assurance-tunnel
assurance-tunnel
assurance-tunnel
assurance-tunnel

PERFORMANCE ISSUES... GTX 1080 TI performs worse than stock Radeon 5770



  • 1. been testing extensively and finding that my new Nvidia GTX 1080 Ti is actually performing slightly worse than the stock radeon 5770 on my mac pro 2012 ... the 1080 Ti should be far superior ... my patch uses a combination of CI and GPU effects but no cpu ones.. so all should be processes on GPU.. I'm running at 640 x 480 and using the hapq codec.. but even when Im using only live capture from 1 usb cam and no video playback the results are the same.. worse performance on the 1080 Ti.. How could I test this card to rule out a hardware issue.. I have run every GPU benchmarking apps I can find: cinebench R20, compubench CL, GFXBench GL, benchmark Valley by unigine, and I'm running the System indicators app that monitors performance on the GPU and it appears to be working.. (i'm using the latest nvidia web driver and cuda driver)..

    2. in general I dont understand bottleneck I'm running into with my patch.. when I monitor CPU, Ram, disk r/w, and GPU.. none of these hardware aspects are being taxed significantly but isadora's load monitor can shoot up 100% when I kick on only a few effects, but ram CPU useage= 10%, ram= 5gigs, disk read 25mb/sec, GPU useage 30%.. it seem that for some reason isaodra is not utilizing the full capabilities of my hardware.

     has anyone done an A/B test with a 1080 Ti or any other high end vid card?

    Any help would be much appreciated.

    MAC PRO 2012 5,1 12 CORE XEON 3.46GHZ, 64G RAM
    EVGA GEFORCE GTX 1080 Ti
    PCIE USB3, PCIE ESATA, PCIE SSD
    MAC OS 10.13.6 (17G10021),  ISADORA 3.0.7



  • @demetri79

    @DusX -- can you help out on this one?

    Best Wishes,
    Mark


  • Tech Staff

    @demetri79

    I can only make a guess since your system is a custom build.
    But, what I expect is that the video card is somehow swapping the video on and off the GPU. Maybe from the CPU, maybe from the GPU shared memory (that tends to be rather slow).
    Something between the macOS and the video card is not optimized well.
    Does nVidia offer a macOS driver?
    I think it could be something in the OpenGL implementation, but again, that's a guess.



  • Thanks for the replys.

    Yes I am using the nvidia driver designed for mac.. and the latest version.. this is  called web driver for some political reason. But it seemed like lots of people were using these 1080's for isadora .. I am I wrong about that and all the isadora users of this card are on PC?

    That card should be massively powerful.. but it's has turned out to be a useless expensive "upgrade" .. I guess I will need to test with some high end AMD cards instead.



  • @demetri79 do you have both cards attached? If so remove the AMD card and leave the NVidia.

    Also can you do some other tests, to test pure GPU performance take an HD h264 video and apply a lot of GLSL effects, this is pure GPU, no translation. I use a 1080ti in a modern windows machine and it is a very powerful card, I have no trouble with it and have run a lot of heavy patches, multiple capture devices, heavy GLSL effects and the machine - particularly the GPU is great. I am guessing either something with your config (the version of web drivers is picky), your patch or your config is up. This card itself and Isadora's ability to use it have no issues in my experience.

    Other parts of the motherboard in these machines are quite slow in comparison to modern boards (8+ years of technological advancement is a lot) so there may be other factors like bus speeds disk speeds and pcie transfer speeds that are limiting your patch and hence you get the same results as you did with the old card.


  • Tech Staff

    @fred said:

    @demetri79 do you have both cards attached? If so remove the AMD card and leave the NVidia.

     Yes, it's important to note that Isadora does not support the use of multiple video cards simultaneously.


    @Fred said:

    Other parts of the motherboard in these machines are quite slow in comparison to modern boards (8+ years of technological advancement is a lot) so there may be other factors like bus speeds disk speeds and pcie transfer speeds that are limiting your patch and hence you get the same results as you did with the old card.

    Yeah, customizing those ancient cheese-grater Mac Pros can be a bit of a glass-cannon situation. Sometimes you'll get great results, sometimes it all sort of just blows up in your face


    @Demetri79 said:

    my patch uses a combination of CI and GPU effects but no cpu ones

    @mark Are the CI effects GPU or CPU?


  • Tech Staff

    @Fred 

    I suspect you are right. The slowdown may well be another bus involved.



  • @Fred @Woland Thanks everyone for chiming in.. this has been a long frustrating process.. but nothing good comes easy:

    so I only have the 1080 connected not both cards..  as I have every pie slot occupied.. 1080, NMVE ssd, esata adaptor, usb3 adapter

    i read somewhere on the site that CI are processed by the GPU.

    i read up on all the components and the 1080 Ti thoughput is not choked by the PCIE bus on the 2012 mac pro.. BUT the results do make me question that though, i will test high end AMD cards to make sure. the nvidia web drivers supposedly dont cause any performance penalties according to users of other identical machines and cards on other video software apps.