Dual GPU AMD/Nvidia

2»

Comments

  • RayDAnt said:

    I'd have to go through my records by 4 that I recall. People who bought budget Radeon cards for gaming and cheap quadros to do AI coding.

    Is it safe to assume that these Quadros were all being set to TCC mode? Because if so, that would explain why you would have never come across any of the brand intercompatability issues being discussed in this thread (since GPUs in TCC mode don't engage with the WDDM subsystem in the first place.) It's also worth reiterating that @lazarus102 is working with a 6+ year old Radeon R9 290 specifically. Meaning that whatever worked/didn't work hardware-compatibility-wise between AMD and Nvidia GPUs circa 2013 is primarily what's relevant here.

    I didn't put them in TCC. The 290 is using the Radeon driver that every other reasonably recent Radeon card does. The annual update just came out.

    Windows doesn't talk directly to any HW except through drivers. So as far as Windows is concerned the 290 could just as easily be a brand new 5700.

  • RayDAntRayDAnt Posts: 1,147
    edited November 2019
    RayDAnt said:

    I'd have to go through my records by 4 that I recall. People who bought budget Radeon cards for gaming and cheap quadros to do AI coding.

    Is it safe to assume that these Quadros were all being set to TCC mode? Because if so, that would explain why you would have never come across any of the brand intercompatability issues being discussed in this thread (since GPUs in TCC mode don't engage with the WDDM subsystem in the first place.) It's also worth reiterating that @lazarus102 is working with a 6+ year old Radeon R9 290 specifically. Meaning that whatever worked/didn't work hardware-compatibility-wise between AMD and Nvidia GPUs circa 2013 is primarily what's relevant here.

    I didn't put them in TCC.

    Why on earth would you install headless Quadro cards in systems for compute purposes and not put them in TCC mode?

     

    Windows doesn't talk directly to any HW except through drivers. So as far as Windows is concerned the 290 could just as easily be a brand new 5700.

    Updated software doesn't change underlying hardware. You can write updated code to tell Windows that an old card is a brand new one, and Windows will happily go along treating it as such up until Windows tells it do do something it simply can't, and you get unresolvable errors. Non-obsoleteable hardware is a contradiction in terms.

    Post edited by RayDAnt on
  • RayDAnt said:
    RayDAnt said:

    I'd have to go through my records by 4 that I recall. People who bought budget Radeon cards for gaming and cheap quadros to do AI coding.

    Is it safe to assume that these Quadros were all being set to TCC mode? Because if so, that would explain why you would have never come across any of the brand intercompatability issues being discussed in this thread (since GPUs in TCC mode don't engage with the WDDM subsystem in the first place.) It's also worth reiterating that @lazarus102 is working with a 6+ year old Radeon R9 290 specifically. Meaning that whatever worked/didn't work hardware-compatibility-wise between AMD and Nvidia GPUs circa 2013 is primarily what's relevant here.

    I didn't put them in TCC.

    Why on earth would you install headless Quadro cards in systems for compute purposes and not put them in TCC mode?

     

    Windows doesn't talk directly to any HW except through drivers. So as far as Windows is concerned the 290 could just as easily be a brand new 5700.

    Updated software doesn't change underlying hardware. You can write updated code to tell Windows that an old card is a brand new one, and Windows will happily go along treating it as such up until Windows tells it do do something it simply can't, and you get unresolvable errors. Non-obsoleteable hardware is a contradiction in terms.

    I don't sell people monitors. I build systems to spec. If someone wants a gaming card and a quadro that's what they get. If they put the Quadro in TCC that is their choice. I certainly don't need the headaches of trying to explain over the phone or in email how to put a card back into WDDM if they decide to connect a monitor to it.

    Drivers just do not work that way. When an application that uses a GPU, or any other hardware, is initializing correct procedure is to query the device to make sure it supports the features you need. So if a card only supports OpenGL 3.2 and not 4+ then the application knows and reports the issue. Every driver call is also supposed to fail with an error code not just try to get the hardware to do something it can't when the call is for something the HW doesn't support. There should never be a crash directly caused by a driver. Sloppy code that doesn't trap errors can crash but that isn't the drivers fault.

    That's the whole reason device drivers exist.

  • nonesuch00nonesuch00 Posts: 18,302

    So as I thought and stated. the problem is because iRay Preview is being used in the viewport and that the nVidia GPU / DS weren't programmed to handle that with an AMD card present. If p0rt's suggestion isn't the correct and only way to handle it besides replacing the AMD card with an nVidia card then there is unlikely to be another solution.  

  • PadonePadone Posts: 3,790

    @nonesuch00 it works just fine for me.

  • RayDAntRayDAnt Posts: 1,147
    edited November 2019

     

    I don't sell people monitors. I build systems to spec. If someone wants a gaming card and a quadro that's what they get. If they put the Quadro in TCC that is their choice. I certainly don't need the headaches of trying to explain over the phone or in email how to put a card back into WDDM if they decide to connect a monitor to it.

    So you have experience building these sorts of systems, but not necessarily operating them. That's a fair distinction to make.

    Speaking as someone who daily-drives that kind of a system (beefy TCC-capable Nvidia GPU for compute purposes, dramatically cheaper other-brand alternative graphics device for display output/other purposes) one of  the very first things you do, if you want to make a return on your investment, is activate TCC mode since it instantly eliminates a whole boatload of inter-GPU compatibility issues (see eg: this thread) and increases both VRAM use efficiency and GPU compute performance with the Nvidia card in all tasks.The reason why you have never apparently come across any of these issues in your work is because you've never had to deal with people using the specific hardware (non-Quadro/Titan GPUs) where these issues matter.

     

    Drivers just do not work that way. When an application that uses a GPU, or any other hardware, is initializing correct procedure is to query the device to make sure it supports the features you need. So if a card only supports OpenGL 3.2 and not 4+ then the application knows and reports the issue. Every driver call is also supposed to fail with an error code not just try to get the hardware to do something it can't when the call is for something the HW doesn't support. There should never be a crash directly caused by a driver. Sloppy code that doesn't trap errors can crash but that isn't the drivers fault.

    Notice all of the "shoulds" in this statement. In theory, all operating system/driver code and embedded hardware platforms work perfectly. The reality is that they don't always. If they did, this thread wouldn't exist.

     

     

    So as I thought and stated. the problem is because iRay Preview is being used in the viewport and that the nVidia GPU / DS weren't programmed to handle that with an AMD card present.

    Daz Studio doesn't have the capablity to interfere with which GPUs the Iray plugin can/can't access successfully for rendering (apart from the control directly afforded the user as those checkboxes under "Photoreal Devices".) Eg. Iray has absolutely no problem accessing my Nvidia GPU for rendering (whether for "liveview" or final "Render" purposes) regardless of whether Daz Studio is programmed to use my Nvidia GPU or my Intel iGPU for its own operations.

    It is far more likely, based on your initial description of your problems, that your Nvidia card is currently failing to get used by Iray for rendering under any circumstances (including with the AMD completely deactivated) due to underlying driver issues with Windows. The way to tell definitevely is by looking at the end of the DS log file for the lines labeled "IRAY" after attempting to activate "liveview"/"Render" both with and without the AMD card being used for Daz Studio operation and seeing what it says. Hence why I have been asking you to post that information.

    Daz Studio using your AMD card for processing will not in itself prevent Iray from using your Nvidia card for liveview/final rendering. Using a secondary card (Nvidia branded or not) for display output/Daz Studio operations is actually a recommended use case in DS's own log file.

    Post edited by RayDAnt on
  • Every response I can make could be interpreted by mods as a personal attack. I strongly suggest people investigate previously claims about TCC and DS before believing anything said here.

Sign In or Register to comment.