1080 Ti mining OC and power?

[dagarair]

Quint you don't happen to have a 1070Ti do you?  I am about to start to config one tomorrow but wondering if you have a starting point =)

[1714042678]

1714042678

[1714042678]

1714042678

[1714042678]

1714042678

[1714042678]

1714042678

[1714042678]

1714042678

[QuintLeo]

You didn't see the existing 1070 ti threads (about 3 of them) yet?

 I have a pair of EVGA SC and one Zotac Mini so far - all 3 run stable at +200 core +700 memory in nvidia-settings under LINUX (or Afterburner on the one card I tried under Windows early on) - 104 watts seems to be THE best efficiency point but anywhere in the 100-110 range is very very good (or 60% in Afterburner, which should get you to 106 watts on all of them so far).

[Jasmint]

Memory DOES make a difference on equihash - though not a big one on the 1080 ti.

 On my Aorus (not the extreme), I found that +100 core +100 memory was the best setting for speed/efficiency that was stable - +150 memory gave a few more sol/s but wasn't long-term stable as those cards seem to push the memory pretty hard already, +200 memory gave me LOWER sol/s probably because it pushed the card up into the next "strap" range with higher latency settings.

 50% TDP (125 watts) gave 505 hash - don't bother this is a waste of the card.
 60% TDP (150 watts) gave 620 hash - this seems to be the best efficiency point that still has decent hashrate.
 70% TDP (175 watts) gave 680 hash - efficiency dropping but not a LOT yet
 80% TDP (200 watts) gave 725 hash - efficiency dropoff getting noticeable
 90% TDP (225 watts) gave 760 hash - ditto
100% TDP (250 watts) gave 780 hash - waste of power unless you have VERY VERY cheap power, and even then it's a bit iffy - though the cards still runs pretty cool at this setting as long as it's NOT in a hot room.


 Other cards may vary, though all of the cards I have to date (1 Gigabyte Windforce, 1 EVGA SC, pair of Aorus) all seem to match nearly identically up to about the 80% point, then the EVGA and Windforce start dropping a bit as temp doesn't let them boost the core as much.

I also found 60-70% TDP is best.

[QuintLeo]

Did some testing last night on my 1070 ti cards - and NO hashrate change under LINUX when I moved them from +700 core to +900, +1100, +1200 (didn't bother testing past that point).

It does appear that LINUX works with "effective" clockrates though not with the base clockrate like Afterburner does, based on some testing I did on 1070 cards (none of which would run stable at +550 or higher under Windows, but were ALL comfortable at +900 under LINUX with noticeable ETH hashrate gains).

[Bimmber]

I'm trying to buy 1080 Ti, but there is a lot of available brands MSI, Gigabyte, Founders Edition,... which one do you recommend?
Thank you!

aorus

My experience with Aorus Xtreme is not that good. My Palit Gamerock cards run way better and clock higher.

[QuintLeo]

Palit is not available in some parts of the world, apparently including ALL of North America (and definitely the US) through normal distribution channels.
Makes it hard to compare 2 cards when I've never seen either (all of my Aorus are the "standard" 1080 ti model, NOT the Extreme).

[VicTwenty]

Memory DOES make a difference on equihash - though not a big one on the 1080 ti.

 On my Aorus (not the extreme), I found that +100 core +100 memory was the best setting for speed/efficiency that was stable - +150 memory gave a few more sol/s but wasn't long-term stable as those cards seem to push the memory pretty hard already, +200 memory gave me LOWER sol/s probably because it pushed the card up into the next "strap" range with higher latency settings.

 50% TDP (125 watts) gave 505 hash - don't bother this is a waste of the card.
 60% TDP (150 watts) gave 620 hash - this seems to be the best efficiency point that still has decent hashrate.
 70% TDP (175 watts) gave 680 hash - efficiency dropping but not a LOT yet
 80% TDP (200 watts) gave 725 hash - efficiency dropoff getting noticeable
 90% TDP (225 watts) gave 760 hash - ditto
100% TDP (250 watts) gave 780 hash - waste of power unless you have VERY VERY cheap power, and even then it's a bit iffy - though the cards still runs pretty cool at this setting as long as it's NOT in a hot room.


 Other cards may vary, though all of the cards I have to date (1 Gigabyte Windforce, 1 EVGA SC, pair of Aorus) all seem to match nearly identically up to about the 80% point, then the EVGA and Windforce start dropping a bit as temp doesn't let them boost the core as much.

Which miner are you using?
I'm getting 735 sol/s @ 90% TDP (225 watts).
Aorus (non-extreme. N108TAORUS) and Zotac (AMP Edition ZT-P10810D-10P)
+100 core, +100 memory.
Ubuntu with dstm miner.

[Bimmber]

I`m getting:

777-790 sol/s @ 86% TDP (215 watts).
Palit GameRock 1080ti
+180 core, +500 memory, 80% fan speed

SMOS with Bminer.

[Wetlint]

I`m getting:

777-790 sol/s @ 86% TDP (215 watts).
Palit GameRock 1080ti
+180 core, +500 memory, 80% fan speed

SMOS with Bminer.

How much faster than EWBF with the same setting?

[Bimmber]

Haven`t tried EWBF, only DSTM and Bminer.

[QuintLeo]

I was using EWBF - have played with DSTM a little but so far it gives the same hashrate (give or take 1%) as EWBF on my cards at the settings I use - NOT worth the 2% dev fee.

This is not to say "don't try DSTM yourself", as YOUR cards at YOUR settings might see more than a 2% improvement in hashrate and there is no way to know for sure 'till you TRY it.

I have not tried BMiner and am unlikely to do so as all of my Nvidia systems at this point are LINUX based and I see ZERO reason to move back to the instability of Windows on them.
I only tolerate Windows on my Polaris and Vega AMD based rigs as the LINUX drivers for the recent generation AMD cards are junk and support is WAY worse than for the fglrx (R9 3xx and older) generation cards, which wasn't SUPER great to start with.

[drawingthemoon]

I will share my 1080ti settings for best results in zcash mining with you guys,

Power Target to 65%
GPU Clock Offset +60
Mem Clock Offset need to be figured out be everyone self.

With these Setting I have less power and a better soll to power ratio with my 1080tis

What it does you just undervolt and overclock your GPUs
In some times you need more mem but you can play around with the setting till you find your sweet spot.

[Bimmber]

I will share my 1080ti settings for best results in zcash mining with you guys,

Power Target to 65%
GPU Clock Offset +60
Mem Clock Offset need to be figured out be everyone self.

With these Setting I have less power and a better soll to power ratio with my 1080tis

What it does you just undervolt and overclock your GPUs
In some times you need more mem but you can play around with the setting till you find your sweet spot.

How many sol/s are you getting on zcash with these settings? And how many sol/w ?

[VicTwenty]

@QuintLeo
Which driver and cuda version are you using?

EWBF on lubuntu 16.04 getting 730-740 sol/s, same as dstm.
Power 225w, +100 core, +100 memory

Maybe I lost the 'silicon lottery' on these cards.

[QuintLeo]

No clue on cuda version.
I THINK they are running the 384.98 drivers, as I also have 1070 ti cards in the machine.

Are you getting the same hashrate on the Aorus as on the Zotac?

Also, try +100 core +200 memory - I figured out LINUX works with "effective" clock rates not with "base" clock rates a couple nights back, and the +100 +100 came from testing on a Windows machine.

[VicTwenty]

Aorus 740 sol/s
Zotac 730 sol/s
Sometimes they are getting the same hashrate, most of the time the aorus is +10 sol/s more than the zotac.

I've tried 384.111, 387.34 and 390.30
390.30 gets the best hashrate so far.

Not sure which cuda version the miner is really using. I tried to remove 7.5 and install 9.1
When I run the following commands I get different versions.

$ cat /usr/local/cuda/version.txt
CUDA Version 9.1.85

$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2015 NVIDIA Corporation
Built on Tue_Aug_11_14:27:32_CDT_2015
Cuda compilation tools, release 7.5, V7.5.17

[QuintLeo]

All of the Pascal cards are listed as CUDA Compute 6.0 or 6.1 capability.
I'm not sure where the numbers you are getting are from or about?


Might just be card-to-card variation. Have you tried +100 core +200 memory yet, since you're running it on Linux?

[VicTwenty]

"384.111, 387.34 and 390.30" are the driver versions.

+100 core +200 memory
average is Aorus 730 sol/s and Zotac 720 sol/s.

+75 core +600 memory gives a similar hashrate.
Tried upping the core, found that +105 is the limit. Anymore then that and the miner will crash after a few hours.

[2stout]

Every card is different, even among the same models; however, 1999 appears to be a sweet spot for core OC for 1080ti.

[QuintLeo]

For reference, just moved all 3 of my Aorus 1080 ti cards to one rig together yesterday (long planned change that I finally got around to driven by my moving my Vega into my gaming machine to replace one of the 1080 ti cards).
There is noticeable variation between the cards depending on position IN the rig (the outside card does noticeable better no matter WHICH actual card is in that slot).

Boost clock reaction to temperature of the GPU (and RAM perhaps?) makes a noticeable difference on hashrate AND efficiency for those 3 cards, even though actual card-t0-card variation is very small between the 3.