Stuttering reported with Win10 Nvidia GeForce GTX 10xx cards

Topic

New Reply

I bumped into a report on Reddit that blames game stutters when Standby memory eats up all available memory. https://youtu.be/_lJcv0V4qlY?t=1m56s Conn
[See the full post at: Stuttering reported with Win10 Nvidia GeForce GTX 10xx cards]

Reply | Quote

Replies

I’ve also seen reports that claim the 39x series of drivers have stuttering issues on Win 10 while the 38x series do not. I’m on Win 7 and a flight simmer (not a gamer), so I can’t confirm one way or the other (GTX 1070 OC). Oddly, the 39x drivers are the ones with the nVidia Spectre/Meltdown mitigations.

If you do have issues, I suggest you visit the nVidia driver forum (1st pinned thread):

https://forums.geforce.com/default/board/33/

That thread is monitored by Manual Guzman (nVidia Corp). There’s also instructions on how to file a bug report if needed.

On another note, be aware that as of April, nVidia has dropped support for:

1) 32-bit OS, and
2) Fermi series GeForce GPUs

– Carl –

2 users thanked author for this post.

Great Lake Bunyip, woody

Reply | Quote

I can confirm there is no stuttering in Win 7 while gaming with my 2x GTX 1070s.

 

Rock

1 user thanked author for this post.

FakeNinja

Reply | Quote

I decided to check out Nvidia’s forum and other Nvidia web pages about the issue and what might be causing it. I went to Nvidia’s Product Security page:

https://www.nvidia.com/en-us/product-security/

I then opened bulletin numbers 4610 and 4611. Here are links for both bulletins.

Security Bulletin 4610, dated 01/09/2018 and updated 01/16/2018:

Security Bulletin: NVIDIA GeForce Experience (GFE) Security Updates for CPU Speculative Side Channel Vulnerabilities

The above bulletin addresses GFE code in an effort to prevent the GFE code being used via the computer’s CPU in a Spectre attack against the CPU. This resulted in the release of GFE software version 3.12.0.84.

Security Bulletin 4611, dated 01/04/2018 and updated on 01/16/2018:

Security Bulletin: NVIDIA Driver Security Updates for CPU Speculative Side Channel Vulnerabilities

The above bulletin mentions that the GeForce drivers themselves were updated so that the graphics drivers themselves could not be used to facilitate Spectre attacks against the computer’s CPU. This resulted in the release of GeForce driver version 390.65.

And finally, the latest Nvidia driver version 391.35 fixes several newly discovered security vulnerabilities. See Security Bulletin 4649 dated 03/28/2018:

Security Bulletin: NVIDIA GPU Display Driver Security Updates for Multiple Vulnerabilities

After driver 390.65 was released and before 391.35 was released, several other driver versions were released:

390.77

391.01

391.24

Page 2 for the 391.35 release thread contains a potential solution, since it appears that 391.35 may be the issue:

Official 391.35 Game Ready WHQL Display Driver Feedback Thread (Released 3/27/18)

From what I see on the above page 2, it would appear that rolling back to driver version 391.24 might resolve this issue.

Also see the post at the bottom of page 3 after you visit the above link.

 

3 users thanked author for this post.

jelson, Great Lake Bunyip, woody

Reply | Quote

well that problem has been around for about a year now;since the creator update and has merited a thread over 200 pages long on the nvidia forum but its also seems to be affecting a small number of people with amd too.Of course it fell on deaf ears until they supposedly fixed it after an uproar for the FCU but as you can see the bug is still there for a number of people.

2 users thanked author for this post.

Great Lake Bunyip, woody

Reply | Quote

Since this is here, I guess I’ll ask: when do you guys think it’s best to apply Nvidia GPU security updates? I already avoid upgrades unless they are security related.

Has it been long enough that bugs in the current version under Windows 7 64-bit have fallen out?

1 user thanked author for this post.

Great Lake Bunyip

Reply | Quote

From what I’ve observed, the majority of people don’t have the stuttering issue on Win 10, but those that do can’t seem to resolve it using current drivers. Also, more complaints seem to emanate from laptop users as opposed to desktops (throttling, heat issues?).

Those that don’t have problems never allow Win 10 to update nVidia drivers. They download drivers directly from nVidia, use the “clean install” option, and only install Physx, driver, and audio (if needed) – no to GeForce Experience. This is purely anecdotal however.

In the flight sim world (Lockheed Martin), it’s rare to see people complain about driver induced stutters on Win 7. Those on Win 7 are mostly like me – 4790k, 16-32gb RAM, GTX 1070, 1080, or 1080ti. But, I’m CPU bound (90+% utilization on all 4 cores), not GPU bound (even though I’m using 7.8gb of video RAM).

I do feel for those having Win 10, nVidia issues. It must be frustrating having an expensive video card and being plagued by stutters. I’m definitely not looking forward to the time when I’m forced to downgrade my 8 machines to Win 10 ;).

– Carl –

Reply | Quote

Unless you have a specific reason for not updating to the most recent version of nVidia drivers (391.35), I’d update ASAP. See this security bulletin issued by nVidia on March 28th:

http://nvidia.custhelp.com/app/answers/detail/a_id/4649

If that doesn’t scare you, it should. These issues were addressed in the latest version as stated in the release notes on page 16:

win10-win8-win7-desktop-release-notes.pdf

No problem on any of my Win 7 Pro/Ultimate machines with these drivers.

– Carl –

Reply | Quote

Have just installed 391.35, in place of previous 382.05. Pleased to report a significant improvement in crispness of screen refreshes.

Group A, Windows 7 Pro, SP1, x64, Intel I-7 “sandy bridge”

Reply | Quote

This got resolved (unofficially) by running EmptyStandbyList every 5 min. through a scheduled task. I have around a .bat file to do it automatically if someone needs it.

Reply | Quote

santino wrote:

This got resolved (unofficially) by running EmptyStandbyList every 5 min. through a scheduled task. I have around a .bat file to do it automatically if someone needs it.

Standby memory usage is a Windows performance feature that has been around since Vista, I think, and while it may help to mitigate this issue in particular, it also mitigates the benefits.  I suppose it’s another sign of the times when the “premier” gaming OS fails to the point that such draconian workarounds are needed.

Windows, from Vista on, isn’t designed to free up as much memory as possible.  Empty memory serves no purpose… it just sits there, doing nothing of benefit to anyone, waiting to be used.  Rather than return deallocated or released memory directly to the pool, why not leave whatever is in that memory there for a bit, with whatever is there (.dll, .exe, etc.) indexed for quick reuse, but with that memory also marked as available?  If a program requests more RAM, it takes no more time to find the memory block with the lowest priority cached standby data and allocate that than it would to allocate “free” memory; in a sense, the memory is both in use and free at the same time, depending on whatever the next task asked of it will be (in a manner that I am sure would make Schroedinger’s cat meow with happiness, if it is in fact alive).

A lot of people seem to put a lot of effort into monitoring their memory usage. I keep seeing people discussing how much RAM the new Firefox uses, or how much Chrome uses, on and on, and basing their decision on which one to use upon that memory usage.  Unless I’m noticing a slowdown from hitting my pagefile, I personally don’t care what’s going on with the RAM.  My tongue-in-cheek suggestion to people who prefer (whatever program’s) reduced memory usage but who would prefer to use something else is to close the Task Manager, then use whatever program they prefer.  Now they can’t see which has more RAM allocated; problem solved!

Now, of course, if you are hammering the page file, then RAM usage begins to matter in a real way, but that’s a separate problem than just worrying about how much RAM is reported to be in use by any given program “just because.”  When I had “only” 8 GB of RAM in my desktop PC a couple of years ago, before all of the optimization that Mozilla has done with Firefox, I did have noticeable slowdowns after a while of heavy duty browsing (tons of tabs), and when I used RamMap64 from SysInternals to have a look at the priority pages and the amount of “Repurposed” RAM, I could see that my system was hitting priority levels (into the top third) for repurposing that it really shouldn’t if there is enough physical RAM for the task(s) at hand.  As a result, the system was punishing my page file and the SSD upon which it resides (though even then I would have had another 11 years of rated life on the SSD at that rate).

I added another 8GB and the problem was fixed… of course, with Waterfox (whose current versions are based on FF 56) being far more efficient with memory usage than the versions of a year or two ago, I may have been able to get away with the 8GB I had before.

I digress (often), but the point is that the standby memory management scheme of Windows is there as a performance feature, and if you have to periodically clear it to avoid other issues like stuttering in games, then it is not working as intended.  Standby memory should be available to use just as much as if it were unallocated, and if people are seeing a difference, then it clearly isn’t working as intended.  Hopefully this will be fixed soon.

Dell XPS 13/9310, i5-1135G7/16GB, KDE Neon 6.2
XPG Xenia 15, i7-9750H/32GB & GTX1660ti, Kubuntu 24.04
Acer Swift Go 14, i5-1335U/16GB, Kubuntu 24.04 (and Win 11)

4 users thanked author for this post.

Elly, AlexEiffel, santino, Noel Carboni

Reply | Quote

What most folks don’t realize is that the last time the way computers used RAM was easy to perceive was in the early 1980s.

A virtual memory system like Windows is anything but simple, ESPECIALLY the current versions, and I suspect many of the “cause” and “effect” relationships gamers are perceiving are very loose at best.

Most folks don’t realize just allocating and filling memory takes a significant amount of time in a virtual memory system, and that things like the Standby list are there to help reduce that.

-Noel

Reply | Quote

Now I’m just wondering: won’t EmptyStandbyList cause problems with the behaviour of SuperFetch (if enabled)? Won’t both fight for the memory: one emptying and the other caching programs?

Reply | Quote

anonymous wrote:

Snip >>> … <<< Those that don’t have problems never allow Win 10 to update nVidia drivers. They download drivers directly from nVidia, use the “clean install” option, and only install Physx, driver, and audio (if needed) – no to GeForce Experience. This is purely anecdotal however.

In the flight sim world (Lockheed Martin), it’s rare to see people complain about driver induced stutters on Win 7. Those on Win 7 are mostly like me – 4790k, 16-32gb RAM, GTX 1070, 1080, or 1080ti. But, I’m CPU bound (90+% utilization on all 4 cores), not GPU bound (even though I’m using 7.8gb of video RAM).

I do feel for those having Win 10, nVidia issues. It must be frustrating having an expensive video card and being plagued by stutters. I’m definitely not looking forward to the time when I’m forced to downgrade my 8 machines to Win 10 ;).

– Carl –

Actually the first part above has been true ever since they added all the G-Force experience, Shield controllers, etc., even on Win7Pro-64_SP1. I was hesitating to comment as am not on a Pascal card or Win10, but I only install the video driver and if updated, the newer version of Physx ever since I found a similar experience with nVidia driver packages. G-Force Experience has been a drag on the system and is avoided.

I found that after driver version 375.63, on my 660Ti 3GB card, I would get severe lag and stuttering, especially on OpenGL games. Interestingly that was the last driver before the nVidia Telemetry add-in. The 375.63 driver was 288 MB (302,305,872 bytes), it stands out as a stripped down version with a lot of the bloat removed for some reason, when those on both sides are 344MB (373.06) and 376MB (375.70). The current 391.35 is now up to 403 MB (423,052,664 bytes).

That said the security advisory appears serious, so you takes you chances. I guess it will be “play or pray” and hopefully not “play and be prey.”

I will try the newest and see what happens, but with an older card and no new games, I would prefer not to update.

Reply | Quote

400 Megs on drivers… I remember entire operating systems being smaller than just a single device driver.

Just wandering, please clarify me, what such bloat can be put on device drivers nowadays that make them so humongous?

Reply | Quote

Ah yes, the infamous 375.63 driver. That was the gold fallback driver for quite a long time. I finally deleted my backup of the driver about a month ago to reclaim space on my SSD.

When nVidia added telemetry, I was quite angry at first (I’m a Group B tin-foil guy). Fortunately, nVidia came to their senses and made it easier to disable (“Help” -> “Allow Experience Improvement Program” – in Control Panel). As a precaution, I also disable “NVIDIA Telemetry Container” under Services.

@Noel – Agree with you completely. Stutters can be caused by hardware, software, drivers, OS, over-clocking, bad power supplies, heat, BIOS settings, Win power plans, sync, etc. or any combination thereof. Many gaming folks are too quick to claim cause and effect in my opinion.

Being the old fart, big iron guy that I am, I tend to look at cooling, stability and component balancing under sustained, heavy load before pointing fingers. I pull my hair out whenever I see comments such as “I ran Prime for two hours, therefor it’s the driver and not my system”. Come to think of it, that may be the reason for my hair loss….

– Carl –

1 user thanked author for this post.

Bill C.

Reply | Quote

Simply uninstall telemetry

“C:\WINDOWS\SysWOW64\RunDll32.EXE” “C:\Program Files\NVIDIA Corporation\Installer2\InstallerCore\NVI2.DLL”,UninstallPackage NvTelemetry

See HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows\CurrentVersion\Uninstall\ for details.

Reply | Quote

This nonsense is slowing down everything and causing quite a few problems since the experts at Microsoft messed with memory management in Windows 10. SuperFetch, for example, loads each and everything and the kitchen sink into ‘standby’ memory on each OS start — even if nothing of it is ever used. On the other hand, apps frequently used get kicked from standby memory after a short period of time and need to be reloaded from drive. It’s just ridiculous, but the smart folks at Microsoft with their AI-driven telemetry don’t notice, or just ignore — even if machines crash. As usual, the experts at Microsoft can’t fix what they broke.

Reply | Quote

I’m not a very tech-savvy guy and a Google search still left me confused. What precisely is this “Standby memory”?

1 user thanked author for this post.

Marco

Reply | Quote

I attempted to describe what standby memory is in post 184946, but I guess I still made it too technical.  Stand by while I try again (wink).

Standby memory is just regular RAM in your system that Windows is using to cache data that may be used later.  It’s not exactly free memory in the sense that free memory has been completely released (or hasn’t been used yet at all) and is just waiting to be used, serving no other purpose but to sit there and wait until Windows needs it.  Even though it is not free memory, standby memory is still available to be allocated (given) to some process (program, generally) as soon as Windows asks, just as free memory is.

When Windows is first booted, there is no standby memory yet.  Some memory is in use by Windows itself, and some is in use by the things that Windows starts at boot time (the various programs that have notification icons in your system tray, for example).  The rest of the memory is free, at that point.

When you run a program, the Windows memory manager assigns it some memory from the free memory pool to load itself into and to store its data within.  It all happens automatically, but there’s a constant give and take between every program and the memory manager… a program needs more memory, so the memory manager finds and assigns that memory, or perhaps you close a bunch of tabs and the program releases that memory back to Windows.

When that happens, or when the program is closed, the memory manager reclaims the formerly allocated (used) memory, but it doesn’t just add it to the pool of free memory and forget anything was ever there.  It keeps the things that were already in that memory loaded in case they are needed again, but it also marks that memory as available.

In addition, Windows looks at how you’re using the system and tries to predict what data you will need next, and it will attempt to preload those things into standby memory while the PC has some free time so that if those data are requested by a program later, they will already be there in memory, saving considerable time.  This is prefetching, or Superfetch as Windows calls it.

As you can see, this strategy seems to be designed to fill the memory to capacity with unnecessary data, which seems counterintuitive.  As it turns out, that is in fact what it is trying to do.  Caching data to RAM instead of having to load it from the relatively slow disk each time results in a faster system, and if done correctly, it has no downside.

When another program is run, or one that is already running wants more memory, Windows will first try to find what it needs from the free memory pool.  If it finds enough memory there, it just assigns it, and its work (for the nanosecond, heh) is done.  It’s when it does not find enough free memory that things get interesting!

As soon as Windows memory manager requires more memory than is available in the free memory pool, it simply finds some standby memory with low-importance data cached in it and kind of tosses whatever was in it out the window, like dumping the contents of a drink cup out of a car window while hanging onto the cup… voila! Instant free memory for whatever process wanted more RAM.  It takes no more time to do that than to assign free memory, and allows the system to hang on to data that may or may not be useful until then.

So that’s standby memory in a nutshell.  I’ve taken some slight liberties with the details to make it simpler, but this is how it basically works.

In this particular case, regarding the stuttering in Windows 10 with nVidia cards, it seems that when all of the free memory is gone (which doesn’t mean the system is out of memory, just that all of the free memory has been converted to standby memory), games begin to stutter.  It should not be like this; there should not be any noticeable effect of having all of the free memory converted to standby memory.  That premise is the entire basis of of the whole caching strategy of Windows.  If there is a penalty associated with the standby system, something’s not working.

I can’t say whose fault this is.  At first glance, it looks like Microsoft’s responsibility, since they wrote the memory manager in the first place, and that the stutter only happens on Windows 10, not earlier versions of Windows.  Without more info, it’s impossible to say, though: it is also possible that Windows 10 changed the way that it handles memory management in a necessary/important/good way, but that the nVidia driver fails to accommodate the new method.

You might be led to question why MS seems to be doing so much changing of things like that, and I am right there with you if you do… this kind of thing is happening with alarming frequency.  There have been reports of Windows 10 drivers being broken (so that they are no longer even compatible with Windows 10 in its latest versions) for each of the last two releases, and for a platform that owes much of its success to its backwards compatibility as Windows does, it is a disturbing trend indeed.

Even so, when MS does make a change, the hardware vendors have to adapt.

 

Dell XPS 13/9310, i5-1135G7/16GB, KDE Neon 6.2
XPG Xenia 15, i7-9750H/32GB & GTX1660ti, Kubuntu 24.04
Acer Swift Go 14, i5-1335U/16GB, Kubuntu 24.04 (and Win 11)

2 users thanked author for this post.

Cybertooth, Elly

Reply | Quote

anonymous wrote:

I’m not a very tech-savvy guy and a Google search still left me confused. What precisely is this “Standby memory”?

I have the same question too.

Plus, if i buy a gtx 1070ti is a bad idea now?

Thanks for the answers =)

Reply | Quote

Standby memory is cached memory. In theory, if the Free memory is very low, standby memory is released instantly, but there is always latency involved. You should aim for free memory to be at least few GBs for best interactive experience.

2 users thanked author for this post.

Marco, Cybertooth

Reply | Quote

The reason the driver package is so big is that it contains a whole lot more than a simple display driver. Looking through the install folder, I have the following:

1. Display Driver
2. Nview (desktop manager software)
3. Optimus (power settings)
4. Update checker
5. Crash Analyzer
6. GForce Experience (a sort of hub to activate the other software)
7. Video streaming service (allows streaming video across a network)
8. HD Audio drivers (for the HDMI port)
9. Miracast (allows for streaming audio across a network)
10. MS Visual Studio Runtime (standard Windows library)
11. nodejs (JavaScript–likely needed for UI stuff)
12. 3D Vision (for 3D glasses compatibility)
13. 3D Vision USB version
14. NvBackend (possibly part of GForce Experience)
15. NvCamera (I think it’s for webcams, with Instagram-like filters)
16. NvConainer (Seems to be necessary to display dialogs)
17. NVidiaInstallCore (the actual installer, I think)
18. Telemetry
19. VirtualAudioDriver (probably allows settings for audio driver)
20. VirtualHostDriver (Unclear. Google says it could be many different things. )
21. Windows Manager Interface (for administration over a network)
22. PhysX (the physics simulator, used in games)
23. ShadowPlay (allows recording screen output with minimal lag)
24. Shield Wireless Controller driver (i.e. a game controller)
25. Update Core (helps with updating old drivers, keeps old settings)

Installed the latest driver with no problems. Previous driver had problems after a reboot and had to be reinstalled, so I checked that.

Reply | Quote

anonymous wrote:

Ah yes, the infamous 375.63 driver. That was the gold fallback driver for quite a long time. I finally deleted my backup of the driver about a month ago to reclaim space on my SSD.

When nVidia added telemetry, I was quite angry at first (I’m a Group B tin-foil guy). Fortunately, nVidia came to their senses and made it easier to disable (“Help” -> “Allow Experience Improvement Program” – in Control Panel). As a precaution, I also disable “NVIDIA Telemetry Container” under Services.

@Noel – Agree with you completely. Stutters can be caused by hardware, software, drivers, OS, over-clocking, bad power supplies, heat, BIOS settings, Win power plans, sync, etc. or any combination thereof. Many gaming folks are too quick to claim cause and effect in my opinion.

Being the old fart, big iron guy that I am, I tend to look at cooling, stability and component balancing under sustained, heavy load before pointing fingers. I pull my hair out whenever I see comments such as “I ran Prime for two hours, therefor it’s the driver and not my system”. Come to think of it, that may be the reason for my hair loss….

– Carl –

I just installed the 397.31 driver (now 64-bit only), and disabled telemetry in Services and the nVidia CP. This is a great driver, no stutter, no tearing and better color than my previous favorite 375.63.

Still bloated, but I only activate the driver and PhysX.

Reply | Quote