Thank you very much for your attention and assistance.
I will try your suggestions.
Hello,
I am trying to push my ZOTAC 1080 Ti mini to its limits. I failed to make it run anything higher than 45.93MH. It is T-Rex miner I use.
I just cannot be sure about @jannoke suggestions. I appreciated it if anybody correct my wrongs below.
What I understand is I need to check memory clock mhz using nvidia-info at command prompt when there is no "mt":"value" option applied to T-Rex miner and when there is no over clock settings in place, no pill etc.
When I do that, I get below output and read 5005mhz as my memory clock.
root@Home_Rig:~# nvidia-info
Sat Apr 10 15:33:26 +03 2021
=== GPU 0, 01:00.0 GeForce GTX 1080 Ti 11178 MB ===
Bios 86.02.40.00.33, PCIE Link Gen 3, PCIE Link Width 16x
S/N -, UUID GPU-1fdd0800-589f-1814-8837-5734f899eb9b
Power Limit: Min 125.0 W, Default 250.0 W, Max 300.0 W, Current 250.0 W
Frequency: GFXCore 1911 MHz, MEMCore 5005 MHz
Memory: Total 11178 MB, Used 4430 MB, Free 6748 MB, Micron GDDR5X
Utilization: GPU 100 %, MEM 76 %
PSTATE P2, PWR 209.0 W, Temp 58 °C, Fan 99 %, RPM 2926
root@Home_Rig:~#
Then, in order to adjust it to ~4900mhz I apply -105 to my memory. Add miner option "mt":"6" (I have no other Nvidia card), set core to 80 for starter, and finally set my power limit to 250 as my current aim is to get most MH out of the card. Doing that and I get around 45.66MH as can be seen below.
At this stage,
nvidia-info reports my memory clock as 4957mhz.
Then I try to increase core by 20 and give it a couple minutes to settle. However, that is not increasing my MH but decreasing it as below
At this stage,
nvidia-info reports my memory clock as 4957mhz
I really could not get a hold of this card OC settings and I’m constantly failing to increase it up to anything like 46MH or more.
I appreciate any help.
Thanks & Regards,
Ertan
Do as suggested at first!!! After that start experimenting. If you are not following the lead then how can you complain that you don’t get same result.
do not too far beyond 200w for starters. it might make the card to overclock too much depending on settings. Also it will make the card unnecessarily hot and it might internally thermal throttle. Especially as you are running card with lower cooling capacity.
All that said - i instructed to hit 4900mhz. You are hitting 4957. Don’t deviate from my suggestions then we can get somewhere. If i set my clock to 4957 I will lose 0.5mhs also. As long as you have not found the memory sweetspot you won’t gain with other things.
Note: every time you change overclocks it will take around 2 minutes for t-rex to recompute it’s algorithm for the timings and it will not immediately yield better results. it will try to get back in sync and in around 2 minutes you can be sure it’s ok (also add hiveos lag into equation, I use miner output directly from console for faster feedback).
it is also said, that not all cards handle mt6. if you have problems start with lower value. dev of trex says to start with 1. I have high end cards (gigabyte aorus extreme watercooled and gainward phoenix) so I jumped to mt6 and have good results. Since you have budget card your mileage may wary. Although - it is said that higher mt should give invalid shares or crash, not lower hasrate.
There is also a subtle hint that your memory is not maxed out - it shows 76% utilized. It should go 100% usually. I do have seen on some 1660super cards with Samsung memory that it will not reach 100%, but all of my other cards (15 total) will hit 100% mem utilization on eth.
1 Like
Hello,
Well, I did not understand what you instructed earlier clearly, and I’m trying to learn. I thought I did explain my situation in my initial post.
This I didn’t know “putting higher power limit might make card overclock too much”. I will set smaller watt limits like 210w
This is the part that I cannot be sure. I already adjusted my memory to be exactly 4900mhz in my first trial. What I understand now is that, I need to observe nvidia-info output after each core increase, and adjust memory clock to be 4900mhz. Do that after each change to core. Is that right?
Using mt6 I do not have invalid shares, I do not have a crash. Actually, I did not have such problems with that card at all. It did happen when I was playing to get more MH. However, once I am done playing with OC settings, it works without any problem for days.
I am not sure what to do about this. Maybe it is not 100% because I still could not do a proper over clock. I will see if it changes once I get this card to run with higher MH.
I am going to sum up all these and give it another try tomorrow afternoon.
Thank you.
Here is my current status.
mt is set to 1
nvidia-info at that settings:
=== GPU 0, 01:00.0 GeForce GTX 1080 Ti 11178 MB ===
Bios 86.02.40.00.33, PCIE Link Gen 3, PCIE Link Width 16x
S/N -, UUID GPU-1fdd0800-589f-1814-8837-5734f899eb9b
Power Limit: Min 125.0 W, Default 250.0 W, Max 300.0 W, Current 200.0 W
Frequency: GFXCore 1873 MHz, MEMCore 4908 MHz
Memory: Total 11178 MB, Used 4430 MB, Free 6748 MB, Micron GDDR5X
Utilization: GPU 100 %, MEM 100 %
PSTATE P2, PWR 201.5 W, Temp 56 °C, Fan 99 %, RPM 2926
root@Home_Rig:~#
I cannot make it 4900 mhz exactly. It either goes 4896 or 4908.
If I let power go up, it goes up to 215w. Increasing PL is not increasing MH though.
BTW, I have absolutely no idea how to find memory sweet spot at all.
1 Like
So for starters now can you keep everything as it is and increase the mt. Does it make any difference?
Also now you have mem utilization @100% which is more like it should.
You won’t get it 4900 indeed. you get it as you said, which is same as my output.
I have two settings for you on gigabyte aorus extreme WB:
This will give me 49mhs @215W, making it 216khs/per watt which is really poor performance (1660 super goes around 410khs/per watt). Notice big OC on memory.
This one is using negative memory speed to achieve almost the same result:
Yelding 47.1x mhs giving me around 250khs/per watt.
This will keep memory cool and for 2mhs I don’t see point to add additional 35W of power.
Also as you can see I have really big negative for memory and small positive for core , because it’s high end card with factory OC. That is why it’s not very informative if someone says “minus this or that”. The actual base clocks depend of the model of the card.
For you , I would try to get core as high as possible (around 2000mhz, it will crash on me on 2200) and then start playing with mt values.
Also : I use nbminer, because it displays clocks and other data in a nice table so I can check the clocks from miner. Maybe give it a try and it works better 
I use these settings on 2 different 1080ti’s and they work similarily.
in conclusion : use nbminer, get core around 2000Mhz, leave memory 4908Mhz and if getting decent results then try increasing mt 1 => 3 => 5 => 6 (I don’t think there is point on checking every one mt. check every other, it’s faster).
2 Likes
If you are interested in performance only, take a look at these settings. I have two Zotac 1080ti mini so it cannot get any closer
T-Rex is the miner, with following extra config arguments:
“kernel”:“5”
“mt”:“4”
Now, I have not really looked into mt settings difference, it kinda looks all the same to me - but I only tried 3 and 4. If you see the difference please share the findings.
DEVICE #0:
GPU CLOCK: 1974 MHz
MEM CLOCK: 5386 MHz
DEVICE #1:
GPU CLOCK: 1987 MHz
MEM CLOCK: 5386 MHz
Also hashrate will vary depending on the temperature. My rig is in a VERY cold room under 10C so it can achieve 61C without much work from the fans. If you hit 70C expect hasrate to fall to 45mhs. In my opinion 60-62C is a sweetspot for these cards (being minis) but again if you notice something different please share.
I have tried also the settings that janoke suggested (with slight deviation) cc+170 mc-105 pl190 and I got 45+ on both cards. I saved that settings for the summer to look into
I cannot understand what figure is what without a caption. Is it possible you share overclock details with caption, please?
Thanks
I still cannot understand what is 75 number in your initial post.
Fan speed was set to 75%. I changed it to 0 (Auto) after I’ve set up AutoFan settings.
I cannot seem to get constant MH rate with my card. I am not changing any over clock settings, but I see card can get 46.66MH. After sometime (an hour, 10 minutes, several hours) it can go down to 43 MH. Again, after sometime it can increase to 45.3MH etc. Card runs around 59-61c
My watt also changes in parallel with MH changing, if MH goes down, watt decrease, if MH goes up watt increase.
I do not have such MH fluctuation for my other card. It is only with that ZOTAC 1080 Ti mini
This card is from year 2017. It is not attended like thermal pad renewal. Can this be a reason?
Anyone has any idea?
wow this was great info, thank you. It kinda sux tweaking remotely not knowing if the rig is crashing right away!! but i just managed to squeeze 22 more mh out of my 5 cards
You’re welcome.
When tweaking, bare in mind that 1080ti differ one model to another. In my experience Zotac minis cannot go beyond 2000Mhz for core clock for longer periods then maybe several seconds and they can crash because of that. I know that AMP version can go past 2100Mhz in most cases which means slightly better hashrate.
If you have minis and want to be on the safe side lower clocks a bit until you feel confident in them. For example cc +150, mc +700. You’ll loose a little hasrate (around 1mhs) for sure but they should run fine with that - hitting core clock just below 2000Mhz at any point is the key.
So my question now is, where does that sweet spot number come from, how do I tune my 3080 the same way?
Check out YT videos (all that you can find) and see what the people are using and use that as your starting point. Most recent ones are usually more relevant in my experience.
I had to go the hard way through trial and error since my cards are specific and none of the settings that I could find at the time gave me results that I wanted and knew that could achieve.
And I used Windows and Afterburner for that since it was easier to tinker that way even with differences between Windows and Linux. Also, it helps if you have IGP for video output so the card that you are testing will not cause whole system to freeze up or reboot (it may happen anyway but rarely).
And factor in the silicon lottery - if you get up to 95% of the set goal and the card refuses to cooperate further you may need to consider that as a limit.
But with a 3080 I would not complain with any result that I get
1 Like
I finally understood all advises. It is only after @sarmagedon own OC sharing.
Now, I can run my card stable at 47.48MH
It is not stable if I increase core up to 190. So, This is the best I can approach to 2000mhz core speed.
I am still on T-Rex miner as I checked my frequencies in console. I am just not sure if that “kernel” argument helps. I left it there for the moment.
Now, I am going to test different mt values, but I suspect there will be much difference.
I understand that my card cannot do more than this probably because difficulty of ETH mining and other factors that I do not know.
One thing that I cannot still understand is, All 1080 Ti chips are produced by Nvidia almost same. So, what makes a card to be “high end” or “low end” as long as chip is the same?
As far as I see, “high end” cards can still do 49MH if power usage is omitted. Not my card though.
Thanks.
1 Like
All those cards are working within NVidia specification (think Founders Edition cards). Since some chips contain less impurities, they respond better to OC and manufacturers (3rd party) usually mark those models as OC or Gaming. They all sport higher then reference clock and memory speeds. Zotac minis most likely are better quality chips with tiny cooling solution since they are similar to Founders Edition in performance but enduring higher temp and slight OC. What we do here is abuse them to the limit outside their specification. If you ran them on Nvidia stock speed they’ll all perform within margin of error from each other.
There is a review on rx5700 on pcgamer that has that explained quite nicely I think:
Modern microchips are made using large (300mm) silicon wafers, cut from a large silicon crystal cylinder. There are impurities in the wafers used for making CPUs and GPUs, and these can lead to errors or differences in performance characteristics. My understanding is that chips from the center of the wafer tend to be ‘better’ than those near the edge, so they may require less voltage to hit the same clockspeed, or they may run at higher clockspeeds at the same voltage. Chips with impurities or errors don’t fully work, so modern processors build in redundancies and some portions of the chip can be disabled to produce a working chip—but one that’s not as fast as a fully enabled chip.
After each wafer gets cut into individual chips, each is tested to determine how good it is in a process called binning. Thus, the best chips usually get sold as the fastest, most expensive parts—the Core i9-9900K, Ryzen 7 3800X, or RX 5700 XT. Meanwhile, the functional but perhaps not quite as good chips are harvested and sold as lower tier parts. Depending on how intensive and accurate the binning process is—and how mature the manufacturing process is—the gap between the ‘best’ and ‘worst’ functional chips from a wafer may be relatively small.
1 Like
Доброе! Подскажи пожалуйста опыт настройки gtx1080ti.Ошибка разгона в последнее время происходит,случилось все после технической работы в HiveOs,до этого все работало на 150,1400,250ват.Риг больше месяца у меня в работе