New NVIDIA Geforce WHQL Driver 353.06

Thanks for this golden information! I have always hated Optimus because I am always wondering hmm, is this app, video player, browser, etc, running of my crappy Intel HD Graphics or my dedicated GPU? If I buy a performance gaming laptop, I sure as heck want everything to run at max performance (ie the dedicated GPU)

sorry zb0t

 

You have another benefit. Desktop capture is now available for Shadowplay. They're trying to work on a solution for this for Optimus users, but since the iGPU is displaying the desktop, the only way to do it would likely be (my guess, at least) to force the dGPU to number crunch even the desktop display, but then that kills the battery life benefits of Optimus.

 

Thanks for this info Ultima!

 

Recording and windowed/desktop capture is a non-issue with Optimus since you can simply use QuickSync or NVENC in Afterburner. Stuff like downsampling/DSR, refresh rate overclocking, and G-Sync are the real bummers.

 

Yes, all the issues with compatibility of new technologies has led me away from laptops

 

Unless you have a 4K screen - then only two downsides

I really would like G-Sync though. NVIDIA should partner with Intel to support Optimus based systems.

 

That offers them no benefit though... they'll be giving away their proprietary tech for GPUs that can't handle them just so people can have Gsync and battery life. I may be an Optimus hater, but I still cannot logically see why so many compromises should be made for Optimus. Improving the tech is one thing, sure, but sacrifices still need to be made. I'm starting to ask them to let us turn on/off SLI without restarting our PCs cuz desktops (and desktop modified drivers) can do it.

But those don't get the shadowtime ability XD.

 

Afterburner does. It's called prerecord and it's better than shadow time because you have complete control over the duration and can put the rolling buffer into RAM instead of disk.

 

Can't wait to try out the rog swift tomorrow

 

1) So....? Consumer wins.
2) Wait, what compromises?

 

No:

G-Sync
DSR and downsampling
VR
3D
120Hz
Refresh rate overclocking

 

Ah but what I meant was what compromises must be made specifically in regards to G-Sync to get it functioning on Intel GPUs? Can't see how how it would impact non-Optimus machines and it would benefit millions of NVIDIA consumers. I say NVIDIA + Intel partnership to enable G-Sync would be a wonderful thing.

 

You can still set the duration for shadowtime and the location of the buffer, but I can't set it into RAM you're right. That being said, MSI AB's pre-record thing isn't driver level so it requires some CPU power (however minimal). Also the capture type differs; some games REALLY hate being captured in certain ways, and shadowplay doesn't seem to trigger those ways (likely because it doesn't hook to the games).

Yeah, consumer wins, but they don't make money off it. In fact they lose money more than anything else. It's not like people are buying AMD GPUs for mobile when they want performance, and all these thin laptops cannot even use AMD's cards, so it's not like they're gonna sell more.
Compromises like... well everything Octiceps just said above. I don't see why nVidia and Intel should fight so hard just to get these working for everyone when they're not going to make any money off of it.

 

Doesn't sound optimistic: http://anandtech.com/show/9303/nvidia-launches-mobile-gsync

 

Haven't used ShadowPlay in ages. Can you specify exact time limit (in s) and file size limit (in MB) for the prerecord buffer like you can in AB?

Funny, all the stories I've read of games not being recordable (like TW3 at release) involved ShadowPlay not AB.

 

You can specify minutes from 1 to 20 for shadowtime, and it's always active. The MB needed is shown below the minutes, and you can custom-set how much bitrate, resolution and FPS you want shadowplay to use (and that affects the pre-record buffer). No way through the program itself to set down to the seconds.

TW3 was a rare case, yeah. I heard people complain about it. But I've never seen it be a problem in other games, really. Except for Binding of Isaac Rebirth; it wouldn't grab that game (I assume due to the kind of OpenGL he coded it in). I will admit it could do with better audio settings and such though. But games like Dark Souls 2 don't like being grabbed and recorded all that much, and other games like BF4 etc show some issues with conventional grabbing at higher FPSes for certain programs' capture methods. If you read up on the nVidia driver tweaking thing that was linked some time ago, it shows that Shadowplay only grabs the entire framebuffer and doesn't actually hook to a game, which is why it can't be used for windowed mode. But desktop recording for borderless windowed games would be pretty sweet, I won't lie.

 

OK I see. ShadowPlay hasn't changed then. AB lets you set time and file size limits for the prerecord buffer independently, whichever it hits first. No 20min/1min upper/lower limit and such.

And yeah, ShadowPlay's audio quality and settings are a joke. I know people who are forced to use AB for that reason alone. AB is just a much better recording utility for power users. The last time I used ShadowPlay, AB blew it away in terms of customization and functionality. I don't know how much ShadowPlay has improved since then but it doesn't sound like much if at all.

 

You think with all the coding experience they have at nvidia they would be able to make a decent recording program for their own hardware.

 

You would be surprised.

I don't doubt that at all. I mean, I could do it with Playclaw using NVENC and pre-record buffers too etc. I just mean that for absolute minimal performance hit and lack of needing to set a pre-record starting, shadowplay be it. I hate its barebonesness though. I'm waiting for Playclaw 5.5 to come out. Looks like they're really getting into it, which is good since it hasn't updated in about a year.

 

"Coding experience"

Recent Nvidia drivers

Does not compute

 

Just put my 780ti's up for sale on eBay, this will fund my second Titan X.

 

Cool story bro

 

Really didn't want to part with them

 

Please continue

 

Well... I'm giving up on the M18xR2 working right with these drivers. I did make progress. Modded the 353.06 desktop driver and used strings for GTX 980 and the throttling is much improved, but it's still there and still hurts performance. I am able to overclock the snot out of a single 780M now, so it is definitely something they added to the drivers to restrict power and stay within some sort of arbitrary limit. I am guessing that code was in the BIOS all along and only recently did they add something to their drivers to stay within an arbitrary threshold. Sucks. If they do not fix their drivers and knock of this foolish nonsense soon will be selling all of my Alienware laptops. It's probably safe to assume this is what is happening with 980M SLI upgrades as well. They have incorporated filth for 980M "power compliance" that is ruining my Kepler experience.

[parsehtml] <blockquote class="twitter-tweet" lang="en"><p lang="en" dir="ltr"><a href="https://twitter.com/ManuelGuzman">@ManuelGuzman</a> <a href="https://twitter.com/nvidia">@nvidia</a> <a href="https://twitter.com/NVIDIAGeForce">@NVIDIAGeForce</a> Here is a brief 3 minute video demonstrating the problem Please fix it. <a href="https://t.co/mhDRD2qBl3">https://t.co/mhDRD2qBl3</a></p>&mdash; Mr. Fox ([USER=165514]@Mr_Fox_Rox[/USER]) <a href="https://twitter.com/Mr_Fox_Rox/status/605557782494904320">June 2, 2015</a></blockquote> <script async src="//platform.twitter.com/widgets.js" charset="utf-8"></script>[/parsehtml][parsehtml]<iframe width="1280" height="720" src="https://www.youtube.com/embed/zkqOfN0AQ7k?rel=0" frameborder="0" allowfullscreen></iframe>[/parsehtml]

 

Has anyone experienced BSOD with the new drivers?mine 336 or 346mhz overclock was fine for project cars in 350.12 but now it crashes to desktop or BSOD(happen 2-3 times),They are making it less stable to for us upgrading.I read today that amd fiji will start at 599$.I wonder how it will compare to 980 or 980ti.Windows 10 is made for AMD.Nvidia will a lot of customers cause of the new windows and their behaviour

 

It better perform at least as well as a 980ti or AMD are dead in the water

 

Are you on an Alienware and/or have a Killer wireless card?
I tried a +158/500 Overclock today and it wasn't really an issue but I didn't exactly leave it for long

 

No no.I got acer v3 772g which is a old system haha but it lasted long enough until nvidia screwed it up but next year i buy a desktop which is more suited for gaming

OFF TOPIC:Has anyone ever heard for this game.Looks really cool and comes today out? http://store.steampowered.com/app/346110/

 

Here's what I get for stock 780M SLI with 345.20... side-by-side with yours. http://www.3dmark.com/compare/3dm11/9884524/3dm11/9884990

 

1. Did you at least know that Graphics score is what means for estimating GPU's performanse? Now look again why you got lower scores with 353.06. I would retest it again.

Well, I feel cool that 10 minute video sacrifice just 1MB for audio and it sounds OK. But yeah, more options is better.

 

Well these drivers brang back the strange image doubling blur phenomenon on occasions on my desktop 980 but also have experienced on 980m . Not to mention, one of the dirtiest installations ever.. Black screen after install with a strange blinking white icon on the bottom left, reboot, black screen, reboot and ok.
No thanks Nvidia... No thanks.
Geez, going onto 1 year of rubbish drivers.

 

AW 18: 880M SLI, 4910MQ CPU Oc'ed to 4.3GHz, 32GB Corsair Vengeance RAM, 1TB Samsung 850 Evo m-Sata SSD... On my 4.3GHz overclock profile it crashes Project Cars. Was fine before the latest driver. But going down to the 4.2GHz profile flawless gameplay, VERY nice increase in FPS with this new driver. Average 10-15FPS on various tracks.

 

This was really not much to brag about. The difference between 353.06 vs 352.86 WHQL drivers is minimal... 0.2% increase in graphics scores with 3DMark Fire Strike is a joke. LoL. This is all Nvidia could boast of... http://www.3dmark.com/compare/fs/4977946/fs/4934209

 

Nvidia promised game performance not benchmark performance

 

Sure... But it had been a little more fun with a larger increase than 0.2% in benchmarks. But 0.2% increase is better than lower scores. LoL.

 

Ehh... 0.2 is well within the margin of error... These drivers suck again.

 

I got BOTH massive game performance increase and benchmark as well!!

3DMark Run with previous driver

3DMark Run with the latest driver

 

Kurwa!

Do you mean GTX 850/860/960M, i.e. GM107 users don't get those tessellation improvements too, and are stuck in the same boat as Kepler users?

And despite all that, I finally managed to get 45-50 FPS in TW3 at medium-high settings with this driver - also dragged down the resolution slider to 1600x900. The smoothness is more than worth the loss in visual fidelity.

And yet again, GTA V performance is best with 350.12. I cannot achieve frame-rates higher than 40 FPS with the last two drivers when I used to easily reach 60 FPS at high-very high settings with 350.12.

Nvidia, get your **** together.

 

@SRSR333



Too tired/bothered to compare with Kepler GPUs in-class, but I'm sure octiceps will be more than happy to help you out

 

The whitepaper says only Maxwell 2 has PolyMorph 3.0 but Maxwell 1 still has tessellation optimizations over Kepler.

Okay... What settings are they using besides x64 though? Because when I run it on my 980M, the results are quite BAD. 253 FPS with a +100 core and memory overclock... This is also an OpenGL benchmark... its not going to be relevant for DX11 tessellation performance.

EDIT: Nevermind... "The latest version of TessMark is designed to focus on tessellation via OpenGL 4. We run the latest version of the benchmark using the high resolution map set at maximum tessellation while 1080p full screen, reporting the average FPS."

Power limit throttle brings the core down to around 1100



Okay... Anand did something different... According to http://www.anandtech.com/show/7457/the-radeon-r9-290x-review/17 I should be getting over 2k FPS on my 780 Ti, however...

 

Maxwell 2 is PolyMorph Engine 3.0:

Spoiler: Maxwell 2 SMM diagram

Maxwell 1 is PolyMorph Engine 2.0 like Kepler:

Spoiler: Maxwell 1 SMM diagram

Spoiler: Kepler SMX diagram

However, based on these TessMark results, it seems Maxwell 1 has improved tessellation performance over Kepler like Maxwell 2 does.

980 (GM204): 2321 FPS
770 (GK104): 1625 FPS

750 Ti (GM107): 796 FPS
650 (GK107): 412 FPS

Notice how 980 scores about 3x as high as 750 Ti, which makes sense as the GM204 GPU (16 SMMs) is about 3x as big as GM107 (5 SMMs).

Interesting to note, Maxwell 2 is nowhere near 3x as fast as Kepler in tessellation despite what Nvidia claims in its GTX 980 whitepaper.

 

^ I think the fairest comparison of Kepler vs Maxwell would be 980M vs GTX 770, as they have the same amount of cores. That'd show the direct Maxwell improvements. Of course you'd likely have to clock the 770's memory down to match, but still.

 

The core count is not directly comparable because they're different architectures. The "a 128 CUDA core Maxwell SMM has 90% the performance of a 192 CUDA core Kepler SMX" explanation that Nvidia gave (which would mathematically translate to each Maxwell core being 40% stronger than a Kepler core). So it's not fair to compare a neutered GM204 vs. a fully enabled GK104 just to match CUDA core counts.

There is one PolyMorph Engine per SMM/SMX. GM204 has double the tessellation engines of GK104 because GM204 has 16 SMMs while GK104 has 8 SMXs. That's where GM204's tessellation advantage comes from.

Similarly, GM107 has 5 SMMs and tessellation engines while GK107 has 2 SMXs and tessellation engines.

 

Welp. That's a GG I guess. Now if only Maxwell had stable voltage and kept proper clocks in SLI and stuff. That'd be great.

 

I'll be damned. This driver really improved TWD3's performance. I can play it now with these settings, rarely dropping below 30 FPS in town and stable 40 FPS in outside with GTX 870M. 3DMark 11 score is almost at same level (though there is a little drop, but it happens all the time). This performance should've been there in the first place but I'm glad that NVIDIA cleaned their mess. But I must say nevertheless, the graphics don't look so good even on ULTRA. I mean it is good but it is not as anticipated. It is like something is unnatural.

 

I take it you haven´t read the last couple of pages.. Nice it´s working for you though.

 

Anyone knows why the latest 353.06 WHQL nVidia drivers are available for cards such as GTX 860M, 850M and so on and not for the GTX 960M? Looks very strange to me...

 

Indeed, I get improved performance in TW3. Your GPU is more powerful than mine - you should consider moving up a few of your settings (especially water and texture quality). Barely any FPS impact, and the game looks a lot batter. Also, I notice you don't have AO switched on. Use SSAO - it makes all the rough corners stand out a little less.

And finally, you might consider sacrificing 1080p resolution for 900p, to get a boost of up to 20 FPS - yes, 20.

Yet, Nvidia nuked performance in everything else. GTA V is still a mess (-25 FPS on average). Damn it Nvidia, I want to play all games on a single driver, not switch between drivers for different games. This isn't the 1990s.

 

TessMark is useless because it is OpenGL 4 tessellation performance, not DirectX 11. A comparison of Heaven results and measuring the performance hit the respective cards take with the extreme tessellation option would be a much better example.