PSA: RX480 and unpowered risers (exceeds 75w ATX spec)

2»

Comments

  • HarryBoyHarryBoy Member Posts: 1
    Hi guys. I tried to run 3 x Sapphire RX480s last night on an Asus P8P67 Pro motherboard (from i7-2600K days).

    Using either unpowered PCIEx16 risers, powered PCIEx16 risers, or the fancy powered USB cable risers, I was unable to get anything other than 1 of the cards working. Sometimes I would get lockups in Windows, or endless reboot cycles.

    After a few hours, I tried the cards directly in the PCIEx16 slots (no risers). This is working well. However, the main point was to use powered risers to avoid drawing too much power thru the motherboard.

    I am using a 1600W Enermax power supply (something I had lying around from mining a few years ago). Has anyone else had problems with risers and the RX480 ? I have no history with this motherboard and multiple GPUs on risers, so I have no previous experience to go by.
  • Marvell9Marvell9 Member Posts: 593 ✭✭✭
    dlehenky said:

    What I don't get is that the rx 480, a 14 nm process chip, is drawing as much power as the Nano, which is 28 nm process chip. It should have been a slam-dunk for AMD to get *magical* power reduction numbers for the rx 480, but apparently not. I can only assume, at this point, that the rx 480 represent AMD's first-pass effort at moving to the 14 nm chip process, and as such, the chip design is anything but optimized for the new process. In other words, it appears to me that the rx 480 is a "let's get something out the door quick" offering, and doesn't show anything near what can be accomplished on a robust/mature 14 nm process chip design. I'd call it a rev 0.9 type effort, if that.

    Die shrink does not automaticlaly = lesser power. The newer chips actually pack far more silicon into a smaller area and the new chips are more powerful by a factor of 3 than the old 28nm chips.

    Youre getting 290x perfomance from half the amount of silicon and cheaper fabbed proccesses so I dont think you can expect a 50% power reduction as people seem to expect.

    The 480 cards are cheaper becuase they use a cheaper 14nm procces less physical silicon and they do hash at around 30% less power than their older Hawaii brethen.

    Also keep in mind ive seen some posts of rx 480s hashing at 33mhs at 150 watts, so we will prbaby see a better hash/watt rate when better drivers come out.
  • SIRacer09SIRacer09 Member Posts: 246 ✭✭
    Marvell9 said:

    dlehenky said:

    What I don't get is that the rx 480, a 14 nm process chip, is drawing as much power as the Nano, which is 28 nm process chip. It should have been a slam-dunk for AMD to get *magical* power reduction numbers for the rx 480, but apparently not. I can only assume, at this point, that the rx 480 represent AMD's first-pass effort at moving to the 14 nm chip process, and as such, the chip design is anything but optimized for the new process. In other words, it appears to me that the rx 480 is a "let's get something out the door quick" offering, and doesn't show anything near what can be accomplished on a robust/mature 14 nm process chip design. I'd call it a rev 0.9 type effort, if that.

    Die shrink does not automaticlaly = lesser power. The newer chips actually pack far more silicon into a smaller area and the new chips are more powerful by a factor of 3 than the old 28nm chips.

    Youre getting 290x perfomance from half the amount of silicon and cheaper fabbed proccesses so I dont think you can expect a 50% power reduction as people seem to expect.

    The 480 cards are cheaper becuase they use a cheaper 14nm procces less physical silicon and they do hash at around 30% less power than their older Hawaii brethen.

    Also keep in mind ive seen some posts of rx 480s hashing at 33mhs at 150 watts, so we will prbaby see a better hash/watt rate when better drivers come out.
    Dang, 33mhs! What settings are they using? I've only got my XFX 8GB 480s to 28.5. :)

  • Marvell9Marvell9 Member Posts: 593 ✭✭✭
    @SIRacer09 not me some guy on altcoin mining forums posted some pics its a linux driver miod
  • dlehenkydlehenky Member Posts: 2,249 ✭✭✭✭
    @Marvell9 Of course, you are correct, 14 nm != less power; it's a design tradeoff. 14 nm has a *much* higher potential from lowering power, or denser circuity, or higher clocks, compared to 28 nm, but if the design priority is to just stuff more transistors onto the silicon and/or pump the clocks up, then there goes the opportunity for lower power. I'm just saying AMD, at least in this first 14 nm offering, didn't attempt to improve power much beyond what they had already done with the 28 nm Nano. Then again, they've had several years to mature their 28 nm designs. I expect as they get further down the 14 nm road, we'll see the same, or better, circuit density and clock frequencies, but also with reduced power consumption. In other words, the rx 480 is obviously a young (immature) 14 nm design, at this point.
  • workwork Member Posts: 2,084 ✭✭✭✭
    @dlehenky I would bet a big part of the power draw difference isn't the chip tho. HBM on the Nanos vs savagely high-clocked GDDR5 on the RX 480 have pretty massively different power requirements.
  • dlehenkydlehenky Member Posts: 2,249 ✭✭✭✭
    Ah, yes, I hadn't even thought about the memory differences. I've never actually run anything but the Nanos, to this point, but sure, the GDDR memory's got to suck more juice than HBM.
  • HelioxHeliox Member, Moderator Posts: 634 mod
    dlehenky said:

    Ah, yes, I hadn't even thought about the memory differences. I've never actually run anything but the Nanos, to this point, but sure, the GDDR memory's got to suck more juice than HBM.

    Very true.

    Knowing that i can lower the voltage for hbm with -175mV and still be stable enough to run it at 400Mh.

    Also knowing that the voltage controller on the new 480's is the same as on the fiji series. This means that i'll be able to reduce the voltage for the memory (not the controller, the memory itself) as well, just not sure if it will handle -100+mV .. :tongue: (pretty sure it won't .. :smile: )
  • peepeedogpeepeedog Member Posts: 32
    edited July 2016
    Alright,

    My H97 is in and i tossed out the old fried H81.

    My stock RX480 is getting a 24.8 hashrate with the latest beta 16.7.1 drivers running off Claymore 4.7, using powered risers.

    Even though the new drivers should in theory put less stress on the PCIE slot, i.e. bring it to spec, it's still drawing a higher wattage compared to previous gen cards. So please use powered risers guys, I'm a great example of what happens when you don't, on an aging board.

    *EDIT*

    Did some tweaking:

    Stock clocks: 24.8

    Memory overclocked to 2200
    Undervolt to 900

    27.5 MHs

    So far so good, pretty stable, and i think it's a decent mining card for the price/power.


    Cheers
    Shaun


    Post edited by peepeedog on
  • bbcoinbbcoin Member Posts: 377 ✭✭✭
    edited July 2016
    I'm running 16.6.2 4GB sapphire 480's

    Undervolt to 860 memory to 800
    -20% core frequency (around 1013mhz)

    And I'm getting stable 24.1mhs with about 100W- 110W at the wall excluding CPU and motherboard.

    Rig has been running for a few days now and it's very quiet + stable. I'm really happy with it. Used ribbon powered risers.
  • pantadorpantador Member Posts: 46
    Are non ref RX 480/470, especially with 8 pins, draw less power from the pcie slot?
    Maybe for 4GB model, that's currently "locked" @1750MHz if they use lower memV.

    If not then must use the same slot/risers caution like the ref cards I think.
Sign In or Register to comment.