Mac mini 2018 eGPU TB3 bandwidth

Discussion in 'Mac mini' started by avxkim, May 12, 2019.

  1. avxkim macrumors member

    avxkim

    Joined:
    Nov 5, 2017
    Location:
    Kazakhstan
    #1
    I'm using samsung T5 + Sonnet box with RX 580, what TB3 i have to use, so my external SSD won't cut eGPU's bandwidth?
     
  2. Stephen.R macrumors 65816

    Stephen.R

    Joined:
    Nov 2, 2018
    Location:
    Thailand
    #2
    Just plug them into different ports on the mini.
     
  3. avxkim thread starter macrumors member

    avxkim

    Joined:
    Nov 5, 2017
    Location:
    Kazakhstan
    #3
    I mean, there's only 2 TB3 controllers on the mac mini, for example, i plugged eGPU in a port 1, then external SSD on port 3. Is it ok? I just noticing that my FPS in games, are below the averages, e.g. dota 2 on max settings @1080p are hovering in 120-130FPS on iMac 5k Radeon Pro 580, but RX 580 is a bit better and i get 80-100 FPS.
     
  4. Zdigital2015 macrumors 65816

    Joined:
    Jul 14, 2015
    Location:
    East Coast, United States
    #4
    Yes, Port 1 and 2 are handled by one controller and Port 3 and 4 are handled by the other controller, so the eGPU in Port 1 and external SSD in a Port 3 are probably the best.

    I cannot speak to FPS differences as I do not game, at least not on a Mac.
     
  5. brentsg macrumors 68040

    Joined:
    Oct 15, 2008
    #5
    What averages? TB3 gives the GPU 4 PCIE lanes vs the usual 16 so there is always some compromise.
     
  6. avxkim thread starter macrumors member

    avxkim

    Joined:
    Nov 5, 2017
    Location:
    Kazakhstan
    #6
    Is it worth to upgrade RX 580 to a Vega 56 in order to increase FPS?
     
  7. Zdigital2015, May 12, 2019
    Last edited: May 12, 2019

    Zdigital2015 macrumors 65816

    Joined:
    Jul 14, 2015
    Location:
    East Coast, United States
    #7
    For a game? You mean 80-100fps is not enough? Yeah, I am going to have to defer on that one before I get really snarky and says some something I should not...

    Seriously...I would not spend money on a Vega 56 with Computex and AMD’s 7nm Navi GPU announcement right around the corner!

    Source: https://wccftech.com/amd-navi-radeon-rx-gpu-rumors-navi-20-2020-rx-navi-graphics-cards/

    Good luck!

    EDIT: You also need to make sure the Thunderbolt 3 cable you are using can handle 40Gbps and is not a 20Gbps cable. Make sure you are using the cable that Sonnet shipped with the eGPU. If not, chances are you have a cable that can only handle 20GBps and not one of the more expensive active cables that sell for $40-$50.
     
  8. richpjr macrumors 68040

    richpjr

    Joined:
    May 9, 2006
    #8
    Depends on the game, but a 580 is not going to give 80-100fps on any demanding games. And since we don't know when the new Navi will actually be released considering a Vega 56 is not at all an unreasonable thing to do if an increase is needed today, instead of sometime in 2020.
     
  9. rmdeluca macrumors 6502

    Joined:
    Oct 30, 2018
    #9
    As others have said, you’re connecting it right by keeping the eGPU on its own TB3 controller.

    The higher you push frame rates, the more that PCIe bandwidth matters. On current gen hardware, TB3 will only give the eGPU about 22Gb/s (10Gb/s of the 32Gb/s is reserved by the controller). This is less than 3 full PCIe 3.0 lanes, compared to what the iMac’s GPU gets (at least 8, likely 16? Don’t quote me on that) or a full desktop that gets 16 lanes. Point being, you’re almost certainly bandwidth constrained.

    If you’re booting into Windows, you can check this easily using GPU-Z, which can log bus utilization to a text file while you’re playing the game. If you’re regularly seeing instances of 100% utilization, those are moments when the FPS is bandwidth constrained.

    If you want to get equivalent frame rates you’ll have to use a higher class of card (like a Vega). But this is an avenue of diminishing returns - you’re trying to compensate for the reduced bandwidth by cutting down frame render time by throwing more GPU at the problem. At some point the bus becomes the dominant factor.
    --- Post Merged, May 13, 2019 ---
    I know it seems absurd, but gaming (especially competitive multiplayer) is moving towards 120-144Hz. For many of us, it’s about time. The visual difference between 60 and 120Hz is remarkable.

    As for competitive gaming, with double and triple buffering, the difference between 60Hz and 120Hz is human perceptible in terms of input lag. Granted, for 99%+ of players their skills are the bottleneck, not their frame rates, but it still “feels” nicer regardless.
     
  10. avxkim thread starter macrumors member

    avxkim

    Joined:
    Nov 5, 2017
    Location:
    Kazakhstan
    #10
    @rmdeluca, thanks for the detailed response. I'm still thinking, is it worth to upgrade from RX580 to a Vega 56...
     
  11. Zdigital2015 macrumors 65816

    Joined:
    Jul 14, 2015
    Location:
    East Coast, United States
    #11
    Make sure your eGPU can handle the power requirements of a Vega 56 before you buy one...
    --- Post Merged, May 14, 2019 ---
    For paying work that would benefit from a faster GPU, helping get the job done faster, make more money, sure...for some extra FPS on a game...no...unless that’s your profession.
     
  12. richpjr macrumors 68040

    richpjr

    Joined:
    May 9, 2006
    #12
    I would say a hard core gamer would absolutely pay for extra FPS, but in reality a hard core gamer would likely get a Windows gaming PC and not play on a mini with an eGPU.
     
  13. Zdigital2015 macrumors 65816

    Joined:
    Jul 14, 2015
    Location:
    East Coast, United States
    #13
    Different strokes for different folks...faster editing, more effects, layers, LUTs, etc., quicker turnaround, more quality time and money - I'm there...probably why I don't pine for the return of NVIDIA GPUs, because unless it can speed up a workflow, I could care less how many FPS it can push in Far Cry 5, Anthem, Fortnite, et al.

    My sympathy extends to people who want a Mac and need CUDA, but that's about it.

    Again, different strokes for different folks...
     

Share This Page

12 May 12, 2019