choose 470 over 480 anytime!

ursul0ursul0 Member Posts: 54
I have mixed rig here: 4*470*4GB (3*Sapphire refs and 1*Asus Strix) and 1*480*8G (MSI ref) all on raisers.
1) all flashed with tighter timings from 1500
2) memory and core set to 0.857v
3) all mem runs @ 2000MHz
4) 480 drives display (do note that if you try to stick display in any of 470s - you'd need to increase the voltages, otherwise won't even start mining)
5) using watttool to push mem voltage
6) only the drivers from 16.9.1. No wattman gui - it feels more stable to me this way.

Using claymore dual, each 470 gives 28.3MH and 480 does 29.1MH, total power off the wall is 825W:




P.S. same results on B150 Pro4/D3

Comments

  • ursul0ursul0 Member Posts: 54
    side bonus - this setup (no proper AMD driver install) allows to run fine without any display connected, you just won't see the sensors data :) ... can be useful, if you are sure it's all running smoothly
  • josechudmjosechudm Member Posts: 63
    i got 4 asus strix 4 gb 470 but im affraid to flash cause it doesnt have dual bios, what do you recommend?
  • RatRat Member Posts: 25
    @ursul0 : This is true mostly until POS. after that ? 8gb is definitely better option. cryptonite hashing definitely is better in 8gb. I have 470 4gb and 8gb. both modded. 4gb gives me 28mhs better than 8gb which gives 27mhs.
  • techtottechtot Member Posts: 339 ✭✭✭
    The Mh/s is good, but 825 is really high for watts. my 4 card 480 rig only draws around 550 watts at wall in a system.
  • mjaaaymjaaay Member Posts: 97
    Interesting. What exactly does the -7% on Power Limit do?
  • RavinderDhillonRavinderDhillon Member Posts: 74 ✭✭
    edited September 2016
    <
    P.S. same results on B150 Pro4/D3 >


    Do you have 5 cards running on the Asrock B150 Pro4/D3? I was considering buying that board.
    Post edited by RavinderDhillon on
  • ursul0ursul0 Member Posts: 54
    josechudm said:

    i got 4 asus strix 4 gb 470 but im affraid to flash cause it doesnt have dual bios, what do you recommend?

    make a backup of the original and try to mod yourself. but hey that's what I would do...
  • ursul0ursul0 Member Posts: 54
    Rat said:

    @ursul0 : This is true mostly until POS. after that ? 8gb is definitely better option. cryptonite hashing definitely is better in 8gb. I have 470 4gb and 8gb. both modded. 4gb gives me 28mhs better than 8gb which gives 27mhs.

    hmmm... interesting. I need to educate myself more on the matter.. I've got pretty vague idea on what POS arrival would actually mean. does anyone?
    techtot said:

    The Mh/s is good, but 825 is really high for watts. my 4 card 480 rig only draws around 550 watts at wall in a system.

    are you sure yours runs in dual mining mode? what are the settings?
    mjaaay said:

    Interesting. What exactly does the -7% on Power Limit do?

    lol... it makes me feel good:) well... and maybe saves a few watts. isn't it suppose to reduce the max amperage the card can pull?
    -7% is actually around the very bottom the cards would allow without hashrate reduction under these settings. 480 is doing fine with -15%


  • ursul0ursul0 Member Posts: 54
    @RavinderDhillon nope, not yet. It was actually my testing rig where I still have 3 cards (one of each type mentioned)... but I see no reason why it wouldn't do just fine.
  • RatRat Member Posts: 25
    @ursul0 In POS proof of stake , v don't mine new coin's. V stake our coins to get coins. Atleast for now it looks like our gpu will be useless if pos happened. v need to move on to other coins.

    Better explanation https://forum.ethereum.org/discussion/8411/why-are-miners-scared-of-pos
  • techtottechtot Member Posts: 339 ✭✭✭
    techtot said:

    The Mh/s is good, but 825 is really high for watts. my 4 card 480 rig only draws around 550 watts at wall in a system.

    are you sure yours runs in dual mining mode? what are the settings?

    Not doing dual mode, but I think that only adds about 20-30watts per card? I guess that would put me @670 if dual mining... I have heard the 470 draws more juice, even though it is a cheaper card , you may pay the difference in electricity.. ?

  • ursul0ursul0 Member Posts: 54
    edited September 2016
    @techtot Not sure you can trust these numbers, but looks like you are correct


    I'd guess the temps are to blame
  • ursul0ursul0 Member Posts: 54
    yep, the sensor data look proper.
    meaning that 480 draws ~130W and 470 ~155W
  • Marvell9Marvell9 Member Posts: 593 ✭✭✭
    the 480s can function at a lower voltage thus save power thatway since any 480 with 8gb ram is usually high asic quality and samsung ram
  • ursul0ursul0 Member Posts: 54
    Marvell9 said:

    the 480s can function at a lower voltage thus save power thatway since any 480 with 8gb ram is usually high asic quality and samsung ram

    maybe so but in this case the voltages are the same .85, check out the image above with sensors data.

    BTW anyone knows what "GPU memory errors" are? maybe it is a factor of sorts...
  • techtottechtot Member Posts: 339 ✭✭✭
    Are you adding GPU VRM Power in + GPU Power to get the total? If so , then yes, looks like the 480 runs 20-30 watts lower. Probably not that significant unless you have more than 20 cards..
  • RavinderDhillonRavinderDhillon Member Posts: 74 ✭✭
    edited September 2016
    ursul0 said:


    BTW anyone knows what "GPU memory errors" are? maybe it is a factor of sorts...

    Memory errors signify that the Ram is being stretched too far ( either by tight timings / high overclocks or low voltages) and its basically leaking data, and your numbers are pretty high. This could cause more operations/cycles for the memory until it generates accurate results, thus slowing down the output.

    I tried the 1375 memory timing strap on my cards and while they were stable for almost a week, hashrates at the pool were consistently 10% lower at least, far too long for it to be luck, so that made me curious. To test this thoroughly I tried gaming on these cards with the 1375 strap and while game benches did not crash, I could see frame drops everywhere and stuttering telling me that the memory was struggling to keep up, and having to possibly rework, which could explain my consistently 10% lower pool hashrate.
    This was fixed by increasing voltage significantly, but that increased power draw as well, so I returned to the 1500 timings strap reducing my hash from 27.5 to 26.8 which I can live with. Very few, if any memory errors now on any of my cards, and pool hashrate has improved.

    Pool hashrate increase could ofcource just be pure luck as claymore reported hashrate wasn't impacted by memory errors, and I did not get any rejected shares, but I chose the safe stable way out.
    Do let us know if you also see a reduced poolside hashrate with high memory errors over time.
  • ursul0ursul0 Member Posts: 54
    edited September 2016
    techtot said:



    Not doing dual mode, but I think that only adds about 20-30watts per card? I guess that would put me @670 if dual mining... I have heard the 470 draws more juice, even though it is a cheaper card , you may pay the difference in electricity.. ?

    You should absolutely run claymore in dual mode, even if with intensity for subcoin around 10. Claymore writes about dual mode: "Effective Ethereum mining speed is higher by 3-5% because of a completely different miner code - much less invalid and outdated shares, higher GPU load, optimized OpenCL code"
    And it seems to be so.

    It's magic - it gives me almost 1MH ETH on the 470s compared to -mode 1, and all my cards are colder!

    GPU1 is 480, rest is 470, and ignore GPU4( it's R9 370 nitro, I'm utilizing the x16slot)

    mode -1:
    ETH: GPU0 27.954 Mh/s, GPU1 28.944 Mh/s, GPU2 27.806 Mh/s, GPU3 27.928 Mh/s, GPU4 14.187 Mh/s, GPU5 27.954 Mh/s
    ...
    ETH: GPU0 27.999 Mh/s, GPU1 29.010 Mh/s, GPU2 27.932 Mh/s, GPU3 28.006 Mh/s, GPU4 14.156 Mh/s, GPU5 27.963 Mh/s

    and now:
    -dcri 12, 12, 12, 12, 12, 12
    ETH: GPU0 28.733 Mh/s, GPU1 26.368 Mh/s, GPU2 28.652 Mh/s, GPU3 28.738 Mh/s, GPU4 12.670 Mh/s, GPU5 28.743 Mh/s

    ...for some reason the 480 actually dropped down from 29

    -dcri 12, 18, 12, 12, 12, 12 - this makes things a bit better:
    ETH: GPU0 28.720 Mh/s, GPU1 28.704 Mh/s, GPU2 28.644 Mh/s, GPU3 28.696 Mh/s, GPU4 12.637 Mh/s, GPU5 28.711 Mh/s
    ...
    ETH: GPU0 28.698 Mh/s, GPU1 28.680 Mh/s, GPU2 28.637 Mh/s, GPU3 28.697 Mh/s, GPU4 12.644 Mh/s, GPU5 28.710 Mh/s
    SC: GPU0 172.189 Mh/s, GPU1 258.123 Mh/s, GPU2 171.821 Mh/s, GPU3 172.179 Mh/s, GPU4 75.863 Mh/s, GPU5 172.257 Mh/s


    All cards run several degrees colder, and you get another coin for few more watts.

    Edit: total power for the whole rig is 880-900W
  • ursul0ursul0 Member Posts: 54

    ursul0 said:


    BTW anyone knows what "GPU memory errors" are? maybe it is a factor of sorts...

    Memory errors signify that the Ram is being stretched too far ( either by tight timings / high overclocks or low voltages) and its basically leaking data, and your numbers are pretty high. This could cause more operations/cycles for the memory until it generates accurate results, thus slowing down the output.

    I tried the 1375 memory timing strap on my cards and while they were stable for almost a week, hashrates at the pool were consistently 10% lower at least, far too long for it to be luck, so that made me curious. To test this thoroughly I tried gaming on these cards with the 1375 strap and while game benches did not crash, I could see frame drops everywhere and stuttering telling me that the memory was struggling to keep up, and having to possibly rework, which could explain my consistently 10% lower pool hashrate.
    This was fixed by increasing voltage significantly, but that increased power draw as well, so I returned to the 1500 timings strap reducing my hash from 27.5 to 26.8 which I can live with. Very few, if any memory errors now on any of my cards, and pool hashrate has improved.

    Pool hashrate increase could ofcource just be pure luck as claymore reported hashrate wasn't impacted by memory errors, and I did not get any rejected shares, but I chose the safe stable way out.
    Do let us know if you also see a reduced poolside hashrate with high memory errors over time.
    Yes my thoughts exactly. I'd guess that mem controller is working like crazy trying to perform error correction.

    I also notice effective hashrate on the pool to be 'round 10% lower the actual, on average. But I don't think it has anything to do with the overloaded cards. Well... at least I cannot imagine how it may actually be that way. (I think the pool's inner-working is the reason for that)
  • ursul0ursul0 Member Posts: 54
    @RavinderDhillon you think it can be somehow possible that submitted shares somehow do not correspond to the hashrate at which the hashes are produced...?
Sign In or Register to comment.