Nvidia Has a Driver Overhead Problem

Discussion in 'Videocards - NVIDIA GeForce Drivers Section' started by RealNC, Mar 15, 2021.

  1. RealNC

    RealNC Ancient Guru

    Messages:
    4,893
    Likes Received:
    3,168
    GPU:
    RTX 4070 Ti Super
    It seems the original thread was posted in the wrong forum section and thus closed? So reposting here:



    Apparently something very similar was discovered some years ago and NVidia released a driver with less CPU overhead.
     
    akbaar, yobooh, cloakengaged and 9 others like this.
  2. Chrysalis

    Chrysalis Master Guru

    Messages:
    373
    Likes Received:
    90
    GPU:
    RTX 3080 FE
    I will repeat my reply then.

    to me this is not proven to be a generic driver overhead issue as from what I can tell he only tested DX12 games, we already know AMD has a DX12 edge.

    The reviewers need to get into the habit of testing a mixture of games from dx9/dx11/dx12/vulkan all 4 api's.
     
  3. PQED

    PQED Active Member

    Messages:
    92
    Likes Received:
    32
    GPU:
    MSI 1070 Gaming X
    They stickied a comment regarding this on the very video in question, as well as leaving a comment on this LTT video:

    I too was confounded regarding DX11 titles, they could definitely have been clearer on that point. However, I found no trouble comprehending that he was talking about DX12 as well as Vulkan in the video itself.
     
    Smough, BlindBison and Dragam1337 like this.
  4. RealNC

    RealNC Ancient Guru

    Messages:
    4,893
    Likes Received:
    3,168
    GPU:
    RTX 4070 Ti Super
    That's over an hour long :p Better link to the specific timestamp using code tags.
     

  5. PQED

    PQED Active Member

    Messages:
    92
    Likes Received:
    32
    GPU:
    MSI 1070 Gaming X
    It's a comment below the video.;) (one by Hardware Unboxed, that is)

    If you want to hear what Linus and Luke has to say about it there are timestamps down in the comments for that as well.
     
  6. Chrysalis

    Chrysalis Master Guru

    Messages:
    373
    Likes Received:
    90
    GPU:
    RTX 3080 FE
    I will quote his sticky comment (the relevant part).

    So it is as I suspected. Misleading title in both video title and this thread.

    PQED I suggest you quote the linus comments, its unreasonable to expect people to search for it.
     
    Last edited: Mar 15, 2021
    Stormyandcold likes this.
  7. PQED

    PQED Active Member

    Messages:
    92
    Likes Received:
    32
    GPU:
    MSI 1070 Gaming X
    I will agree that the video is somewhat misleading, but the title of this thread makes no claim as to what API is affected; only that there's and overhead problem, which there appears to be.

    RealNC didn't appear to know either, and I can't blame him considering that the relevant info to clear it up can only be found in the comments of the two videos linked. You should never need to search outside the relevant media for pertinent information.
    This one's on Hardware Unboxed.
     
    Smough likes this.
  8. Chrysalis

    Chrysalis Master Guru

    Messages:
    373
    Likes Received:
    90
    GPU:
    RTX 3080 FE
    Well given the vast majority of games are not DX12, on that basis I consider it misleading. :)
     
  9. PQED

    PQED Active Member

    Messages:
    92
    Likes Received:
    32
    GPU:
    MSI 1070 Gaming X
    Linus's comment is in the video, which is timestamped in the comment right at the top of the video comments (34:19), but that's not the one I was referring to as I noted above.


    This is Hardware Unboxeds' comment on LTTs' video I was referring to (it has been edited, and who knows if it will be again):

    "Just a bit of extra info, this mostly applies to DX12 and Vulkan games when CPU utilization is maxed out, though you can also see it with DX11 titles as well, again when the CPU is maxed out. The comments from Linus on how to test GPUs (low-end or high-end) are spot on in my opinion. There is really no need to change how graphics cards are tested and as Luke and Linus said, this could be special feature type content, but again the way reviews are done doesn't need to change."
     
    Chrysalis likes this.
  10. PQED

    PQED Active Member

    Messages:
    92
    Likes Received:
    32
    GPU:
    MSI 1070 Gaming X
    Quite harsh choice of words considering it wasn't intentional, but you are of course entitled to your opinion.
     

  11. Chrysalis

    Chrysalis Master Guru

    Messages:
    373
    Likes Received:
    90
    GPU:
    RTX 3080 FE
    I never said it was intentional, just that it is misleading, which it is, I am sorry if you think its harsh. :/, and yes its just my opinion.

    However it is good he has made the sticky post and he is doing a follow up video.
     
    PQED likes this.
  12. PQED

    PQED Active Member

    Messages:
    92
    Likes Received:
    32
    GPU:
    MSI 1070 Gaming X
    I didn't mean to imply that you said it was intentional, I was just saying that in the context of what RealNC knew when he created this thread I felt it a bit harsh. The title is correct after all, just lacking a bit of information which we've already corrected for.

    But we've probably gone off-topic enough with this already. No hard feelings on my part, so no need to apologize. :)
     
  13. The Goose

    The Goose Ancient Guru

    Messages:
    3,057
    Likes Received:
    375
    GPU:
    MSI Rtx3080 SuprimX
    So basically...... Nvidia need to remove all the background running crap from there drivers.
     
  14. TheDeeGee

    TheDeeGee Ancient Guru

    Messages:
    9,624
    Likes Received:
    3,409
    GPU:
    NVIDIA RTX 4070 Ti
    Those arn't even using 1% of the CPU.
     
    The Goose likes this.
  15. Stormyandcold

    Stormyandcold Ancient Guru

    Messages:
    5,872
    Likes Received:
    446
    GPU:
    RTX3080ti Founders
    Needs way more Intel cpus to see the whole picture.

    Also, I'm sure I've seen videos around GTX900 series era that did similar comparisons and it showed that it was Nvidia who were better than AMD if you had a lower-end cpu.
     

  16. Nastya

    Nastya Member Guru

    Messages:
    185
    Likes Received:
    86
    GPU:
    GB 4090 Gaming OC
    Latest video on the topic:
     
    akbaar, Smough, Xtreme512 and 4 others like this.
  17. ScreamerRSA

    ScreamerRSA Member

    Messages:
    44
    Likes Received:
    19
    GPU:
    ASUS 3090 TUF OC
    Does this not explain Ampere's worse performance (relative to AMD 6800 series) at 1080/1440p even on high end CPUs. Usually running those resolutions will result in high framerates, and increased CPU usage. If the driver is using up a lot of the available CPU cycles, needed to get high framerates, surely that will affect it?
     
  18. Chrysalis

    Chrysalis Master Guru

    Messages:
    373
    Likes Received:
    90
    GPU:
    RTX 3080 FE
    Possibly if its DX12 games.
     
  19. ivanosky

    ivanosky Active Member

    Messages:
    95
    Likes Received:
    18
    GPU:
    NVIDIA RTX 3060
    Digital Foundry has mentioned AMD's driver issues with lower end CPUs for years on a lot of their videos and articles: Search • Eurogamer.net
     
    Stormyandcold likes this.
  20. BlindBison

    BlindBison Ancient Guru

    Messages:
    2,404
    Likes Received:
    1,128
    GPU:
    RTX 3070
    I am curious whether or not games like Crysis 3 which are DX11, but also scale with core/thread count much better than most DX11 games would still exhibit this issue. The Nerdtech video that Hardware Unboxed linked in the comments that went over the differences between AMD's hardware scheduler/Nvidia's software scheduler makes it sound like this behavior would probably occur in any games that saturate all cores and threads whereas Nvidia's approach would yield better results if a game only used a few cores/threads.

    Video for context:

    Since going forward I expect DX12/Vulkan to become more standard and especially since modern consoles use Zen2/RDNA2 and all that jazz, my bet would be that these HU findings will become increasingly more relevant unless Nvidia reworks their driver/scheduler, but I'm not knowledgeable enough to understand what that would entail. 18-30% difference when CPU bound is quite a lot.

    It is of course possible that I am misunderstanding certain aspects of this behavior, but particularly for 1% lows which can manifest as microstutters, I think this is an important topic. To my recollection, Digital Foundry was still experiencing some cpu bound microstuttering in their 3900X and 2080S videos from awhile back at 1440p so perhaps more relevant than some are giving credit seems to me.
     
    Last edited: Mar 15, 2021
    tfam26, CPC_RedDawn, itpro and 2 others like this.

Share This Page