EVGA

1080ti power efficiency

Author
Edome
iCX Member
  • Total Posts : 271
  • Reward points : 0
  • Joined: 2017/10/14 13:31:21
  • Status: offline
  • Ribbons : 1
2017/10/15 03:11:28 (permalink)
Tested today the power efficiency for various voltage points @ EBWF zcash mining:
 
2025MHz 1080mV 766sol/s 338W 2.26sols/W
 
1999MHz 1042mV 760sol/s 303W 2.51sols/W
 
1936MHz 962mV 740sol/s 253W 2.92sols/W
 
1835MHz 875mV 703sol/s 205W 3.43sols/W
 
1721MHz 800mV 661sol/s 168W 3.93sols/W
 
Undervolting works like magic for 1080ti!
#1

20 Replies Related Threads

    QuintLeo
    SSC Member
    • Total Posts : 946
    • Reward points : 0
    • Joined: 2016/04/16 23:05:09
    • Status: offline
    • Ribbons : 3
    Re: 1080ti power efficiency 2017/10/17 20:24:53 (permalink)
    Did some testing about a week back under the most recent EBWF version.
     
     All at +100 core +100 memory in afterburner (seems to be the best RELIABLE spot in Afterburner, +150 core worked better but wasn't long-term reilable) on a Aorus (not the extreme) on a Windows 7 machine with 382.53 drivers.
     
     No undervolting (haven't gotten it to WORK on my 1080ti yet).
     
     125 watt TDP (50%) = 505 Sol/sec
     150 watt TDP (60%) = 620 Sol/sec
     175 watt TDP (70%) = 680 Sol/sec
     200 watt TDP (80%) = 725 Sol/sec
     225 watt TDP (90%) = 760 Sol/Sec
     250 watt TDP (100%) = 780 Sol/Sec
     
     My 175 and 200 watt TDP-limited figures are very close to your undervolted similar wattage figures via undervolting - I suspect we're ending up at close to the same voltage due to TDP limiting.
     The power figures I saw were what Nvidia-SMI reported, the actual figure and hashrate bounced around a bit like on any mining.
     
     I'm curious WHICH model of 1080ti you pushed past 300 watts power draw and what you used to estimate it with, as well as which driver version and OS you were using.
     

    Now that vorsholk has stopped his abuse, I'm returning to folding.
     I no longer MOO due to abuses by certain "whales" in the Gridcoin community - so I now work the Distributed.net project directly again.
     
    #2
    Chris21010
    FTW Member
    • Total Posts : 1587
    • Reward points : 0
    • Joined: 2006/05/03 07:26:39
    • Status: offline
    • Ribbons : 9
    Re: 1080ti power efficiency 2017/10/17 21:05:20 (permalink)
    i would strongly think that the wattage he listed is system power from the wall and not GPU power. there is no way a 1080ti even under cooled water could get close to 300W much less surpass it.


    #3
    Edome
    iCX Member
    • Total Posts : 271
    • Reward points : 0
    • Joined: 2017/10/14 13:31:21
    • Status: offline
    • Ribbons : 1
    Re: 1080ti power efficiency 2017/10/18 03:27:07 (permalink)
    QuintLeo
     I'm curious WHICH model of 1080ti you pushed past 300 watts power draw and what you used to estimate it with, as well as which driver version and OS you were using.
     

    This Palit Gamerock Premium 1080ti has 2x 8pin with a 350W power limit, the measurement is from wall minus the idle consumption (70-80W)of the system, the driver used was 385.69 WHQL and OS is Win10.
    The PSU efficiency is 90% @ 50% load.
     
    In Timespy Extreme Graphics Test 2 gpu voltage drops to a minimum of 1000mV with the 350W limit.
    Now I'm using Asus XOC bios so I have eliminated any power limits and the peak power draw from wall with 1100mV is 490W(- idle load) so the PC actually uses 440W with the 90% efficiency, the GPU peaks @ 420W as estimated by Hwinfo. I'm using custom 120mm high pressure fans on the heatsink so temps are never a problem.
     
    The closest estimation of the actual GPU power peak is (490W-idle consumption 80W) * 90% = 369W
     
    #4
    Chris21010
    FTW Member
    • Total Posts : 1587
    • Reward points : 0
    • Joined: 2006/05/03 07:26:39
    • Status: offline
    • Ribbons : 9
    Re: 1080ti power efficiency 2017/10/18 05:57:35 (permalink)
    Wow, bios swapping... Very very risky. But it is interesting to hear what a possible unlocked gpu could do.


    #5
    Edome
    iCX Member
    • Total Posts : 271
    • Reward points : 0
    • Joined: 2017/10/14 13:31:21
    • Status: offline
    • Ribbons : 1
    Re: 1080ti power efficiency 2017/10/18 10:03:23 (permalink)
    With 1080ti it's pretty safe to crossflash BIOS.
     
    The only known softbrick is if you flash a HOF or Lightning BIOS that uses 3x 8pin.
    Then you need a spare graphics card or iGPU to restore the original BIOS.
     
    Also my Palit card has a bios switch similar to FTW3 if all else fails :)
    post edited by Edome - 2017/10/18 10:11:42
    #6
    QuintLeo
    SSC Member
    • Total Posts : 946
    • Reward points : 0
    • Joined: 2016/04/16 23:05:09
    • Status: offline
    • Ribbons : 3
    Re: 1080ti power efficiency 2017/10/18 11:01:12 (permalink)
    Aorus models also have dual 8 pin connectors - I've never run mine past 275 watts but it runs COOL enough it should handle a lot more.
     
     Quite a few other higher-end 1080 ti models also have dual 8 pin connectors.
     In theory that allows for up to 375 watts without exceeding PCI-E specs for power....
     
     

    Now that vorsholk has stopped his abuse, I'm returning to folding.
     I no longer MOO due to abuses by certain "whales" in the Gridcoin community - so I now work the Distributed.net project directly again.
     
    #7
    Edome
    iCX Member
    • Total Posts : 271
    • Reward points : 0
    • Joined: 2017/10/14 13:31:21
    • Status: offline
    • Ribbons : 1
    Re: 1080ti power efficiency 2017/10/18 11:39:33 (permalink)
    Both 
    QuintLeo
     In theory that allows for up to 375 watts without exceeding PCI-E specs for power....
     

     
    The spec is overprotective :)
    Normal 18awg 6pin (2-3x 12V yellow lines, most PSUs got 3) can easily do 12V*8A*2= 192 watts without heating and 8pin 12V*8A*3= 288 watts
     
    Some PSUs have 16awg which can do safely even more.
    post edited by Edome - 2017/10/18 11:40:53
    #8
    Chris21010
    FTW Member
    • Total Posts : 1587
    • Reward points : 0
    • Joined: 2006/05/03 07:26:39
    • Status: offline
    • Ribbons : 9
    Re: 1080ti power efficiency 2017/10/18 11:51:00 (permalink)
    you have to remember to take those numbers and multiply by 80% due to constant power and not intermittent power.


    #9
    QuintLeo
    SSC Member
    • Total Posts : 946
    • Reward points : 0
    • Joined: 2016/04/16 23:05:09
    • Status: offline
    • Ribbons : 3
    Re: 1080ti power efficiency 2017/10/19 09:55:32 (permalink)
    The spec IS massively overprotective - the actual 6-pin connector used by the PCI-E spec is rated at 8 amps continuous per pin (13 amps free-air max per pin but that is derated some for heat retention by the connector), or 24 amps (3 +12VDC and 3 ground) = 288 watts at 12VDC per connector, so there is ZERO derating required for 24/7 continuous operation as the connector is ALREADY derated to barely over 25% of it's actual rated power capacity by the PCI-E spec (75 Watts per PCI-E).
     If they are only using 2 of the designated +12VDC pins for power (and the other for sense), that's 16 amps or 192 watts which STILL leaves the connector at well under 40% of it's rated capacity.
     
    The PCI-E 8-pin connector spec still only uses 3 of the pins for +12VDC which STILL leaves the connector at barely over 50% of it's rated power capacity (150 watts  per PCI-E)
     
     WIRING, on the other hand, needs to be watched at derated capacity in 24/7 usage as many power supply makers cheap out a bit on that and push it a LOT closer to it's actual specs (PCI-E doesn't specify the wiring size AFAIK).
     
     While the wire used by some power supplies is 16 AWG, that's normally used on 2-connector lead sets not just on singles and you have to watch what the rated draw is from BOTH connectors - and 16 AWG is only rated for 10 amps max anyway, or 8 amps for continuous usage.
     
     

    Now that vorsholk has stopped his abuse, I'm returning to folding.
     I no longer MOO due to abuses by certain "whales" in the Gridcoin community - so I now work the Distributed.net project directly again.
     
    #10
    gt3911
    New Member
    • Total Posts : 5
    • Reward points : 0
    • Joined: 2014/09/23 08:52:52
    • Status: offline
    • Ribbons : 0
    Re: 1080ti power efficiency 2017/11/18 07:55:43 (permalink)
    Have any of you guys found much to be gained by reducing the power target to like... 60%? To  help find an efficient sweet spot?
    #11
    Chris21010
    FTW Member
    • Total Posts : 1587
    • Reward points : 0
    • Joined: 2006/05/03 07:26:39
    • Status: offline
    • Ribbons : 9
    Re: 1080ti power efficiency 2017/11/18 08:37:45 (permalink)
    that is basically the same thing just with a different metric. lowering the power target will in turn lower the max voltage to limit the max power.


    #12
    gt3911
    New Member
    • Total Posts : 5
    • Reward points : 0
    • Joined: 2014/09/23 08:52:52
    • Status: offline
    • Ribbons : 0
    Re: 1080ti power efficiency 2017/11/18 08:59:59 (permalink)
    Ah I see! I'm quite new to this. 
    I've been lowering my power target by 5% and checking for stability issues but I'm still a bit confused. 
     

     
    I don't really understand how I'm running quicker than my boost clock with no overclocking yet define on the offset.
    #13
    QuintLeo
    SSC Member
    • Total Posts : 946
    • Reward points : 0
    • Joined: 2016/04/16 23:05:09
    • Status: offline
    • Ribbons : 3
    Re: 1080ti power efficiency 2017/11/18 11:13:14 (permalink)
    Lowering the target TDP does work to improve efficiency - and is more widely available than direct undervolting (LINUX has power limit control, but I've yet to find a way to set undervolting in software under LINUX).
     
     In my testing with my Aorus GTX 1080 ti (and later spot checking with my Gigabyte and EVGA 1080 ti cards), it seems that pretty much any 1080 ti is capable of a bit over 4 efficiency using EBWF at +100 core +100 memory and TDP set to around 60% (150 watts) - but hashrate starts dropping fast much lower than that, and by the time you get to 125 watts (50% TDP) the efficiency starts DROPPING again because the core clock is getting severely power starved.
     
     I'm not claming this as the ultimate efficiency setup as I've not played with the 1080 ti as much as my 1070 ti cards, but I suspect it's fairly close.
     
     Boost clock has quite a few factors that go into it - it's the CARD that decides how high it can boost, the published figures you see for cards is a specified MINIMUM that any card of that model should be able to achieve on stock settings and pretty much ALL cards will be able to exceed that spec. How far a specific card will go depends a lot on the so-called "silicon lottery", which is a way of saying "how few flaws the wafer had in the area that your particular GPU came out of".
     

    Now that vorsholk has stopped his abuse, I'm returning to folding.
     I no longer MOO due to abuses by certain "whales" in the Gridcoin community - so I now work the Distributed.net project directly again.
     
    #14
    Chris21010
    FTW Member
    • Total Posts : 1587
    • Reward points : 0
    • Joined: 2006/05/03 07:26:39
    • Status: offline
    • Ribbons : 9
    Re: 1080ti power efficiency 2017/11/18 12:51:07 (permalink)
    gt3911 
    I don't really understand how I'm running quicker than my boost clock with no overclocking yet define on the offset.



    because of this: https://www.geforce.com/hardware/technology/gpu-boost/technology
     
    all nvidia cards will boost past factory settings until it reaches either thermal or power limits.


    #15
    QuintLeo
    SSC Member
    • Total Posts : 946
    • Reward points : 0
    • Joined: 2016/04/16 23:05:09
    • Status: offline
    • Ribbons : 3
    Re: 1080ti power efficiency 2017/11/19 12:44:07 (permalink)
    The factory rated "boost clock" for a card model is a guarentee that the card will get AT LEAST that high on boost clock, most will do better and some quite a bit better.
    It also depends on the type of workload - some types like ZEC mining load a card down more than FAH or ETH mining or Dnet RC5-72 work (FAH loses some GPU loading when it's passing mass data to/from the CPU, ETH is heavy memory limited on most cards, DNet is entirely core limited and barely touches memory at all).
     
     
     

    Now that vorsholk has stopped his abuse, I'm returning to folding.
     I no longer MOO due to abuses by certain "whales" in the Gridcoin community - so I now work the Distributed.net project directly again.
     
    #16
    gt3911
    New Member
    • Total Posts : 5
    • Reward points : 0
    • Joined: 2014/09/23 08:52:52
    • Status: offline
    • Ribbons : 0
    Re: 1080ti power efficiency 2017/11/19 13:03:19 (permalink)
    Great information, thanks heaps :)
    #17
    igoricom
    New Member
    • Total Posts : 1
    • Reward points : 0
    • Joined: 2017/11/27 16:47:07
    • Status: offline
    • Ribbons : 0
    Re: 1080ti power efficiency 2017/11/27 16:58:23 (permalink)
    Does anybody know if cryptocurrency mining voids the 5years and 10years extended warranty from EVGA?
    Just purchased 3x1080ti because the vendor assured me it covers mining since you keep it original and don't OC.
    Thanks
    #18
    ipkha
    CLASSIFIED Member
    • Total Posts : 2308
    • Reward points : 0
    • Joined: 2015/02/10 21:01:40
    • Location: Indiana
    • Status: offline
    • Ribbons : 6
    Re: 1080ti power efficiency 2017/11/28 05:32:32 (permalink)
    Yes, as long as you don't make any hard mods to the card you will be fine for warranty. Long term mining, especially when undervolting for power efficiency, is unlikely to negatively impact your card.


    #19
    Slegir
    New Member
    • Total Posts : 11
    • Reward points : 0
    • Joined: 2017/09/25 08:30:50
    • Status: offline
    • Ribbons : 0
    Re: 1080ti power efficiency 2017/12/19 07:22:05 (permalink)
    igoricom
    Does anybody know if cryptocurrency mining voids the 5years and 10years extended warranty from EVGA?
    Just purchased 3x1080ti because the vendor assured me it covers mining since you keep it original and don't OC.
    Thanks




    99% of the time, OC'ing your card makes no sense for mining. you want to underclock, not overclock.
    #20
    Chris21010
    FTW Member
    • Total Posts : 1587
    • Reward points : 0
    • Joined: 2006/05/03 07:26:39
    • Status: offline
    • Ribbons : 9
    Re: 1080ti power efficiency 2017/12/19 10:34:29 (permalink)
    you are wrong, but also right.
     
    you DO want to OVER CLOCK but also UNDER VOLT at the same time. sense higher clock speeds at a low power draw is the ultimate goal for mining.


    #21
    Jump to:
  • Back to Mobile