dont jump so much like with any 280.xx , like i said in the first page these 280.28 are much better then any 280.xx atm but still its not it like it was before.


Noq Im really curious what are they gonna do about it, I hope its ok again.

 

Hmm also 275.89 stays very long in P8 for example even with 0 load occurring for a long timeframe when not using vsync + triple buffering in Firefox with smooth scrolling.


Though i don't used Afterburner for making sure it's sane i use the Watt meter which has a overload function which i set on a given wattage if it goes over that a red light indicates it and overload is shown in the display.
So if i smooth scroll and the red light goes on i know its in P0 state and if i stop the mouse i can count the time it takes for the red light to go of (sure there is latency involved here but still you can count the time) and with vsync and tripple buffering reaction time so back into the older state is very fast with 280 see afterburner graph above im trying currently the same with 275 gonna report the difference and the same graph for it above.

So what im doing to optimize is very easy i count how many times the red light goes on (overload) and how long it stays on until i stooped smooth scrolling
I can definitely say without V-sync its overloading a lot of times (which seems logic as it isn't limited to a Framerate anymore and so Frames explode and you don't really see anything more smooth without V-sync the contrary is the case it stucks on scrolling (depends on scroll speed). Everything seems so logic it's the same you would expect from a Direct3D Game on XP
And as Layers is based on Direct3D the behavior without V-sync makes absolute sense (lower power efficiency)

Though in Afterburner you would think something is totally strange as you see 0 GPU load just the clock switch happening to full 3d, that was looking indeed very strange @ first to me.

 

P8 is fine, that doesn't heat up much compared to 280.xx's fast switching between P8 and P0 with a much longer P0.


Must be Winxp specific and your gpu specific then.

 

As you can see in the excellent optimization with 280 and Firefox the highest switch time is roughly 5 seconds from P0->P8 and look at how often that is triggered (under the v-synced load) definitely with 275.89 i have times like 14 seconds before it switches from P0->P8 and it gets much more often triggered so the higher response time is really a good thing but the Application scenario needs to be optimized (in Firefox Layers case you need V-sync or it blasts any Power save for nothing the best with Triple Buffering to keep Performance) you can really save something here

 

This driver heating up cards business certainly looks huge right now.

Anyone actually lost a card, noticed an increase in electricity bill, or forum discussions is where the most of energy went?

 

if you don't ave anything constructive to say stay out of this thread, we don't need your trolling attempts here..

There is nothing excellent about multiple P8 to P0 switching - its this what heats up the gpu in the first place,

If you brought it down to "old monitoring" by enabling vsync it could be winxp and your gpu specific.

Win7 is different i see no frame tearing when watching flash videos unlike by winxp.
And in winxp it didnt help if i forced vsync i still saw frame tearing, so that what you said about vsync doesnt really work imo, or vsync would be kicking in and i wouldn't saw any frame tearing.

But i leave you with your attempts of fixing it, gl lol


I'll wait for a fixed beta.

 

Hmm this definitely helps for XP (with Firefox Layers) btw there is also a Adobe Flash Player profile.



Though this amazed me also it seems incredible and it not only looks that good the benchmark is also on screen much faster then with 275.xx



Though their is 1 thing i really find odd with the Driver nad P-state managing but also 275.xx shows this behavior many applications cause a P0 state just by starting up Gpu-Z for example or Firefox with Layers It looks strange but in the case of 275.xx it takes also over 15 seconds before it goes down after Application init that can't be right

It doesn't happen with every Application Afterburner isn't affected for example but GPU-z definitely is Also Adrenalines Crysis 2 Benchmark just opening his GUI triggers it (im pretty sure it's a Nvidia API initialization thing, this was reported a long time ago could be maybe a regression)

Initialization bug ???



Same for Nvidia Inspector though not so heavy



Unwinder is still the Hero here with his low level (Kernel) coding Sniff Sniff Rivatuner

 

^
afterburner spikes to full3d too when i start it, but in only in win7 (any driver so far), winxp is fine and stays at low2d (be it 275.33 or 280.26).