No announcement yet.

AMD Announces The Radeon PRO W6000 Series - RDNA2 Workstation GPUs

  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    Originally posted by khnazile View Post

    I'm using mine in an actual SFF case, and I was able to hit critical temperature when it starts thermal throttling. The fan never spins at full speed. And even if you undervolt/downclock it, it would still reach temperatures above 90C because fan would be spinning at 20% of full speed or so. I'm really not happy with default fan setting as it makes whole pc case hot to touch.
    That's indeed interesting. In fact fan spin on full throtle during the boot here. But well, not sure if this is not due to having CSM still on and boot is using old BIOS.
    BTW: using ubuntu 20.04 with whatever drive is intended for the card:

    ii libdrm-amdgpu1:amd64 2.4.102-1ubuntu1~20.04.1 amd64 Userspace interface to amdgpu-specific kernel DRM services -- runtime
    ii xserver-xorg-video-amdgpu 19.1.0-1 amd64 X.Org X server -- AMDGPU display driver

    Are you sure your drive is not buggy?


    • #32
      Originally posted by simburde View Post

      AMD plans to add RDNA support to ROCm this year
      I'm well aware of this and the myriad other similar threads on their tracker. The problem is this line:

      "We can not add details of future plans in our wiki/docs as per the process."

      Well great, see ya then because I'm not going to make purchasing decisions based on vague timelines on a project that has repeatedly failed to deliver on them. If getting more transparency is a problem, that's a you issue and should be worked out internally. No demonstration of forward progress and no community/PR strategy means that it's fair game to assume nothing exists and nothing will be done. FFS, at least put up a github milestone or something so it looks like you care about keeping the community engaged. Open source is about more than throwing code over the wall.


      • #33
        Originally posted by Mark625 View Post
        Which was kind of my point. Another poster found the AMD specs and the FP64 rate is 1.11 Tflops, or 1/16th of the FP32 rate. This FP64 crippling has come to be expected for consumer grade cards, but one would expect better from a PRO card. FP32 is actually fine for my field (image processing), but I think for the prices they are charging we should get non-crippled FP64. If not 1/2 of FP32 rate, at least 1/4th.
        "Crippled" can mean a lot of things in this context but the most common usage is fusing off hardware functionality so that FP64 throughput is less than what the silicon allows. That is not the case with the RDNA workstation parts - the hardware is only designed for 1/16 rate, as is the case with almost all of our cards (both workstation and consumer). This is still generally faster than corresponding NVidia parts.

        We typically include fast FP64 in one chip every generation or two - Cayman (1/4), Tahiti (1/4), Hawaii (1/2) and Vega20 (1/2) so far. The Hawaii and Vega20 consumer SKUs did have fuse-reduced FP64 throughput - 1/16 for Hawaii and 1/4 for Radeon VII vs 1/2 for workstation - but workstation SKUs always run FP64 as quickly as the underlying silicon allows.
        Last edited by bridgman; 09 June 2021, 08:46 PM.
        Test signature


        • #34
          RDNA id for gaming. I doubt it can manage fast FP64. CDNA is for FP64 compute.


          • #35
            Originally posted by flashmozzg View Post
            RDNA id for gaming. I doubt it can manage fast FP64. CDNA is for FP64 compute.
            Not everything needs FP64.