none
Windows 10 does not let CUDA applications to use all VRAM on (especially secondary) graphics cards. RRS feed

  • Question

  • I'm a 3D Graphics artist. I'm using nvidia graphics cards for 3D rendering using CUDA computing.

    Problem takes place after upgrading to Windows 10 (from win7).  We did not have this issue with windows 7.

    The application (octane render) reports an amount of unavailable GPU memory that can not be accessed.

    This is a very well know problem people in my industry.

    This problem occurs using in every nvidia graphics card. Not just display cards. But also all GPU's that are not connected to any display, just acting as dedicated CUDA computing (render) cards are suffering from this issue.

    The video cards are not connected as SLI configuration. They are working as separate cards but rendering software 

    we use gets all the GPU's work on the single rendered image..

    if you have a gpu card with 4GB then you end up having 700MB VRAM unavailable 

    if you have an 8 GB (1070) card  then you end up having ~1.3GB VRAM unavailable 

    if you have a 24 GB Quaddro or Tesla GPU you end up having ~4.5 GB VRAM unavailable 

    And this limits our workflow.

    I doubt if it just effects nvidia cards. I have no knowledge of other manifacturers like AMD having the issue.

    GPU oriented computations is mainly done by using nvidia cards. That is by CUDA computing.

    I also found an article about Process residency budget regarding the changes in WDDM 2.0

    new Windows 10 driver architecture. 

    https://msdn.microsoft.com/en-us/library/windows/hardware/dn932169(v=vs.85).aspx

    In our (me and people who are doing GPU rendering) workflows. We generally leave the display card(s) just for displays. And the other secondary (up to 12 or more ) gpu's are dedicated to rendering purposes. I can not understand the dynamics behind the keeping resident video memory That are not helping with the display.

    I hope you can come up with a solution or even a workaround to help us out.

    I need Windows developer team reply to this issue that we're having with Win 10.

    Would be great if we could regain the unavailable memory of our GPU's for our renders.

    Already posted at...

    https://answers.microsoft.com/en-us/windows/forum/windows_10-hardware/windows-10-does-not-let-cuda-applications-to-use/cffb3fcd-5a21-46cf-8123-aa53bb8bafd6?tm=1476693874292&auth=1

    My system is

    i7 cpu + 32GB ram + nvidia Titan 6GB + nvidia GTX 1070 8GB + nvidia GTX 1070 8GB running on Win10 Pro

    Tuesday, October 18, 2016 10:54 AM

All replies

  • Hi oguzbir,

    Since the issue occurred after we upgraded to Windows 10. There may be a compatibility issue. Please ensure the machine is up to date and you have got the latest application.
    According to my research, the latest CUDA version should be 8.0.

    "For Direct3D 12 applications, the budget is handled completely by the application. The size of the budget is meant as a cue to let the application know what to size itself to. By using the budget size as a hint, the application can decide how many resources to keep resident, what resolution and quality of resources to keep. "
    For the "Process residency budget", it should be more related to the application as the link pointed out. So we`d better to ask the application developer support for help.

    Best regards


    Please remember to mark the replies as an answers if they help and unmark them if they provide no help.
    If you have feedback for TechNet Subscriber Support, contact tnmff@microsoft.com

    Wednesday, October 19, 2016 2:03 AM
    Moderator
  • Hi MeipoXu,

    Thank you for your contribution.. 
    FWIW I do have the latest application, latest GPU drivers, latest windows 10 updates and Cuda 8 for sure.. But this is completely unrealated to the issue we're having. This does not just happen to me and my single computer. 

    This is a wide range problem. Has to be solved either by application developer or Windows Technical developers Which I do not know yet the solution to.

    So, I'm so sorry but I do not understand what you're saying here.

    The rendering software I among others using having this problem. Including our software developers. 
    Say it's more than thousand userbase with all ranging hardware software are using this software. 
    And the ones making the move from Win7 to Win10 is affected..

    The program (octane render) developers which they see this problem as it can't be solved on application programming side. They are already in talks with nvidia. And as they call it. It's just because the Microsoft Windows 10 Display Driver Model (WDDM) v2..
    I as a user for this software looking for ways to understand the problem. 
    If a Windows Tech developer could describe this problem to us. Then we'll see what we can do.

    If a windows developer says that it's the first time he/she hears this and investigate it. That's fine.
    Or maybe he can tell the CUDA application has to do some programming for this. Then I will ask the our developers to work harder and solve the issue at hand.

    If the Windows developer says this is know issue. And wont be fixed near future.
    Then I and other user probably will make the move back to Windows 7..

    There is no need to address the part of Direct3D 12 section from the Process residency budget article or what ever causing the problem. Because as far as we see it CUDA enabled applications are suffering this problem globally. Direct3D 12 is a completely different thing.

    Also not just spesific render engine like Octane render has this issue.
    VRAY, ARNOLD, FSTORM, IRAY are also render engines that uses CUDA computing and has this problem as well. So you can add upto a ~100K userbase. Some are unaware of the problem because haven't made the move to Win10 yet.

    I imagine there must be other industries where they use CUDA computing softwares.(like Agriculture or Petrol searching, Architecture mabe the police )

    Since the problem we're having is the unavilable memory during 3D Render. It might not be much of a problem where they use Cuda computing only. 
    But 3D CGI Rendering engines use the most of the hardware including the GPU computing power and all of it's memory when needed in bigger scenes.

    So It would be really very kind of you if you can help us elevate this issue to a technical person who can resolve or give advices for this matter.

    Cheers,

     
    Wednesday, October 19, 2016 2:25 PM
  • Hi oguzbir,

    I will try to escalate the issue on my side. The whole process may take a long time, please be patient and take an eye on this thread.

    Best regards


    Please remember to mark the replies as an answers if they help and unmark them if they provide no help.
    If you have feedback for TechNet Subscriber Support, contact tnmff@microsoft.com

    Monday, October 24, 2016 7:00 AM
    Moderator
  • Hi MeipoXu,

    Thank you for this.

    Is there any other way of contacting Windows 10 techincal developers directly 

    so we can speed this up?

    Since new Windows Display Driver Model is causing us trouble here. It would be great to get in touch as much as directly it can be.

    Or is this the only way (forum) we can reach the developers?

    Monday, October 24, 2016 4:21 PM
  • Update / Edit:

    Added additional information how the described issue affects system setups that use 1 GPU for the display and additional GPUs for rendering tasks.

    - - -

    I have also observed that since the upgrade to Windows 10 a large amount of VRAM is not available for usage in GPU rendering software.

    To illustrate a practical example:

    compare screenshot*

    - System setup:

    3x GTX Titan

    1 GPU is assigned to the display

    2 GPU are assigned to the render engine

    - VRAM on all three cards: 6144MB

    - VRAM available in rendering application: 5052 MB
    - VRAM used by 3d scene: 146.1 MB

    Applications open in windows 10:
    OctaneRender 3.04.1
    GPU-Z 1.12.0

    -> Why is only 5052 MB available?

    The scene is only 146.1MB large and there is
    6144 MB VRAM on the graphics card.

    Assumption:

    -> part of the VRAM is reserved by Windows 10 and cannot be used by other applications
    -> the larger the VRAM on the card the larger the amount that is reserved by Windows 10

    -> It seems Windows 10 is making no difference between the VRAM of the GPU who is assigned to the display and the additional  GPUs who are used by GPU rendering applications.

    - - -

    *Unfortunately I cannot seem to post an image or link until my account is verified.

    Hopefully this workaround link is acceptable:

    daz3d.com/forums/uploads/FileUpload/2c/15ff5f35e7f98d78041331a6c7151b.jpg

    - - -

    Suggested improvements:

    1) If Windows 10 needs to reserve VRAM of the GPU assigned to the display this should not in any way affect the available VRAM of other GPU installed on the system.

    2) Instead of windows 10 using a percentage of maximum available VRAM on the graphic cards it would help if users could indicate a fixed amount of VRAM that is reserved.

    3) In general it might be helpful if in windows 10 diagnostic tools would be included that clearly indicate

    - GPU assigned to the display, GPUs assigned to other applications

    - maximum available VRAM on each of the installed cards

    - amount of VRAM currently reserved by Windows 10

    - amount of VRAM currently used by different applications

    - amount of VRAM actually available for use

    - - -











    • Edited by linvanchene Monday, October 24, 2016 7:07 PM
    Monday, October 24, 2016 6:14 PM
  • While discussing this topic on other forums some questions were raised.

    I post the relevant information also here:

    - - -

    What is the difference between a display GPU and a rendering GPU?

    A video card aka GPU (= Graphical Processing Unit) is used on most computers to handle the display.

    Because of that part of the VRAM of the display GPU may be reserved by applications active on the system.

    For rendering 3d scenes it is important that all VRAM of a GPU is available for use.

    Because of that most users of 3d rendering software have installed 1, 2 or even 3 additional GPU just for rendering.

    This would allow users to render with the render GPUs while the system would still be responsive for other tasks because a separate GPU was assigned to handle the display and all other applications.

    With Windows 8 or Windows 7 such multi GPU setups were working as intended.

    The available VRAM of the display GPU would not affect the available VRAM of the rendering GPU in any way.

    - - -

    - - -

    How do we know that under Windows 8 and 7 rendering GPUs had the full amount of VRAM available to render?

    In the OctaneRender 2 manual a screenshot of the "GPU quick information" is shown

    docs.otoy.com/Standalone_2_0/?page_id=138

    When OctaneRender 2 was released in June 2014 Windows 10 was not yet around.

    So that old screenshot was very likely made on Windows 8 or Windows 7.

    the numbers there read: 122 / 5978 / 6144

    VRAM on card 6144 MB

    Scene size 122 MB

    The available VRAM was still around 5978 MB !

    Now with windows 10 we only have around ~5000 MB of a 6144 MB card available for use.

    - - -

    - - -

    Did you make sure that you just used a rendering GPU and not a display GPU when testing?

    Yes, I made multiple tests with different scenes and different combinations of GPU activated.

    - - -

    Here is another example that features a more detailed look how the VRAM is used:

    Test scene car:

    Please once again compare the screenshot:

    daz3d.com/forums/uploads/FileUpload/b0/c1c1eae0d971e8f8e41538f0da268d.jpg

    - VRAM on the GTX Titan that is a dedicated rendering GPU not set up as display card: 6144MB

    - VRAM available in rendering application: 5077 MB
    - VRAM used by 3d scene: 440.7 MB

    This 440 MB  of the 3d scene include:

    Engine run time data: 140.3 MB

    Geometry: 106 MB

    Image Textures: 182 MB

    etc.

    -> Unavailable Memory 1 GB !

    - - -

    The GPU used for rendering is not a display GPU.

    -> Why would there be any unavailable VRAM at all if the only other applications open are File Explorer and Internet Explorer?

    - - -

    I tested this scene with all 3 GPU individually.

    The two rendering GPU used alone had 1 GB of unavailable VRAM.

    The display GPU had 1.1 GB unavailable memory.

    (Tested while File Explorer and Internet Explorer were open).









    • Edited by linvanchene Tuesday, October 25, 2016 4:57 PM
    Tuesday, October 25, 2016 4:49 PM
  • Hi linvanchene,

    Thanks for this. Could you please let me know which other forum are you discussing this issue?

    I'd like to contribute as much as I can.

    As for the above Q&A.  I agree with all the answers....

    If we can pass this Questioning and quizing phase and have this issue escalate to ranking developer. That would be great. 

    It could also be a problem with nvidia drivers. Maybe in order to keep the Windows WDDM 2.1 intact.

    Nvidia could enable (if possible) all GTX cards as TCC not WDDM. That way geforce cards could be able to ditch this problem as being able to be seen as compute cards as well.

    We'll see.

    Tuesday, October 25, 2016 11:32 PM
  • @ other threads discussing this topic

    One thread was posted on the Otoy forum in December 2015 by official OctaneRender staff:

    render.otoy.com/forum/viewtopic.php?f=12&t=51992#p260054

    This topic became active again in the last weeks after Cuda 8 was released and people started to purchase new graphic cards of the Pascal series.

    - - -

    I created another thread on the DAZ3D forum to gather some information if those using Nvidia Iray may have encountered a similar issue and to motivate people to participate in discussing this issue and raise some awareness.

    daz3d.com/forums/discussion/120541/windows-10-reserving-vram-links-to-posts-at-microsoft-forum/p1

    Unfortunately unlike in OctaneRender the DAZ Studio Iray integration does not have advanced tools to monitor VRAM usage. Most users seem to rely on GPU-Z to get an estimate.

    Some users on the DAZ3D forum report not to be affected by this issue.

    It is unclear if this may be because Iray features a hybrid CPU - GPU mode.

    - When OctaneRender runs out of VRAM the render simply fails.

    - When Iray runs out of VRAM it falls back to the slower CPU mode.

    - - -





    • Edited by linvanchene Thursday, October 27, 2016 9:44 AM
    Thursday, October 27, 2016 9:34 AM
  • Hi

    We need this to be fixed soon. Windows 10 is great but it swallows 2GB vram.

    I have 4xTitan X which has 12GB VRAM PER CARD. 

    I am only getting 8GB. I use Octane Render and after contacting octane they have made it very clear it can only fixed from windows side. Octane has around 10,000 users worldwide. This is one render engine. So were looking at around 100,000 people affected by this issue.

    The sad thing is if I buy the new 1070s then I wont even get 8GB but only 6GB Vram. This is abysmal. 

    Please can this be fixed asap as its been awhile since windows 10 release now.

    Thanks

    Christian

    Thursday, February 2, 2017 9:05 AM
  • I recently moved from windows 7 to windows 10, and have the same issue. Under windows 7 I was able to use 95% of the GPU memory for running Theano in deep learning applications, by setting cnmem=1.0, under Windows 10 I am limited to cnmem=0.8, which is equal to 80% of the available GPU memory (3.2 of 4 GB, Nvidia GTX 960). The display is connected to the on-board Intel graphics, the GPU is used for computations only.

    Win 10 seems to be allocating (and blocking) GPU memory for a display that is not there! Win 7 was smarter.

    It means that all deep learning programmers moving from Win 7 to Win 10 will lose (at least) 15% of their GPU memory capacity.

    Thursday, February 16, 2017 8:06 PM
  • you can change VRAM Settings 2 ways in the settings untick the first 3 boxes that uses the aero stuff then if you have extra hard drive space you can change your VM from system managed to from anything higher then the standard overtime if you track this using control alt delete keys and going to system performance you will see that windows is smart enough to use VM and not all your system resources SYSTEM RAM i split my drive in to partitions and use a whole 20 gigabyte partition for PAGING

    but understand paging is so slow if your running a new SSD you shouldn't even use paging

    but adding another stick of ram and making 16 gigabytes instead 8

     
    Friday, February 17, 2017 7:39 PM
  • I have the same issue with GTX 980 TI on WIN10. I can only allocation maximum 82.3% memory for computation, which means >1GB memory cannot be assessed, however nvidia-msi only show 40MB memory is in use. This card is only for calculation purpose, there is another card for display.

    MSFT team, please help to fix this issue asap. This issue has been widely spread in the past more than one year --- the first discussion on this issue I found was on June'2015 in Nvidia forum, it has been more than one year but it seems there is no progress made.

    Tuesday, February 28, 2017 6:15 AM
  • Microsoft, this is a huge deal, not only affecting users of 3D Rendering software, but also anybody who uses parallel computing in general. There are farms of 100's to 1000's of machines who are now only able to use 80% of their available VRAM. In these use-cases, they need every megabyte of available VRAM to successfully complete our projects.

    This has been a known issue for almost two years now and really, really needs attention.

    Windows 10 is an otherwise very pleasant upgrade, but I may be forced to go back to 8.1 and halt recommendations to my peers to upgrade unless this is addressed.

    Can you please commit to addressing this issue?

    Friday, March 10, 2017 5:53 PM
  • Hi,

    Same problem here, 6gb of video ram and only 4.8gb available.

    this is creating lots of problem, I have upgraded the cards (4 cards, almost 1000$ each) from 3gb (2.5 usable) to 6gb in order to have 3.5gb more of memory, and instead i have gained only 2.3gb, this is so disappointing.

    In a 3d production pipeline 1.2gb are a LOT of memory that we cannot use, now I need to upgrade to 11gb cards, so I assume I will have only 8.8 gb of actual free video ram considering that the 20% of the memory is allocated we don't know where..

    can someone solve this issue?

    Thursday, March 23, 2017 7:54 PM
  • I too have this problem and I 'm not alone. If you head to the Octane render forums (or any gpu render forum) you will see that everybody that is serious about GPU rendering is sticking to Win7. Nobody is going to pay (myself included) top dollar to buy 2-3-4 or even 7 top Nvidia cards (I have 7 ) and then install win10 to lose 30% of their investment in available video memory. GPU is being also used to more and more things today and it's the rendering workhorse of the future. We are not alone in this and it's a pity that a legion of users that would otherwise use win10 is not going to until this is clearly fixed. We are waiting for a clear answer on this for a long time now. 

    Regards,

    KOD


    PiNkFlUiD

    Friday, March 24, 2017 8:20 PM
  • I also use an Nvidia card for rendering and the amount of VRAM reserved for Windows is ridiculous: over one gigabyte on an 8 GB card.
    Friday, April 14, 2017 5:10 PM
  • Please Microsoft, do something about this. I'm going back to windows 7 to render bigger stuff, but it's a pain in the a..
    Sunday, September 3, 2017 3:07 PM
  • Do you have any success to contact Microsoft developers? The issue is really serious. I'm going to uninstall Windows 10.
    Monday, September 11, 2017 8:11 PM
  • Hi,

    I also have observed this with a GTX 1060 (compute GPU) beside a GTX 1080 (render GPU), from the 6GB available on the 1060, 1086MB are always impossible to use.

    This is a big problem as memory is a scarce resource when computing on GPU. Does someone have some information about this problem?

    Can we at least have a confirmation whether this is a problem on Windows's side (Windows taking resource it does not actually require) or on Nvidia's (bug in the driver)?

    Thank you for your attention and, I hope, your answer.

    Friday, October 6, 2017 1:19 AM
  • Has there ever been a response from Microsoft regarding this? Is the issue the same with OpenCL/AMD? Because at least then we could rule out nvidia driver issues.
    Wednesday, November 8, 2017 11:00 AM
  • MeopoXu I don't think you fully understand just how big this issue is.  To clarify GPU rendering is now a HUGE part of the crypto-currency especially with coins like Etherium.   We are talking 10's of thousands 6 card GPU computer rigs.

    Mostly on windows 10 1000's of users have been on a wild goose chase trying to work out why GPU cards aren't seen, picked up, crash, 1 or 2 out of 6 cards returning an error on drivers ... the list is long. The when they finally get windows 10 to mine or render multiple GPU cards (in my case after countless (+30) blue screens of death) they are then left scratching their heads as to why 1 of their GPU cards is showing poor hash rate performance.

    At the moment few get this far (this thread) 

    My issue at the moment is I now have no way now of getting windows to use a generic driver to render windows from RAM because I can't get into only has the default windows driver disappeared from device manager as an option in my device list (probably after windows preferring the new intalled GPUS to default to)  So now i can't even get into my bios to set the initial display output to the generic display driver which I no longer seems to have anyway.  Reason being boot skips bios screen when initial display output is set to a PCI lane occupied by a GPU.  

    I think someone at microsoft needs to sit down and install a mining rig with 6 GPU's from start to finish.  Not kidding.  Because people are kinda giving up on MS in the alt coin mining industry (which is huge).  You guys have really fallen asleep at the wheel on this one!  

    Saturday, November 11, 2017 7:58 PM
  • DaDoodle, since you're a miner, it would be nice if you could say whether you have AMD cards or specifically if it is an AMD card that is missing the GPU RAM.

    I am "only" trying to run tensorflow on windows and my models simply don't fit into whatever windows 10 leaves me as GPU RAM. What the hell. So if I want to stay on windows I have to downgrade to win 7/8 with constant reminders to upgrade to win10 or skip this shit and use linux?! Well played.


    • Edited by MuhMuh Monday, November 20, 2017 2:02 PM
    Tuesday, November 14, 2017 12:10 PM
  • Just ran into this at our production shop, please fix!

    A long series of CAD render projects were done on Windows 7 on multiple clusters of 11GB 1080ti cards where most of the scene sizes were heavily optimized to fit into the roughly 10.3-10.4 GB of available VRAM. 

    We've since upgraded to Windows 10, but the client recently asked to update the scenes. To our surprise the scenes no longer render as we only have ~9GB of usable vram under windows 10. It is not feasible at all to rebuild the assets to lower spec as it will take way too much time, and in addition degrade the overall quality. 

    Turned into a huge problem and we're stuck currently transplanting the render cluster to windows 7 machines in order to render the scenes

    Please fix this, there is no reason VRAM allocation should be percentage based and also be imposed on non display render cluster cards.

    Monday, December 18, 2017 5:30 PM
  • Windows 10 consumes almost 2 GB. Totally insane, especially in spite of the fact that it's the second GPU, which is not used for display. It's windows 10 pro, v 1703, build 15063.786

    
    Wednesday, January 3, 2018 4:27 PM
  • What a thread. I'm building a render box. I'd thought to use W10, until I read this. Maybe I'll just stick with W7 for now.
    Monday, February 5, 2018 3:14 AM
  • I also had the same problem, my CUDA application can only end up using part of video memory. I thought this was intentional since my original system only have 1 graphics card and it's probably because it's reservation for display or something. But now I have 1 GTX 980 for rendering and display and one GTX 1080 8GB dedicated for CUDA workflow, still, I can only use about 6.7 GB of video memory on my 8GB GTX 1080. This is a big deal for us because it limits the availablility dataset for our GPU based algorithm.

    It's been 1 year and a half since the problem was reported and I'm now using the latest Windows 10 with CUDA 9.2 and the latest driver from NVidia but the problem is still there. Because of this and the horrible WDDM overhead we have to rewrite a Linux version for the benchmark. Microsoft should think seriously of those problems which explains why developers and researchers often consider Linux as their best platform for development. 

    Sunday, May 27, 2018 7:27 PM
  • So many time has passed and Microsoft didn't do anything. In meantime, only solution was to reverse to Windows 7 as all this (and often data loss due to forced restart after windows updates) leaves impression Windows 10 isn't production ready. :(
    Saturday, June 16, 2018 12:28 PM
  • This is unbelievable, it's been since 2015 that this problem is known and what has been done?? absolutely nothing...
    How is this possible ?? what are we supposed to do now ? move back to windows 7 and hope Microsoft doesn't completely kill it off soon?

    Monday, July 2, 2018 10:04 PM
  • I am just running into this problem with my GTX 1080. 12Gb on the card, tensorflow reports 8.0Gb total and 6.0Gb free. I just ordered a new ML rig that is being built out. I guess that I will end up installing Ubuntu on it and move the GTX 1080 to it. Really getting sick of all the issues with Windows. Takes forever just to get anything up and running. Had to manually hack the Autokeras code just to get it to run on Windows. Microsoft is really falling behind.
    Thursday, November 8, 2018 6:46 PM
  • I have a GTX 1080 Ti and my Tensorflow models keep going Oom (Out of memory). nvidia-smi tells me that I have more than 3 GB Free. Bought Windows 10 Pro for this machine but am now regretting it. With Microsoft soon stopping support for Windows 7, I don't think I will have any option but to switch to Ubuntu if I would like to continue serious machine learning work.
    Thursday, November 22, 2018 2:56 PM
  • Same issue with the GTX 1080TI.  Getting out of memory warnings and like everyone else, we lose 20% of VRAM from the start.  We should all assume Microsoft has no intention of being a player in any regard in the AI/ML market.
    • Edited by johnpjust Thursday, December 20, 2018 3:37 PM
    Thursday, December 20, 2018 3:37 PM
  • I agree 100%  It's been years and windows 10 still eats 20% of vram. Fix it Microsoft
    Thursday, February 14, 2019 3:00 PM
  • Please vote in Feedback Hub aplication: https://aka.ms/AA4mjmm
    Monday, March 25, 2019 1:03 PM
  • The link https://aka.ms/AA4mjmm doesn't seem to be working.
    Tuesday, March 26, 2019 5:32 AM
  • Well yeah, if you have a GTX 1080, you will have 8GB max. Not 12 GB. Or do you have a GTX 1080 TI? Then it would be possible to have 12 GB.

    As for Microsoft: SHAME ON YOU! Seriously, how is it possible that you still haven't done ANYTHING to fix this issue? 

    Last few years Microsoft has made a lot of decisions which are ANTI-CUSTOMER.

    WINDOWS 10 is not production ready, it's not even ready for personal use.

    A MCITP/MCP

    Friday, April 26, 2019 10:41 AM
  • This link to feedback hub, you need windows 10 and installed application "feedback hub" (installed on windows 10 by default).

    Maybe this will work?
    feedback-hub:?contextid=70&feedbackid=232767ca-71ff-45a1-841b-3dcb87362566&form=1

    Saturday, April 27, 2019 2:20 AM
  • MICROSOFT - GET OFF YOUR ____ AND DO SOMETHING ABOUT THIS PROBLEM
    Saturday, June 29, 2019 9:45 PM
  • I can't believe it has been as long as Windows 10 has been around, and yet they've done absolutely nothing to resolve this. We spend thousands of dollars to buy bigger GPUs for extra VRAM only to lose 1-4.5 GB to windows 10. I think it's time to start looking for other operating systems that may get the job done.

    Microsoft clearly doesn't give one single ....

    Wednesday, December 4, 2019 11:02 AM
  • Yep... Almost three years from the original post, and not a single reply from Microsoft.

    I just gave up waiting for Microsoft to fix the problem. I replaced Windows 10 for a Linux distro called Pop!_OS which allows me to use all my GPU VRAM with my deep neural networks. The extra 1.3 GB in VRAM is almost like day and night difference when working with deep learning applications.

    There is a similar issue in the Feedbak Hub app, but it only has around 6 upvotes. So, I don't think this is a priority to Microsoft.


    • Edited by LangoLango Friday, December 6, 2019 9:49 PM Add more info.
    Friday, December 6, 2019 9:48 PM